Handbook of Psychology, Volume 5, Personality and Social Psychology

Structure of Attitudes 303

1995), and (c) affect is more closely related to the importance
attached to social values (Maio & Olson, 1998).

The Dimensionality of Attitudes

The three-component model and the expectancy-value model
describe the manner in which attitudes are related to beliefs,
feelings, and behavior. Neither model, however, specifies
precisely how attitudes summarize positivity and negativity
in memory. There are two prominent perspectives on this
question: the unidimensional model and the bidimensional
model.

Unidimensional Model

Guiding Assumptions The traditional perspective regards
attitudes as being unidimensional evaluations, which express
sentiments ranging from extreme unfavorability toward the
attitude object to extreme favorability toward the attitude
object. In other words, the unidimensional perspective as-
sumes that attitudes can take the form of (a) favorability,
(b) unfavorability, or (c) neither favorability nor unfavorabil-
ity toward the attitude object. Thus, a person may feel either
positively or negatively about the object, but not both at the
same time.

Measurement The most common measures of attitudes
are based on the unidimensional perspective. These measures
include bipolar semantic-differential scales, which are an-
chored by a negative adjective at one end (e.g.,bad) and a pos-
itive adjective at the other end (e.g.,good). For example,
respondents could be asked to rate their attitude toward cen-
sorship using a 7-point scale from3(very unfavorable)to
3(very favorable), with 0 (neither favorable nor unfavor-
able) in between. Respondents may be given many semantic
differential scales, anchored by different adjective pairs (e.g.,
goodvs.bad; negativevs.positive). To yield an overall index
of attitudes, responses are averaged across the scales.
Another common procedure uses Likert-like scales. This
technique utilizes many statements expressing varying
degrees of favorability or unfavorability toward the attitude
object. Examples might be Censorship unfairly restricts
access to informationandCensorship is necessary to keep
obscene material from children.People respond to each item
on a scale from2(strongly disagree)to2(strongly agree).
To yield an overall index of attitudes, responses to the items
that imply unfavorability toward the attitude object are reverse
coded (e.g.,2 changes to2), and responses to all items are
then averaged.

Evidence To some extent, the unidimensional model is

supported by findings that unidimensional measures of atti-

tude exhibit substantial criterion validity. That is, semantic-

differential and Likert scales yield attitude scores that predict

behavior (Ajzen & Fishbein, 1977; Kraus, 1995). In addition,

the unidimensional model is consistent with Judd and Kulik’s

(1980) observation that people are faster at identifying their

agreement or disagreement with extreme attitude positions

than with neutral attitude positions. These researchers argued

that this result should occur if people easily represent strong

positivity (without any negativity) and strong negativity

(without any positivity) in their minds.

Bidimensional Model

Guiding Assumptions The bidimensional model rejects

the notion that attitudes exist only on a single evaluative

continuum from negativity to positivity. Instead, the bidimen-

sional model suggests that attitudes subsume an evaluative

tendency that varies in positivity and a separate evaluative

tendency that varies in negativity. Consequently, attitudes

can take the form of (a) favorability, (b) unfavorability, (c) nei-

ther favorability nor unfavorability, and (d) both favorability

and unfavorability toward the attitude object.

Measurement To measure attitudes from the bidimen-

sional perspective, the positive and negative responses must

be assessed separately. Kaplan (1972) suggested that any sin-

gle semantic-differential scale could be split to yield separate

positive and negative dimensions. For example, researchers

could use a semantic-differential scale from3(very bad)to

0(neutral) and a semantic-differential scale from 0 (neutral)

to3(very good), rather than use a single semantic-differential

scale from3(very bad)to3(very good). In this manner, sep-

arate negative and positive dimension scores are obtained.

This approach prevents ambiguous neutral responses (Kaplan,

1972). That is, in single semantic-differential and Likert items,

neutrality may stem from an absence of both positivity and

negativity toward the attitude object, or it may stem from

the simultaneous presence of both positivity and negativity;

the split scales can differentiate between these two types of

neutrality.

Split scales may be unnecessary when an attitude measure

includes many items that assess both positive and negative

attributes of the attitude object. For example, open-ended

measures of attitude ask participants to list their beliefs about

an attitude object and the emotions that the object elicits in

them (see Esses & Maio, 2002; Haddock & Zanna, 1998).

Using a traditional semantic-differential scale, participants

then rate the valence of each response. This approach enables