Chapter 5

advertisement
Correlational Designs
Causal Modeling
Quasi-Experimental Designs
How do quasi-experiments differ from actual
experiments?
Quasi means “seeming like.” Quasiexperiments superficially resemble
experiments, but lack their required manipulation
of antecedent conditions and/or random
assignment to conditions.
Correlational Designs
How do quasi-experiments differ from actual
experiments?
They may study the effects of preexisting
antecedent conditions—life events or subject
characteristics—on behavior.
A quasi-experiment might compare the incidence
of Alzheimer’s disease in patients who used
ibuprofen since age 50 and those who did not.
Correlational Designs
How do quasi-experiments differ from actual
experiments?
In experiments, researchers randomly assign
subjects to antecedent conditions that they
create.
An experiment might randomly assign subjects
to either daily ibuprofen or aspirin use, and then
measure their incidence of Alzheimer’s.
Correlational Designs
When should we use quasi-experiments instead of
experiments?
We should use quasi-experiments when we
cannot or should not manipulate antecedent
conditions.
Quasi-experiments could study the effect of
spouse abuse on the frequency of child abuse.
Correlational Designs
Describe the properties of a correlation.
A Pearson correlation coefficient is used to
calculate simple correlations (between two
variables) and may be expressed as: r (50) =
+.70, p = .001.
Correlation coefficients have four properties.
linearity, sign, magnitude, and probability.
Correlational Designs
Describe the properties of a correlation.
Linearity means how the relationship between x
and y can be plotted as a line (linear
relationship) or a curve (curvilinear relationship).
Sign refers to whether the correlation coefficient
is positive or negative.
Correlational Designs
Describe the properties of a correlation.
Magnitude is the strength of the correlation
coefficient, ranging from -1 to +1.
Probability is the likelihood of obtaining a
correlation coefficient of this magnitude due to
chance.
Correlational Designs
What does a scatterplot show?
Scatterplots are a graphic display of pairs
of data points on the x and y axes.
A scatterplot illustrates the linearity, sign,
magnitude, and probability (indirectly) of a
correlation.
Correlational Designs
How does range truncation affect correlation
coefficients?
Range truncation is an artificial restriction of the
range of X and Y that can reduce the strength of
a correlation coefficient.
Correlational Designs
How do outliers affect correlations?
Outliers are extreme scores. They usually affect
correlations by disturbing the trends in the data.
Range truncation removes outliers.
Correlational Designs
Why should we compute the coefficient of
determination?
The coefficient of determination (r2) estimates
the amount of variability that can be explained by
a predictor variable.
For example, Chaplin et al. (2000) showed that
handshake firmness accounted for 31% of the
variability of first impression positivity.
Correlational Designs
Why doesn't correlation prove causation?
Since correlational studies do not create multiple
levels of an independent variable and randomly
assign subjects to conditions, they cannot
establish causal relationships.
Correlational Designs
Why doesn't correlation prove causation?
There are three additional reasons that
correlations cannot prove causation:
(1) casual direction
(2) bidirectional causation
(3) the third variable problem
Correlational Designs
Why doesn't correlation prove causation?
Causal direction
Since correlations are symmetrical, A could
cause B just as readily as B could cause A.
Does insomnia cause depression or does
depression cause insomnia?
Correlational Designs
Why doesn't correlation prove causation?
Bidirectional causation
Two variables—insomnia and depression—
may affect each other.
Correlational Designs
Why doesn't correlation prove causation?
Third variable problem
A third variable—family conflict—may create the
appearance that insomnia and depression are
related to each other.
Correlational Designs
When do researchers use multiple correlation (R)?
Researchers use multiple correlation (R) when
they want to know whether there is a relationship
among three or more variables.
We could measure age, television watching, and
vocabulary and find that R = +.61.
Correlational Designs
When should we compute a partial correlation?
We should compute a partial correlation when
we want to hold one variable (age) constant to
measure its influence on a correlation between
two other variables (television watching and
vocabulary).
Correlational Designs
When do researchers use multiple regression?
Researchers use multiple regression to predict
behavior measured by one variable based on
scores on two or more other variables.
We could estimate vocabulary size using age
and television watching as predictor variables.
Correlational Designs
What is causal modeling?
Causal modeling is the creation and testing
of models that suggest cause-and-effect
relationships between behaviors.
Path analysis and cross-lagged panel
designs are two forms of causal modeling.
Causal Modeling
Explain path analysis.
In path analysis, a researcher creates and tests
models of possible causal sequences using
multiple regression analysis where two or more
variables are used to predict behavior on a third
variable.
Causal Modeling
What is a cross-lagged panel design?
In cross-lagged panel design, a researcher
measures relationships over time and these are
used to suggest a causal path.
Causal Modeling
What is an ex post facto design?
Ex post facto means “after the fact.” A
researcher examines the effects of already
existing subject variables (like gender or
personality type), but does not manipulate them.
Quasi-Experimental Designs
What is a nonequivalent groups design?
A nonequivalent groups design compares the
effects of treatments on preexisting groups of
subjects.
A researcher could install fluorescent lighting in
Company A and incandescent lighting in
Company B and then assess productivity.
Causal Modeling
Describe the longitudinal and cross-sectional
approaches.
In longitudinal designs, the same group of
subjects is measured at different points of time
to determine the effect of time on behavior.
In cross-sectional studies, subjects at different
developmental stages (classes) are compared at
the same point in time.
Causal Modeling
What is a pretest/posttest design?
In pretest/posttest designs, a researcher
measures behavior before and after an event.
This is quasi-experimental because there is no
control condition.
For example: Practice GRE test 1  six-week
preparation course  Practice GRE test 2.
Causal Modeling
Which problems reduce its internal validity?
There is no control group which receives a
different level of the IV (no preparation course).
The results may be confounded by practice
effects (also called pretest sensitization) due
to less anxiety during the posttest and learning
caused by review of pretest answers.
Causal Modeling
What is a Solomon 4-group design?
This variation on a pretest/posttest design
includes four conditions:
(1) a group that received the pretest, treatment
and posttest
(2) a nonequivalent control group that
received only the pretest and posttest
Causal Modeling
What is a Solomon 4-group design?
(3) a group that received the treatment and
a posttest
(4) a group that only received the posttest
Causal Modeling
Download