Professional Documents
Culture Documents
Topics:
Correlation vs. experimentation
Warm-up:
the Hawthorne effect
Correlations can range from +1.00 (scores on one measure increase in direct proportion
to scores on another), to 0.00 (no relationship), to –1.00 (scores on one measure
decrease precisely as scores rise on the other).
Scatterplot – a graphed cluster of dots, each of which represents the values of two
variables.
The slope of the points suggests the direction of two variable’s relationship
The amount of scatter suggests the strength of the correlation (little scatter
indicates high correlation).
Correlational methods for
research: pros and cons
Research methods are a highly-examined FRQ topic for the AP exam (learn the
correct terminology for Modules 4-8 now, rather than later).
Remember correlations tell us that two variables are related, but not why they are. Only
experiments can confirm our explanations. (Understanding this point makes us better
critical thinkers).
On the other hand, statistics and data, especially scatterplots, help us see what the
naked eye often misses.
Correlations can also help us make useful predictions. But no matter how strong the
relationship, correlation can never prove causation!
Two uncorrelated variables can never be causally connected.
Correlation indicate the mere possibility of a cause-effect relationship between
variables. Why can’t it ever prove causation?
3 possible cause-effect relationships
People low in self-esteem are more likely to report depression than are those high in self-
esteem. One possible explanation of this negative correlation is that a bad self-image causes
depressed feelings. But, as the diagram indicates, other cause-effect relationships are
possible.
Illusory correlations
May also be influenced by regression to the mean (the tendency for extreme or
unusual scores or events to regress toward the average).
Scientists carefully measure whether the independent variable brings about a change (in the
dependent variable).
Unlike correlational studies (which show naturally occurring relationships), experiments are
deliberate, and can demonstrate an effect (or a lack one).
If there’s a drug in an experiment, it’s almost always the independent variable.
Double-blind procedure: (used to eliminate biases)
Neither those in the study nor those collecting the data know which group is receiving
the treatment.
Treatment’s actual effects can be separated from potential placebo effect.
Placebo effect
Effect involves results caused by expectations alone.
Random sampling (selection)
Do case studies,
No control of variables;
To observe and record naturalistic
Descriptive Nothing single cases may be
behavior observations, or
misleading
surveys
To detect naturally
occurring
Collect data on two or
relationships; to Cannot specify cause
Correlational more variables; no Nothing
assess how well one and effect
manipulation
variable predicts
another