Transcript PowerPoint

Data Analysis Outline
• What is an experiment?
• What are independent vs. dependent
variables in an experimental study?
• What are our dependent measures/variables in
this study?
Goals for Today
• Learn about the basics of an experiment
• Learn how to create a unit-weighted composite
variable and how/why it is used in psychology.
• Learn how to create composite variables in
SPSS.
• Learn how to compare the difference between
two groups using Cohen’s d.
Composite Scores
• When we have used multiple ways of assessing
a construct (e.g., self-esteem), we often create a
composite that captures the these scores.
• It is assumed that there is variation across
people with respect to the latent variable (i.e.,
self-esteem).
• A “latent variable” is one that we assume exists,
but that we cannot observe directly.
• Most constructs in psychology are latent
variables: memory, extraversion, self-esteem,
intelligence.
•It is also assumed that variation in this latent variable causes variation in
the observed responses (i.e., the ratings of each item).
Item 1
Item 2
Item 3
Item 4
Self-esteem
Self-esteem
Self-esteem
Self-esteem
+
+
+
Latent
Self-esteem
-
Reverse Scored Items
• Some items are negatively related to the
construct of interest.
– Ex: “I feel I do not have much to be proud of. ”
• These items cannot be weighted in the same
fashion as the others when creating a composite
variable.
Unit-weighted composite
• To create a “unit-weighted composite”—the most
commonly used composite in personality
psychology, do the following:
– 1. Reverse-key responses to items that are in
the opposite direction of the construct.
• One way to do this is to use the following
formula:
• Max - X + Min
• Thus, on a 1 (Min) to 5 (Max) scale, like the one
we used, we would use the following equation to
reverse key the responses:
• Rev key response = 5 – X + 1
• 2. Once the appropriate responses have been
reverse keyed, simply average the responses for
each person.
Item
Person 1
Person 2
Person 3
I feel that I'm a person of worth, at least on an
equal plane with others
5
5
2
I feel that I have a number of good qualities.
5
4
3
All in all, I am inclined to feel that I am a failure.
(Reverse)
1 (5)
2 (4)
3 (3)
I am able to do things as well as most other
people.
5
5
2
I feel I do not have much to be proud of.
(Reverse)
1 (5)
1 (5)
4 (2)
Sum
25
23
12
Average
5
4.6
2.4
Qualifications
• This method is the simplest, but there are more complex
ways of creating composites.
– For example, sometimes responses to each variable
are standardized before the averaging takes place.
– In some work, the different variables are weighted
differently. That is, some variables count more than
others.
– In other work, non-linear relationships might be
assumed between the latent variable and an item
response (e.g., Item Response Theory models).
Mean Differences
• The big question in our experiment is whether
people’s self-esteem improves after listening to
a subliminal recording containing subliminal
messages designed to improve self-esteem.
Our Experiment
• Two conditions:
– A. People in the “good” condition were
presented with self-affirming subliminal
messages, such as “You are a good person.”
– B. People in the “bad” condition were
presented with self-defacing subliminal
messages, such as “No one likes you.”
Answering the Question
• One way of addressing the question is whether
the self-esteem of people in the Condition A is
higher than that of people in Condition B. (As
measured after hearing the recording.)
• Everyone has a unique self-esteem score, so we
average the scores (i.e., the composite scores) for
people in Condition A and separately average the scores
for people in Condition B.
• We want two statistics: (a) the mean, which tells us the
average self-esteem value for a person in Condition X,
and (b) the standard deviation (SD), which tells us the
amount of variability there is around the mean in that
condition.
• Mean Difference between conditions:
– (Mean of Group A – Mean of Group B)
– If positive, then Group A > Group B
– If negative, then Group A < Group B
– If zero, then no difference between conditions.
Cohen’s d
• If we divide the mean difference by the average
SD of the two groups, we obtain a standardized
mean difference or Cohen’s d.
d
MA  MB
SD
2
A

 SD / 2
2
B
Pooled standard
deviation
• Cohen’s d expresses the difference between
groups relative to the average standard
deviation of the scores.
Another Way – For Wed.
• We could also ask about the amount of change
that takes place in self-esteem scores from Time
1 (before the recording) to Time 2 (after the
recording).
• Create a composite for the Time 1 scores.
• Create a new variable in SPSS that represents
the Time 2 composite – Time 2 composite
scores.