Single-Factor Experimental Designs Passer Chapter 8

Download Report

Transcript Single-Factor Experimental Designs Passer Chapter 8

Single-Factor
Experimental
Designs
Passer Chapter 8
Slides Prepared by Alison L. O’Malley
What does it mean to have experimental
control?
Here, the focus in on experiments with
one independent variable (i.e., singlefactor designs).
Experimental Control
•Manipulation of one or more IVs
•Measured DV(s)
•Everything else held constant
Need to rule out all possible influences on
the DV other than the IV!
Experimentation: Some Review
•What are the three criteria that must be
met in order to make a causal inference?
•What is a confounding variable?
Causality and Confounds
• What are the three criteria that must be met
in order to make a causal inference?
• Covariation of X and Y
• Temporal order
• Absence of plausible alternative explanations
• What is a confounding variable?
• A factor that covaries with the IV
• Cannot tell whether the IV or the confound
affects the DV
How can confounding factors be reduced?
Confounding Variables
•Many environmental factors can be held
constant
• This is why laboratory settings are so
attractive!
•Those environmental factors that cannot
be held constant (e.g., time of day) can be
balanced across different experimental
conditions
Confounding Variables
•Confounds dealing with participant
characteristics (e.g., personality, age) are
further addressed through experimental
design
•One major distinction to attend to is
whether psychological scientists employ a
between-subjects design or a withinsubjects design
Confounds
Research
Description minimized
Design
through…
Between- Different
Random
subjects participants assignment
in each
condition
WithinParticipants Counterbalancing
subjects encounter
all levels of
experiment
Between-Subjects Designs
Age, cognitive ability, personality, mood, ethnicity, eating habits…and
so on
Level 1
Participant ID:
1
2
3
4
5
Level 2
Participant ID:
6
7
8
9
10
Level 3
Participant ID:
11
12
13
14
15
In theory, randomly assigning participants to levels of the IV (1, 2, or
3) will distribute individual differences throughout the conditions so
experimental groups are equivalent in every way except the IV.
Within- Subjects Designs
Participants serve as their own controls
Level 1
Participant ID:
1
2
3
4
5
Level 2
Participant ID:
1
2
3
4
5
Level 3
Participant ID:
1
2
3
4
5
Counterbalancing ensures that participants experience the
IV levels in different orders.
Manipulating an IV
The Case for Multilevel Designs
A two-group design would miss this nonlinear effect
Experimental and Control Conditions
More experimental terminology
Participants in an experimental condition are
exposed to a “treatment,” whereas participants
in a control condition do not receive the
treatment.
E.g., In goal setting theory research,
participants in the experimental condition are
given a difficult, specific goal to achieve,
whereas participants in the control condition
are simply told to do their best.
Between-Subjects Designs:
Advantages
• No carryover effects
• Less likely that participants will catch
on to the hypothesis
• Exposure to multiple levels of the IV
may be impossible or ethically and
practically difficult
Between-Subjects Designs:
Disadvantages
• Different people in each condition
generate more “noise” (i.e.,
variability), making it more difficult to
establish an effect of the IV on the DV
• More participants required
Between-Subjects Designs:
Independent-Groups
• Participants randomly assigned to
conditions
• What are potential issues with random
assignment?
• Block randomization helps overcome these
issues
• Run through random order of blocks
(rounds of conditions) until desired
sample size reached
Between-Subjects Designs:
Matched-Groups
• Identify a relevant characteristic (a
matching variable) and randomly
assign participants to conditions based
on their standing (e.g., high, average,
low) on this characteristic
• Wise to use possible confounds as
matching variables
Between-Subjects Designs:
Natural-Groups
• Rather than manipulate IV, create different
groups of participants based on naturally
occurring attributes called subject
variables (e.g., age, gender, personality)
• Subject variables often referred to as
quasi-IVs
• But is this truly an experimental design?
Concept Clarification
What is the difference between random
sampling and random assignment?
Random Sampling vs.
Random Assignment
Within-Subjects Designs:
Advantages
• Also called repeated-measures designs
• Need fewer participants
• Can collect more data per condition
• Required to investigate certain
research questions
• Don’t have to worry about nonequivalent groups
Within-Subjects Designs:
Disadvantages
• Greater likelihood that participants will
catch on to experiment’s purpose
• Cannot accommodate all research
questions
• Order/sequence effects occur when
participants are affected by the order
in which they encounter conditions
Within-Subjects Designs:
More on order effects
• Progressive effects occur when
participants are altered by the sequence of
conditions they encounter (e.g., acquiring
experience with the task and thus
improving performance)
• Carryover effects occur when participants’
responses in one condition are affected by
the prior condition (e.g., taste of stimulus
from Trial 1 lingers during Trial 2)
• The cure? COUNTERBALANCING
Within-Subjects Designs:
Counterbalancing Goals
1. Every condition of the IV appears
equally often in each position
2. Every condition appears equally often
before and after every other condition
3. Every condition appears with equal
frequency before and after every other
condition within each pair of positions
in the overall sequence
Within-Subjects Designs:
All Possible Orders
• Also known as complete counterbalancing
• Establish all possible sequences (n!) of IV
conditions, and assign equal number of
participants to each sequence
• Every possible confounding effect is
counterbalanced (hence the name!)
• Requires a large number of participants to
satisfy all counterbalancing goals
Within-Subjects Designs:
Latin Square
• Matrix design wherein matrix
structured so that each condition
appears only once in each column and
each row
Within-Subjects Designs:
Latin Square
• Accomplishes goals of all-possibleorders design except Goal 3
• When IV has odd number of
conditions, cannot construct a single
matrix that accomplishes Goal 2 if
using a Williams Latin Square
• Alternative method is random starting
order with rotation
Within-Subjects Designs:
Random-Selected-Orders
• Subset of orders is randomly selected
from the set of all possible orders
• Each order administered to one
participant
• Refrain from using with small number
of participants
Within-Subjects Designs:
Other Options
• Two designs wherein participants
encounter each condition more than once
• May be more practical than other
counterbalancing designs, lets you
examine reliability of participants’
responses, and extends generalizability of
results
Within-Subjects Designs:
Block-Randomization-Design
• Participants encounter all conditions
within a block and are exposed to multiple
blocks
• Each block contains a newly randomized
order of conditions
How does this block randomization design
differ from that applied in betweensubjects designs?
Within-Subjects Designs:
Reverse-Counterbalancing
• Also called an ABBA-counterbalancing design
• Participants first receive random order of all
conditions
• Participants then receive conditions once
more in reverse order
• Watch out for nonlinear order effects!
Making Sense of Experimental Data
• First compute descriptive statistics (e.g.,
mean for each condition)
• Inferential statistics then used to establish
whether findings are statistically
significant
Are differences between conditions due to chance?
Remember that p < .05 criterion?
Making Sense of Experimental Data
• t test examines mean difference between
two conditions
• ANOVA (analysis of variance) used with
three or more conditions
Would researchers
use a t test or ANOVA
here?
Making Sense of Experimental Data
• An ANOVA would be preferable given that
there are 3 conditions
• If ANOVA reveals a statistically significant
overall pattern of mean differences,
post-hoc tests can
reveal which
which
means differ