CHAPTER 8: III.

Physical
I. Internal Validity in Variables
Experiments
Definition
In experimental research, the goal is to
establish internal validity — the confidence Physical variables are environmental
that changes in the dependent variable are factors that may influence participants’
caused only by the independent variable. performance.

If another factor changes along with the Examples:

independent variable, a confounding
variable occurs. When this happens, • Time of day
researchers cannot confidently determine the • Day of the week
true cause of the results. • Testing room
• Lighting
Rule: Internal validity must be • Noise level
established before results can be • Temperature
generalized. • Distractions

Example of a Confound:

If Group A is tested in a quiet room and

II. Extraneous Group B is tested in a noisy hallway, any
difference in performance may be due to
Variables noise, not the independent variable.

Extraneous variables are any variables

other than the independent variable that may
influence the dependent variable.
IV. Methods for
Major types include:
Controlling Physical
1.
2.
Physical Variables
Social Variables
Variables
3. Personality Variables
4. Context Variables 1. Elimination
This study guide focuses on physical and Remove the extraneous variable completely.
social variables.
Example:

• Testing in a quiet room to eliminate

noise.
 • Posting a “Do Not Disturb” sign to This prevents the testing room from
prevent interruptions. becoming a confound.

Limitation:

Some variables cannot be eliminated (e.g.,

wall color, weather conditions). V. Internal Validity
vs. External Validity
2. Constancy of Conditions Internal Validity

Confidence that the independent variable

Keep conditions identical for all treatment
caused the observed effect.
groups.
External Validity
Example:
Ability to generalize findings to real-world
• Use the same room for all
settings.
participants.
• Give standardized written or
recorded instructions. Trade-Off:
• Provide the same amount of time for
completing tasks. • High control increases internal
validity.
This ensures environmental factors affect all • Too much control may reduce
groups equally. external validity.

However:

Without internal validity, external

3. Balancing validity has no value.
If elimination or constancy is impossible,
distribute the extraneous variable equally
across conditions.
VI. Social Variables
Example:
Definition
Two testing rooms must be used.
Social variables involve interactions
• Half of Group A tested in Room 1 between participants and researchers that
• Half of Group B tested in Room 1 may influence behavior.
• Remaining halves tested in Room 2
• Participants randomly assigned
1. Demand Characteristics • Participants do not know which one
they received.
Definition

Cues in an experiment that lead participants

to guess the purpose of the study and change 2. Placebo Effect
their behavior accordingly.
Definition
Participants may:
Improvement that occurs because
• Try to act like a “good subject” participants believe they are receiving
• Attempt to confirm the researcher’s treatment, not because of the treatment
hypothesis itself.
• Alter responses intentionally or
unintentionally Example:

Example: In depression treatment studies, many

participants improve even when given a
In early sensory-deprivation studies, placebo, demonstrating the power of
participants showed psychological expectation.
symptoms not because of deprivation itself,
but because the experiment was presented as Using a placebo group helps researchers
serious and dangerous. Their expectations separate:
influenced behavior.
• True treatment effects
from
• Expectation-based improvements

VII. Controlling
Social Variables 2. Using a Cover Story
1. Single-Blind Procedure Definition
In a single-blind experiment, participants A cover story is a plausible but false
do not know which treatment condition they explanation of the study’s purpose, used to
are in. disguise the real hypothesis.

Example: It prevents participants from guessing the

experimental goal.
In drug research:

• One group receives the real drug.

• Another group receives a placebo. Example: Hannover & Kuhnan
(2002)
Real Hypothesis: • Violates full informed consent
• Requires strong justification
Clothing style influences how people • Requires debriefing
describe themselves. • Must be approved by an IRB
(Institutional Review Board)
Concern:
Use cover stories only when necessary
Participants might guess: for internal validity.
“You want me to rate myself according to
how I’m dressed.”

Cover Story Used:

VIII. Experimenter
Participants were told:
Bias
• The study was about eyewitness
identification A. Definition
• They were part of a “police lineup”
• They must dress casually or formally Experimenter bias occurs when the
to match a suspect description researcher unintentionally influences
participants’ behavior or data recording.
Afterward, they completed a supposedly
unrelated personality task. This creates confounding.

Results:

• Casually dressed participants rated

themselves as having more “casual”
B. Types of Experimenter
traits. Bias
• Formally dressed participants rated
themselves as more “cultivated” and 1. Subtle Behavioral Cues
“restrained.”
Example:
Why the Cover Story Was An experimenter asks:
Necessary: “Did the circular light move faster?”

Without it, participants might intentionally While saying “faster,” she:

rate themselves in a way that fits their
clothing. • Leans forward
• Raises eyebrows
• Speaks louder

Ethical Consideration Participants are more likely to respond

“faster.”
Cover stories involve deception, which:
 Classroom Study (Rosenthal &
Jacobson, 1966)
2. Emotional Transmission
Teachers were told certain students were
Example: “intellectual bloomers.”
If a treatment is unpleasant, the researcher
may appear anxious. Those students later showed greater IQ
Participants detect this anxiety and feel gains.
anxious themselves.
Why?
If anxiety is the dependent variable → the
study is confounded. Teachers:

• Created warmer climates

• Gave more feedback
3. The Rosenthal Effect (Pygmalion • Provided more input
Effect) • Allowed more output

Definition: Expectations shaped outcomes.

Researchers treat subjects differently based

on expectations.
C. Recording Errors
Rosenthal (1978) found:
Classic Rat Study (Rosenthal &
Fode, 1963) • About 1% of observations were
wrong.
Students were told: • 2/3 of errors favored the hypothesis.

• Some rats were “maze-bright” This shows how bias can influence scoring.
• Others were “maze-dull”

In reality, rats were identical.

Results: IX. Controlling

• “Bright” rats performed better. Experimenter Bias
• Students handled them more gently
and enthusiastically. A. Standardization
Expectations changed behavior.
• Use written instructions
• Use audio/video recordings
• Time procedures carefully
• Follow strict protocols
 X. Personality
B. Double-Blind Variables
Experiments
A. Experimenter
Definition:
Personality
Neither the participant nor the experimenter
knows the treatment condition. Experimenters who are:

This controls: • Warm and friendly → produce better

performance
• Demand characteristics • Hostile or authoritarian → reduce
• Experimenter bias performance

Problem:
If friendliness varies by condition →
Example: Triazolam Study (Berman confounding occurs.
& Taylor, 1995)
Control:
Men received:
• Be consistent across conditions
• Triazolam (sleep medication) • Minimize face-to-face contact
• Placebo • Use multiple experimenters if
possible
Neither subjects nor experimenters knew • Balance experimenters across
who received which. conditions

Dependent Variable:
Shock intensity delivered to a fake
opponent. B. Volunteer Subjects
Results: Volunteers tend to:

• Drug group gave higher shock levels • Be more sociable

(26% highest level) • Score higher in social desirability
• Placebo group gave highest shock • Have more liberal attitudes
only 4% of the time • Score higher on intelligence

Because it was double-blind: This limits external validity.

Results were not biased by expectations.
McNemar (1946):
“Psychology is largely the science of
sophomores.”
 C. Practice and Fatigue
Effects
XI. Context
Over time, experimenters:
Variables
• Become more practiced
A. When Subjects Select • Become fatigued

Experiments This can create instrumentation threats.

Experiment titles can bias samples. Control by:

Example: • Practicing beforehand

• Balancing conditions across time
• “Memory Test Experiment” • Minimizing personal interaction
• “Heavy Metal Music Experiment”

Different types of students will volunteer.

Solution:
Use neutral titles like:

• “Impression Formation Study”

B. When Experimenters
Select Subjects
Avoid:

• Choosing only friendly participants

• Assigning difficult participants to
easier conditions

Must:

• Use random selection

• Use random assignment
• Stick to procedure
 1. Between-Subjects Design
CHAPTER 9:
• Different subjects in each condition
I. What Is an • Compare behavior between groups

Experimental 2. Within-Subjects Design

Design? •
•
Same subjects in multiple conditions
Compare behavior within the same
group
An experimental design is the general
structure or blueprint of an experiment. (This chapter focuses on between-subjects
designs.)
Think of it like a floor plan of a building:

• It shows how many rooms there are.

• It shows how rooms connect.
• It does NOT describe what happens II. Selecting and
inside the rooms.
Recruiting Subjects
Similarly:

• The design shows the number of A. Sampling and External

treatment conditions. Validity
• It shows how subjects are assigned.
• It does NOT describe the specific We rely on samples to test hypotheses.
content of the study.
The more representative the sample → the
better the external validity.

What Determines the Ideal:

Design? • Random sampling (everyone has
equal chance of selection)
Three main factors determine experimental
design:
Reality:
1. Number of independent variables
• Often impossible.
(IVs)
2. Number of treatment conditions
3. Whether the same or different
subjects are used
B. Practical Limits
Researchers often use:
Types of Designs
 • Convenience samples A. Statistical Power
• University subject pools
• Volunteers from campus
Smaller samples = less power to detect
• Friends and family (not
effects.
recommended)
Large samples = better chance of detecting
real differences.
Why NOT Use Friends?

• They may know the hypothesis.

• They may respond to subtle cues. B. Effect Size
• Ethical issue: They may feel
obligated. Effect size = magnitude of the treatment
effect.

• Large effect → 10–20 per group may

be enough
C. Volunteer Bias • Moderate effect → 20–30 per group
• Small effect → need larger samples
Volunteers may differ from non-volunteers.
Most researchers prefer at least 20–30
This limits generalizability. subjects per group.
Without coercion, we can never guarantee a
perfectly random sample.

IV. One Independent

III. How Many Variable: Two-
Subjects Are Group Designs
Enough? The simplest experiment:

• One IV
There is no fixed answer. • Two treatment conditions
The required sample size depends on: This is called a two-group design.
1. Variability in responses
2. Statistical power
3. Effect size
Types of Two-Group
Designs
 1. Two-Independent-Groups Design Without it:
2. Two-Matched-Groups Design
• Groups may differ before the
(We focus here on independent groups.) experiment begins.
• A selection threat may occur.
• Confounding becomes possible.

V. Two-Independent-
Groups Design Example: Kelley (1950)
Students received descriptions of a lecturer:
In a two-independent-groups design:
• Group 1: Told lecturer was “warm”
• Subjects are randomly assigned.
• Group 2: Told lecturer was “cold”
• Each subject participates in only
ONE condition.
After discussion, students rated him.
• Groups are independent of each
other.
Results:

• “Warm” group rated him more

positively.
A. Random Assignment
Why random distribution mattered:
Random assignment = every subject has If all “warm” descriptions were given to
equal chance of being in any condition. students sitting in front (who may already be
more engaged), results would be
Important: confounded.

• Random selection ≠ Random

assignment

Random Selection: VI. Experimental

Improves external validity. Group vs Control
Random Assignment: Group
Protects internal validity. To test an IV, we need at least two
conditions.

Why Random Assignment

A. Experimental Group
Is Important
Receives the independent variable Read passage arguing free will is an illusion.
manipulation.
Control Group:

Read neutral passage.

B. Control Group
Then:
Does NOT receive the manipulation. Participants completed math problems
where cheating was possible.
Everything else must remain identical.
Results:
Experimental group cheated more.

Without a control group:

No-Treatment Control We couldn’t rule out alternative
Group explanations.

Example:

• Exercise group: 15 minutes aerobic

activity
VII. Placebo Control
• Control group: No exercise
In drug studies:
Problem:
What if control group studies during that • Experimental group receives drug.
time? • Placebo group receives inert
substance.
That creates a maturation threat. • Procedures are identical.

Solution: This ensures:

Control group should engage in a neutral The only difference is the presence of the
activity. IV.

Placebo groups are excellent control groups

but may raise ethical issues.

Example: Vohs & Schooler

(2008)
Hypothesis:
VIII. Two-
Believing behavior is predetermined
Experimental-
increases cheating. Groups Design
Experimental Group: Sometimes we compare:
Two different levels of the IV. Problem:
No random assignment.
Examples:
Executives were chosen based on ability to
• 15 min exercise vs 10 min exercise learn task faster.
• Violent game vs nonviolent game
Confound:
Pre-existing differences may explain ulcers.

Example: Anderson & Dill Thus:

Study lacked internal validity.
(2000)
Participants played:

• Violent game (Wolfenstein 3D) X. Example of

Nonviolent game (Myst)
•
Proper Random
Then:
Reaction-time task where they could Assignment: Zajonc
administer noise blasts.
(1969)
Results:
Violent game players gave longer noise Hypothesis:
blasts. Cockroaches run faster with an audience.

This shows difference between two IV Two conditions:

levels.
• Audience present
• Alone

Roaches randomly assigned.

IX. Classic
Even though individual weights differed,
Confound Example: Random assignment produced similar
average weights across groups.
Brady (1958)
This prevented weight from becoming a
Study: confounding variable.
Monkeys either:

• Had control over electric shock

(executives)
• Had no control (nonexecutives)

Executives developed more ulcers.

 Group 1 → Audience present
XI. Why Larger •
• Group 2 → No audience

Samples Help Subjects are assigned using random

assignment.
Randomization
With:

• 2 subjects → impossible to balance

Important Assumption
differences
This design assumes that randomization is
• 10 subjects → better balance
successful, meaning:
• 30+ subjects → much better
equivalence
Treatment groups are roughly equal on all
important extraneous variables before the
Larger samples increase chance that groups
experiment begins.
are equivalent on:
However, random assignment does not
• Known variables
guarantee perfectly equal groups —
• Unknown variables
especially when sample sizes are small.

XII. When to Use a

Two-Independent- When Randomization Fails
Groups Design Even with random assignment:

A two-independent-groups design is • One group may accidentally be

appropriate when: heavier, smarter, older, etc.
• That variable may influence the
• There is one independent variable dependent variable (DV).
(IV). • The extraneous variable becomes a
• The hypothesis can be tested using confounding variable.
two treatment conditions.
• Subjects can be randomly assigned Cockroach Example Problem:
to groups.
• Control group = heavier cockroaches
• Experimental group = lighter
cockroaches
Example: Cockroach • DV = running speed

Experiment If heavier cockroaches run slower, we

cannot determine whether:
Hypothesis: Cockroaches run faster with an
audience than without an audience.
 • The audience caused the difference Purpose of Matching
OR
• Weight caused the difference
Matching ensures:
This is confounding.
• Groups are equivalent on an
important extraneous variable.
• That variable does not become a
confound.
Bernoulli’s Law of Large
Numbers
The larger the sample size: Example: Cockroaches
• The more likely random assignment
Matched on Weight
produces equivalent groups.
Step 1: Measure weight.
Step 2: Form pairs with similar weights.
Small samples → higher risk of unequal
Step 3: Randomly assign one from each pair
groups.
to each condition.

Result:

XIII. The Two- • Experimental group average weight

≈ Control group average weight.
Matched-Groups Confounding from weight is eliminated.
Design
Because randomization can fail (especially
with small samples), researchers may use a: XIV. Matching
Two-Matched-Groups Procedures
Design There are three types of matching:
This design:

• Uses two treatment conditions.

• Matches subjects on a variable 1. Precision Matching
strongly related to the DV.
• Then randomly assigns one member Subjects must have identical scores on the
of each pair to each condition. matching variable.

✔ Most strict
✖ Hard to achieve
 1. Run experiment with random
assignment.
2. Range Matching 2. Measure matching variable.
3. Form pairs afterward.
4. Discard unmatched subjects.
Subjects must fall within a specified range
(e.g., within 3 IQ points).
⚠ Problems:
✔ More practical
• Some subjects may be discarded.
✔ Most common method • Reduces sample size.
• Reduces statistical power.
• May affect representativeness.

3. Rank-Ordered Matching
• Rank subjects from lowest to
highest. XVI. When Should
• Pair adjacent scores.
You Use Matching?
✔ Uses most subjects
Use matching when:
✖ Pairs may differ too much
• Sample size is small.
• A subject variable is strongly
related to the DV.
XV. Matching Before • Literature suggests a key extraneous
variable must be controlled.
vs After the
Experiment
Example: Devlin (2005)
A. Matching Before Study: Training program for teachers
working with severely disabled students.
✔ Preferred method
✔ Groups start equal Matched on:
✔ No need to discard data later
• Intellectual level
• Behavioral characteristics

B. Matching After Result:

Students in trained-teacher group showed:
Steps:
• Greater engagement
• More interaction
Matching allowed researchers to detect the • Match on weight in weight-loss
effect with a smaller sample. study ✔
• Match on weight in spelling study ✖

XVII. Advantages of
Matching XIX. Multiple-
Groups Designs
✔ Controls important extraneous variables
✔ Reduces confounding Sometimes two groups are not enough.
✔ Increases statistical power
✔ Makes treatment effects easier to detect If the hypothesis requires testing multiple
levels of the IV, use a:

Multiple-Groups Design
XVIII. More than two treatment conditions.
Disadvantages of
Matching Example: Exercise & Exam
✖ Difficult to choose correct matching Performance
variable
✖ May accidentally increase differences in Independent Variable: Minutes of exercise
other variables Dependent Variable: Exam performance
✖ Requires additional measurement
✖ Reduces sample size if subjects are Three-Group Design
discarded
✖ If matching variable is weakly related to • Control (0 min)
DV, it can reduce power • 15 minutes
• 30 minutes

Result: Performance increases as exercise

increases.
Important Rule
Only match on a variable that is:
Four-Group Design
Strongly correlated with the dependent
variable. • 0 min
• 15 min
Example: • 30 min
 • 45 min Solution: Block
Result: Randomization
• Moderate exercise improves Steps:
performance.
• Too much exercise decreases 1. Create blocks containing all
performance. treatment conditions in random
order.
This forms an: 2. Assign subjects sequentially within
each block.
3. Ensures equal numbers per
Inverted-U Relationship condition.
Performance increases to a point, then
declines.

XXII. Choosing
XX. Why Use Number of
Multiple Groups? Treatment
Conditions
✔ Allows testing of different IV levels
✔ Detects patterns (linear, inverted-U, etc.) Ask:
✔ Provides fuller understanding
What will I gain by adding extra conditions?

If two groups test the hypothesis sufficiently

→ use two.
XXI. Assigning
More groups:
Subjects in Multiple ✔ More information
Groups ✖ More subjects needed
✖ More time
✖ More complex statistics
Problem:
Random assignment may create unequal
group sizes.
XXIII. Pilot Study
A pilot study is:

A small-scale mini experiment to:

 • Test treatment levels
• Detect procedural problems
• Refine operational definitions

Conducted before the main experiment.