alone, assert that one variable has caused another.
For example, suppose that an investigator discovers
a correlation between being diagnosed with schizo-
phrenia and indices suggesting high levels of the
neurotransmitter dopamine in the central nervous
system. Does this mean that schizophrenia is caused
by excessive levels of dopamine or, alternatively,
that the experience of an episode of schizophrenia
results in changes in dopamine levels? Maybe the
real culprit is a third variable. For example, many
patients with schizophrenia have a long history of
taking psychoactive drugs (e.g., amphetamines);
such long-term use could conceivably affect neuro-
transmitter levels. Therefore, the investigator must
avoid assuming that one variable causes the other
because there is always the possibility that a third
(unmeasured) variable is involved.
A classic example of thethird-variable problemis
the observation that the number of crimes is sig-
nificantly correlated with the number of churches
and synagogues in a community. Does this mean
that religioncausescrime? No. The third variable
that has been neglected in this example is popula-
tion. Both the crime rate and the number of
churches and synagogues are positively correlated
with population, and they are correlated with each
other because both increase as the population
increases.
Correlational methodscandemonstrate that a
cause–effect hypothesis is not valid. If the expecta-
tion is that variable A causes variable B, we should
at least be able to show they are correlated. Failure
to find a significant relationship most certainly con-
tradicts the hypothesis. Also, although causal infer-
ences are not possible from correlation coefficients,
this is not to say that cause–effect relationships do
not in fact exist.
However, such relationships must be demon-
strated through experimental methods (to be dis-
cussed later). There are research methods that can
help vitiate the causation problem, such as matching
participants on other variables that might be con-
tributing to the obtained relationship or using
longitudinal methods that study variables before a
given disorder develops. But these are often cum-
bersome and expensive procedures and thus are less
frequently employed.
Sometimes we are forced to use correlational
methods because we cannot either ethically or prac-
tically manipulate certain variables, such as age, sex,
marital status, or birth order. For example, we cannot
ethically train someone to abuse a child in order to
study the effects of maltreatment on PTSD. Certain32
28
24
20
16
12
8
44 8 12 16 20 24 28 32
Anxiety
r = +.76ControlF I G U R E 4-1 Scatterplot of data from Table 4-2.
T A B L E 4-2 Hypothetical Data for the
Correlation Between Anxiety
and Control
Subject Anxiety Score Control Score
Ann 26 22
Jane 24 28
Tom 20 22
Juan 20 14
Esther 16 18
Indira 12 22
Roberto 12 6
Kevin 10 14
Jasmine 6 12
Ralph 4 2
RESEARCH METHODS IN CLINICAL PSYCHOLOGY 103