496 Adult Development and Aging
investigators beyond the use of repeated measurements
analysis of variance as a method of analyzing cross-sectional
and longitudinal data. The importance of designs with both
components is emphasized in the contemporary literature.
Health psychologists who wish to study aging should con-
sider this important contemporary approach to analysis of
longitudinal and cross-sectional data. Repeated measurement
analyses of variance results are still popular, although it pro-
vides no built-in method for correcting for attrition. It is com-
mon to “nd plots of changes in means over time. Such plots
obscure intra-individual differences, and often no single sub-
ject in the study shows a trend similar to the trend in means
over time. Reporting of changes in means over time obscures
inter-individual differences in rate of change and often pre-
sents an inaccurate picture of change over time. It is possible
to “nd that not a single subject in the study exhibits a trend
similar to the mean change over time. For an excellent
example of how longitudinal data should be presented, see
McArdle and Hamagami (1991). A simple but useful method
for adjusting for attrition is seen in the work of M. Elias et al.
(1998a) on blood pressure.
There are methods of combined cross-sectional and longi-
tudinal data analysis that allow statistical adjustment for
attrition,missing data, and data imputation.All investigators
using longitudinal designs (i.e., studies with multiple wave of
longitudinal data not simply time 1 versus time 2 difference
scores) should take full advantage of these methods. Many
contemporary methods for longitudinal analyses are re-
viewed in Collins and Horn (1991). Illustrations of problems
of attrition in studies of disease are provided in M. Elias &
Robbins•s text (1991b), as well as topics of great importance,
such as missing data, ordinal methods of assessing change,
time series applications, intra-individual differences in intra-
individual change, latent growth curve modeling, and model-
ing incomplete cross-sectional and longitudinal data using
dynamic structural equation modeling. A very creative appli-
cation of survival analysis to longitudinal studies of behavior
is found in Willett & Singer•s work (1991). Solutions to attri-
tion and missing data, as well as the pitfalls of analysis of
variance approaches to change over time and the descriptive
data that accompany these methods, are described in these
texts. An excellent example in AD research is the analysis by
Wilson, Gilley, Bennett, Beckett, and Evans (2000), in which
a large community-based population with AD was followed
and changes in cognition documented (see also Siegler,
Bosworth, & Poon, in press). It is becoming increasingly
clear that rates of health change vary dramatically among in-
dividuals; and as new, sophisticated analyses become more
accepted, researchers will be able to document changes
across individuals.
PERSONALITY AND SOCIAL FACTORSPersonality and social factors are involved in disease etiology,
although there appear to be multiple mechanisms (Contrada
& Guyll, 2001; Contrada, Leventhal, & O•Leary, 1990; Smith
& Gallo, 2001). There is evidence that personality charac-
teristics, such as hostility, operate through risky behaviors
(Scherwitz et al., 1992; Siegler, Peterson, Barefoot, &
Williams, 1992) as well as through reactivity (Williams,
1994), at least in terms of cardiovascular disease (Rozanski,
Blumenthal, & Kaplan, 1999). The evidence is less consistent
for hypertension (B. Jonas, Franks, & Ingram, 1997; S. Jonas
& Lambo, 2000; Levenstein, Kaplan, & Smith, 2000; Spiro,
Aldwin, Ward, & Mroczek, 1995), stroke (Everson, Roberts,
Goldberg, & Kaplan, 1998), diabetes (Niaura et al., 2000), and
cancer (Contrada & Guyll, 2001; Everson et al., 1996;
I. Schapiro et al., 2001). Evidence is accumulating that de-
pression and social support have a signi“cant impact on the
course of coronary heart disease (Barefoot & Schroll, 1996;
Williams & Chesney, 1993) such that a clinical trial called
Enhancing Recovery in Coronary Heart Disease (ENRICHD)
to reduce the impact of depression among myocardial infarc-
tion patients is underway (ENRICHD investigators, 2000).
Most of these studies are conducted on middle-age and
older persons; therefore, there is no doubt that these relation-
ships hold over the adult age span. To date, there has been
little interest in understanding how and why age is such a
powerful risk factor for disease and how it interacts with psy-
chosocial indicators; but this is starting to change.
Williams (2001) has reviewed the literature on these ques-
tions and concludes that after the age of 25, hostility does ap-
pear to be related to CHD incidence. It is also true that the
same characteristic that leads to an increased probability of
disease such as a heart attack, if survived, may also be related
to increased survival after the heart attack. A good example
is the Type A behavior pattern. Williams et al. (1988) showed
that under the age of 55, Type A predicted CHD; but after age
55, it was associated with lower rates of CHD. Kop (1997)
has proposed a theory to explain the stronger impact of psy-
chosocial risk on CHD at younger ages, but there is insuf“-
cient empirical evidence at present to verify his conclusions.
In particular, the National Heart, Lung, and Blood Institute
(NHLBI; 2000) considers age a risk factor for heart disease at
age 45 in men and age 55 in women. This makes the under-
standing of the role of psychosocial factors in CHD before
these ages extremely important.
Research on hostility shows that hostility is operative
across the lifecycle from age 18 to age 100 and that hostility
varies by age, race, gender, and socioeconomic status (see
Siegler, 1994, for review). Barefoot et al. (1987) showed that