Chapter 13 McCrae and Costa’s Five-Factor Trait Theory 385Basics of Factor Analysis
A comprehensive knowledge of the mathematical operations involved in factor
analysis is not essential to an understanding of trait and factor theories of person-
ality, but a general description of this technique should be helpful.
To use factor analysis, one begins by making specific observations of many
individuals. These observations are then quantified in some manner; for example,
height is measured in inches; weight in pounds; aptitude in test scores; job perfor-
mance by rating scales; and so on. Assume that we have 1,000 such measures on
5,000 people. Our next step is to determine which of these variables (scores) are
related to which other variables and to what extent. To do this, we calculate the
correlation coefficient between each variable and each of the other 999 scores. (A
correlation coefficient is a mathematical procedure for expressing the degree of cor-
respondence between two sets of scores.) To correlate 1,000 variables with the other
999 scores would involve 499,500 individual correlations (1,000 multiplied by 999
divided by 2). Results of these calculations would require a table of intercorrelations,
or a matrix, with 1,000 rows and 1,000 columns. Some of these correlations would
be high and positive, some near zero, and some would be negative. For example, we
might observe a high positive correlation between leg length and height, because one
is partially a measure of the other. We may also find a positive correlation between
a measure of leadership ability and ratings on social poise. This relationship might
exist because they are each part of a more basic underlying trait—self-confidence.
With 1,000 separate variables, our table of intercorrelations would be quite
cumbersome. At this point, we turn to factor analysis, which can account for a
large number of variables with a smaller number of more basic dimensions. These
more basic dimensions can be called traits, that is, factors that represent a cluster
of closely related variables. For example, we may find high positive intercorrela-
tions among test scores in algebra, geometry, trigonometry, and calculus. We have
now identified a cluster of scores that we might call Factor M, which represents
mathematical ability. In similar fashion, we can identify a number of other factors,
or units of personality derived through factor analysis. The number of factors, of
course, will be smaller than the original number of observations.
Our next step is to determine the extent to which each individual score con-
tributes to the various factors. Correlations of scores with factors are called factor
loadings. For example, if scores for algebra, geometry, trigonometry, and calculus
contribute highly to Factor M but not to other factors, they will have high factor
loadings on M. Factor loadings give us an indication of the purity of the various
factors and enable us to interpret their meanings.
Traits generated through factor analysis may be either unipolar or bipolar.
Unipolar traits are scaled from zero to some large amount. Height, weight, and
intellectual ability are examples of unipolar traits. In contrast, bipolar traits extend
from one pole to an opposite pole, with zero representing a midpoint. Introversion
versus extraversion, liberalism versus conservatism, and social ascendancy versus
timidity are examples of bipolar traits.
In order for mathematically derived factors to have psychological meaning,
the axes on which the scores are plotted are usually turned or rotated into a specific
mathematical relationship with each other. This rotation can be either orthogonal