PowerPoint - Your Personality

Download Report

Transcript PowerPoint - Your Personality

Goals for Today
• Review 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 mean difference
between two groups using Cohen’s d.
Review
• What is an experiment? What is random
assignment to conditions and why does it
matter?
• What are independent vs. dependent
variables in an experimental study?
• What are our dependent measures/variables in
our subliminal study?
Composite Scores
• When we have multiple ways of assessing a
construct (e.g., self-esteem), we often create a
composite variable that captures the these
scores.
Composite Scores
• Why do we average scores together to create a
composite?
• We assume that a “latent” variable or
“construct”, such as self-esteem, manifests itself
in various ways.
Composite Scores
• Each of those manifestations, however, is an
imperfect reflection of a person’s self-esteem.
• Example: A person may indicate that they feel
good about themselves not because they feel
especially good about themselves per se, but
because they hold others in such low regard.
Composite Scores
• O=T+E
• We assume that our measurement or
observation, O, is a function of at least two
factors: A true score (T: the value that we expect
to observe) and measurement error (E).
• If the measurement errors are random, then
averaging several O’s together should give us a
better approximation of T.
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:
• 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 composite variables.
– For example, sometimes responses to each variable
are standardized (transformed to z-scores) before
the averaging takes place.
– In some work, the measurements might be weighted
differently. That is, some variables might 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.
• [open SPSS]
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
• Did our manipulation have an impact on peoples’
self-esteem?
• One way of addressing the question is by
determining whether people in Condition A had
higher levels of self-esteem than 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 a condition,
and (b) the standard deviation (SD), which tells us the
amount of variability there is around the mean in that
condition.
Mean Difference
• 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.
Standardized Mean Difference
• 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
Standardized Mean Difference
• Cohen’s d expresses the difference between
groups relative to the average standard
deviation of the scores.
• For Cohen's d, an effect size of 0.2 to 0.3 might
be dubbed a "small" effect. Something around
0.5 might be called a "medium" effect. And
values above .80 might be called “large” effects.
• Handy online Cohen’s d calculator:
http://web.uccs.edu/lbecker/Psy590/escalc3.htm
Another Calculation
• 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.