304 Attitudes in Social Behavior
respondents to indicate some beliefs and emotions that are
positive and some beliefs and emotions that are negative.
Using this technique, a positive-dimension score can be de-
rived from the sum or average of the positive ratings, and a
negative-dimension score can be derived from the sum or
average of the negative ratings (Bell, Esses, & Maio, 1996;
Maio, Esses, & Bell, 2000).
Separation of the positive and negative dimensions en-
ables the calculation ofambivalence,which is the simulta-
neous existence of positivity and negativity toward the
attitude object (Kaplan, 1972; Olson & Zanna, 1993). Am-
bivalence is calculated using formulas that are designed to
assess the extent to which there are high amounts of positiv-
ityandnegativity rather than a high amount of positivity or
negativity alone (e.g, Bell et al., 1996; Priester & Petty,
1996; M. M. Thompson, Zanna, & Griffin, 1995). We find it
interesting, however, that the scores derived from these for-
mulas exhibit only moderate correlations with subjective
self-reports of ambivalence (approximatelyr.40; Priester
& Petty, 1996). Thus, although the objective and subjective
measures possess some convergent validity, they must be
tapping psychological processes that are at least somewhat
distinct.
Evidence If the bidimensional view is valid, people’s
favorability toward an attitude object should at least some-
times be largely unrelated to their unfavorability toward the
object. In contrast, the unidimensional view suggests that
there should be a strong negative correlation between posi-
tivity and negativity. In support of the bidimensional view,
past research has found only moderate negative correlations
between positivity and negativity, across a variety of attitude
objects (e.g., different ethnic groups; Bell et al., 1996;
Kaplan, 1972; I. Katz & Hass, 1988; M. M. Thompson et al.,
1995; cf. Jonas, Diehl, & Brömer, 1997).
Cacioppo, Gardner, and Berntson (1997) observed that
positivity and negativity toward an object do not change in
parallel: (a) There is a tendency for people to initially pos-
sess more positivity than negativity toward attitude objects,
and (b) positivity increases more slowly than does negativ-
ity. Therefore, it is plausible that positivity and negativity
summarize different mental processes. Also, if the positive
and negative dimensions are distinct, they should exhibit
somewhat different correlations with other variables. Unfor-
tunately, researchers have not yet systematically examined
this issue.
Finally, if the bidimensional view is valid, the simultane-
ous existence of positivity and negativity (i.e., ambivalence)
should have unique psychological consequences that are
not predicted by the unidimensional model. And, indeed,
researchers have found unique consequences of ambivalence
(see the section on characteristics of attitudes later in this
chapter).Reconciling the Unidimensional and
Bidimensional PerspectivesDespite the empirical support for the bidimensional view, it
should be noted that most researchers have not examined the
correlations between positivity and negativity while simulta-
neously controlling random andsystematic measurement
error. Failure to control for both sources of error can artifac-
tually decrease the magnitude of the observed correlation
(Green, Goldman, & Salovey, 1993), leaving the impression
that the positive and negative dimensions are less strongly re-
lated than they actually are.
Even if future evidence supports the bidimensional model,
it is plausible that the unidimensional model and bidimen-
sional model are valid at different psychological levels. For
instance, the bidimensional model may apply to attitude for-
mation, in which people perceive the attitude object on both
positive and negative dimensions; these dimensions might
then be integrated to form a single, unidimensional evalua-
tion (see Cacioppo et al., 1997). Alternatively, the unidimen-
sional model may lose predictive validity as knowledge
about the attitude object becomes more complex, because it
becomes difficult to integrate the object’s positive and nega-
tive attributes.
Neither perspective on attitude dimensionality explicitly
considers implications of the fact that people can be made
aware of many different exemplars of the attitude object, in
addition to many attributes of each exemplar (Lord & Lepper,
1999). For example, when thinking of their attitude toward
cheese, people can imagine the most recent type of cheese
that they ate (e.g., fresh brie vs. processed cheese slices). The
reported attitude will depend on which exemplar is retrieved
because different exemplars are often associated with differ-
ent attributes and evaluations. Thus, it is likely that attitudes
subsume many different exemplars of the attitude object in
addition to the varied attributes of the exemplars.Alternative Attitude MeasuresPast researchers have most often measured attitudes using
self-report scales. An important limitation of self-report
scales is that they are affected by tendencies to respond in
a socially desirable manner (Paulhus, 1991). For example,
people might be reluctant to report prejudice against ethnic
groups because of the social stigma attached to prejudicial
attitudes.