those cases that are already under treatment or that
have identified themselves by seeking treatment.
We need to sample residences (block by block or
area by area), not just clinics, hospitals, and agen-
cies. Another potential problem with survey data is
that respondents may get caught up in the need to
say“the right thing.”They may want to report
only socially desirable things and deny other, less
socially desirable experiences. For example, respon-
dents may be unwilling to admit to having experi-
enced serious symptoms of psychopathology (e.g.,
auditory hallucinations) because they may be
embarrassed.
In addition, some respondents may be asked to
remember things from several years ago. Such ret-
rospective data can be subject to all sorts of distor-
tions, omissions, or embellishments. For example,
one study (Henry, Moffitt, Caspi, Langley, &
Silva, 1994) found that 18-year-olds who had
been assessed on a regular basis from birth were
not particularly accurate in their retrospective
reports of certain types of childhood experiences
(e.g., family conflict, their own depressive or anx-
ious symptoms, or their own level of hyperactivity).
These findings are noteworthy because clinical psy-
chologists often request this type of retrospective
information from clients or research participants.
The point here is that we should attempt to assess
our clients and research participantsat the time of
interestand not rely exclusively on retrospective
reports.
Recently, several large-scale, methodologically
sound epidemiological studies of mental disorder
have been conducted (e.g., the National Comorbid-
ity Study, Kessler 2005a, 2005b; the Great Smoky
Mountains Study of Youth, Costello et al., 1996).
For example, Kessler et al. (2005a, 2005b) adminis-
tered a structured diagnostic interview to a national
probability sample in the United States to obtain
estimates of the 12-month and lifetime prevalence
of a variety of mental disorders. Further, information
was obtained so that the typical age of onset for each
disorder could be estimated. Some of these results
appear in Table 4-1. Of particular interest are the
differences in age of onset among disorders. In par-
ticular, impulse-control and anxiety disorders had a
younger median age of onset (approximately 11
years old), whereas substance use disorders and
mood disorders typically occurred in late adolescence
or early adulthood (20 years old and 30 years old,
respectively). Although not presented in Table 4-1,
Kessler et al. (2005a, 2005b) reported that women
were at greater risk for developing both anxiety
and mood disorders, whereas men were at greater
risk for developing impulse-control and substance
use disorders. Therefore, being a man is a risk
factorfor impulse-control and substance use disorders.
Risk factors need not be limited to gender but can
involve other sociodemographic features as well
(socioeconomic status, age, urban vs. rural residence,
etc.).Correlational Methods
We have seen that epidemiology often relies on
correlational methods; that is, it assesses the corre-
lates (risk factors) of an illness or disorder. We now
focus more specifically on correlational methods.
These techniques enable us to determine whether
variable X is related to variable Y. For example, is a
certain pattern of scores on an intelligence test
related to specific psychiatric diagnostic categories?
Are particular patient characteristics related to ther-
apy outcomes? Is depression related to gender?The Technique. To correlate two variables, we
first obtain two sets of observations. For example,
suppose we administer two tests to 10 study par-
ticipants. One test measures anxiety and the other
measures a belief in external (rather than internal)
control. These hypothetical data are shown in
Table 4-2. When these data are correlated, the
result is a correlation coefficient; in this case, it is
+.76, indicating a strong positive relationship. As
anxiety scores increase, so do scores on belief in
external control, which means that anxiety and
feelings of lack of control are positively related.
The Pearson product-momentcorrelation coeffi-
cientis a commonly used index to determine the
degree of relationship between two variables. This
is symbolized byr, which may vary anywhere from- 1.00 to +1.00. Anrof +1.00 denotes that the two
RESEARCH METHODS IN CLINICAL PSYCHOLOGY 101