10.1177/0146167204263787PERSONALITY AND SOCIAL PSYCHOLOGY BULLETINFederico / THE NEED TO EVALUATE AND ATTITUDE EXTREMITY
Predicting Attitude Extremity:
The Interactive Effects of Schema Development
and the Need to Evaluate and Their Mediation
by Evaluative Integration
Christopher M. Federico
University of Minnesota, Twin Cities
Research on attitude extremity suggests that schemas containing
more information about a particular attitude domain are more
likely to be associated with extreme attitudes toward objects in
that domain when perceivers' responses toward features of the
domain are evaluatively integrated. The present study argues
that a high need to evaluate may play an important role in determining
when schema development will be associated with the
integrated responses to different domain features necessary for
extremity. Consistent with this argument, data from a nationally
representative survey of political attitudes indicated that the
need to evaluate was associated with increased extremity across
two different indices of the latter; that it moderated the relationships
between schema development (in the form of political expertise),
on one hand, and increased extremity and integration, on
the other; and that the moderating effects of the need to evaluate
vis--vis extremity were mediated by integration.
Keywords: attitude extremity; need to evaluate; political expertise;
attitude structure
Social and political psychologists have long shown an
interest in factors that may lead people to adopt extreme
attitudes (see Abelson, 1995; Judd & Brauer, 1995; Myers
& Lamm, 1976; Tesser, Martin, & Mendolia, 1995;
Sidanius, 1988). In this respect, one variable that has
repeatedly received attention from researchers is the
degree to which people possess well-developed stores of
knowledge about a given attitude domain. More precisely,
it often has been argued that individuals whose
attitude-related schemas contain information about a
greater number of domain features may express more
extreme attitudes toward objects in that domain (e.g.,
Judd & Lusk, 1984; Millar & Tesser, 1986). Evidence consistent
with this argument has come from a number of
different sources. For example, in the context of political
attitudes, a number of studies have demonstrated
that political experts (who tend to have more welldeveloped
schemas for the political domain; see Fiske,
Lau, & Smith, 1990; Lodge & Hamill, 1986) tend to
express attitudes that deviate more extremely from neutral
or centrist positions (Lusk & Judd, 1988; Sidanius,
1988; Sidanius & Lau, 1989). Moreover, a variety of studies
suggest that domain-relevant thought results in increased
attitude polarization, but only among individuals
who possess more extensive stores of knowledge
about the domain in question (for reviews, see Judd &
Brauer, 1995; Tesser et al., 1995).
Effects of this sort are typically explained in terms of
the constraining effects of schema development1
(e.g.,
Millar & Tesser, 1986). Rather than being amorphous
clusters of attitude-relevant information, well-developed
schemas provide perceivers with a structured representation
of social reality and their prior evaluative responses
to it. As a result, information processing that is
guided by well-developed schemas tends to interpret
new, evaluatively relevant information in terms of prior
evaluations. Over time, information that is consistent
1281
Author's Note: Christopher M. Federico is assistant professor of psychology
and political science at the University of Minnesota, Twin Cities.
The author would like to thank the Inter-University Consortiumfor
Political and Social Research for the 2000 National Election Study data
and Mike Sargent, Jon Krosnick, and Agnieszka Golec for their comments
and suggestions. Parts of this article were presented at the annual
meeting of the International Society for Political Psychology,
Boston, MA, July 2003. Please direct all correspondenceto Christopher
M. Federico, Department of Psychology, University of Minnesota, 75 E.
River Road, Minneapolis, MN 55455; e-mail: federico@umn.edu.
PSPB, Vol. 30 No. 10, October 2004 1281-1294
DOI: 10.1177/0146167204263787
 2004 by the Society for Personality and Social Psychology, Inc.
with one's initial evaluation of objects in the relevant domain
is more likely to be added to existing stores of
knowledge, leading to increased extremity (see Judd &
Brauer, 1995). Thus, schema development appears to
promote extremity by virtue of the fact that schemaguided
thought is more likely to be systematic and
directed.
Despite the cogency of this basic account, other studies
have suggested that the possession of well-developed
schemas may not always be associated with attitude extremity.
For example, Patricia Linville and her colleagues
(e.g., Linville, 1982; Linville & Jones, 1980) have
argued that well-developed schemas may actually militate
against the formation of extreme attitudes. According
to this account, the fact that individuals with welldeveloped
schemas characterize objects in a given domain
in terms of information about a greater number of
features should increase the likelihood that their representation
of that domain and objects in it will consist of
both positively and negatively evaluated features. As
such, the positives and negatives should cancel one
another out, resulting in a less extreme evaluation. Consistent
with this argument, Linville and Jones (1980)
found that the same set of characteristics was less likely to
produce an extreme evaluative response when attributed
to ingroup members rather than outgroup members,
a difference that was mediated by the fact that
participants' ingroup schemas were better developed
than their outgroup schemas.
Although these two perspectives would seem to be at
odds with one another, subsequent work has converged
on a useful synthesis. More specifically, a great deal of
research now suggests that the nature of the relationship
between schema development--that is, the possession of
knowledge structures containing information about a
greater number of domain features--and attitude extremity
depends on the degree to which the represented
features are evaluatively integrated or redundant (Judd
& Lusk, 1984; Millar & Tesser, 1986; Tesser et al., 1995;
see also Judd & Brauer, 1995; Liberman & Chaiken,
1991; Sidanius, 1988). When responses to different features
of an attitude domain and objects in it are similar
or highly correlated with one another, knowledge structures
containing information about a greater number of
features tend to be associated with increased extremity.
However, when responses to different features are relatively
orthogonal, schema development is associated
with reduced extremity. This has been frequently demonstrated
in studies that have measured individual differences
in integration (e.g., Millar & Tesser, 1986; see
also Tesser et al., 1995). However, studies that have
manipulated the perceived redundancy of the different
dimensions used to organize a given attitude domain
have produced the same result (e.g., Judd & Lusk, 1984),
providing evidence for the suggested causal role for
integration.
Put another way, these findings suggest that the possession
of knowledge structures incorporating a greater
number of features is more likely to be associated with
extreme attitudes when perceivers are not ambivalent
about the relevant domain and objects in it, that is, when
their evaluative responses to differently valenced domain
features are not in conflict with one another (e.g.,
Lavine, Borgida, & Sullivan, 2000; Liberman & Chaiken,
1991; Priester & Petty, 1996; Thompson, Zanna, &
Griffin, 1995). For example, in the domain of politics,
we might expect experts to display increased extremity
along the liberal-conservative dimension as long as they
respond to liberal and conservative features of the political
domain (e.g., major political figures) in an integrated,
univalent fashion, that is, evaluating liberal
features positively if they evaluate conservative features
negatively, and vice versa.
MOTIVATIONAL INFLUENCES ON ATTITUDE EXTREMITY:
A MODERATING ROLE FOR THE NEED TO EVALUATE?
Research thus suggests that the possession of schemas
that represent a greater number of domain features are
more likely to be associated with extreme attitudes toward
objects in that domain when perceivers' responses
to different features of the domain are evaluatively integrated.
However, existing work offers far less insight into
when these two conditions should co-occur, leading the
possession of well-developed schemas to be associated
with attitude extremity. Although various studies suggest
that individuals whose schemas contain information
about a greater number of features are also more likely
to evaluate those features in an integrated fashion (e.g.,
Zaller, 1992; see also Hamill, Lodge, & Blake, 1985; Judd
& Krosnick, 1989; Sidanius, 1988), the correlation is
far from perfect. For example, to return to the political
case, it is increasingly clear that schema development in
the form of political expertise does not always explain
variance in evaluative redundancy among different political
objects (such as political figures; see Federico,
2003; Goren, 2001). In other words, it is not fully clear
when the conditions necessary for a relationship between
schema development and integration should obtain.
By extension, it is also unclear when conditions
conducive to the emergence of a relationship between
schema development and extremity should be present.
However, research in other domains increasingly suggests
that a variety of motivational factors may determine
how individuals organize and apply attitude-relevant information
(see Cacioppo, Petty, Feinstein, & Jarvis, 1996;
Federico, 2003; Jarvis & Petty, 1996; Kruglanski, 1996;
Lavine, 2002). According to work of this sort, the information
contained in attitude-relevant knowledge struc-
1282 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
tures may serve as a kind of raw material in processes of
attitude formation, whereas motivational variables determine
how this raw material is used to construct evaluations.
Although this interactive perspective has not received
a great deal of attention in the attitude-extremity
literature, some of its insights suggest that a consideration
of these motivational variables may tell us something
about the set of conditions under which schema
development is likely to be associated with attitude
extremity.
In this vein, one motivational factor that has received
a great deal of recent attention is the need to evaluate
(Jarvis & Petty, 1996), which refers to the extent to which
an individual is motivated to spontaneously form evaluations
of various objects as either "good" or "bad." Persons
with a high need to evaluate think more frequently
in evaluative terms and are more likely to have formed
opinions about a variety of objects in a greater number of
attitude domains, whereas those with a low need to evaluate
tend to have fewer evaluative thoughts across objects
and domains. In other words, those with a high need to
evaluate derive enjoyment from the process of assessing
the good and bad features of the things they encounter.
Consistent with this conceptual definition, research suggests
that individuals with a high need to evaluate are
more likely to spontaneously evaluate various objects,
even unfamiliar ones that are difficult to assess in any
straightforward fashion (see Jarvis & Petty, 1996). Moreover,
in the political domain, other analyses suggest that
those with a high need to evaluate are more likely to engage
in behaviors indicative of well-formed preferences,
such as voting and news media use (Bizer et al., 2003).
In addition, other studies have suggested that those
with a high need to evaluate are more likely to represent
their opinions in an "on-line" fashion (Tormala & Petty,
2001); that is, individuals who are strongly motivated to
evaluate objects are also more likely to establish standing
decisions about those objects in memory, which can then
be easily accessed when offering an opinion. As a result,
they respond more quickly when asked to provide an
opinion. In contrast, individuals with a low need to evaluate
are less likely to have established standing attitudinal
decisions, which means they must construct their opinions
in "memory-based" fashion on the basis of whatever
attitude-relevant information is accessible at a given
time. Consequently, those with a low need to evaluate
take longer to express an attitude (see Tormala & Petty,
2001, for a more comprehensive statement).
So, what implications might this variable have for the
development of extreme attitudes? As noted above, the
relationship between schema development and attitude
extremity appears to be contingent on the degree to
which the contents of attitude-relevant schemas are evaluatively
integrated. Thus, variables that encourage perceivers
to use attitude-relevant information in a univalent,
integrated fashion also should strengthen the relationship
between schema development and extremity.
Given its effects in other domains, the need to evaluate
may be one of these variables. As we have seen, a number
of studies suggest that it is generally associated with a tendency
to engage in thought aimed at the formation of
standing evaluative decisions ( Jarvis & Petty, 1996). However,
research also suggests that the need to evaluate may
be very specifically associated with a tendency to use
existing stores of information in an evaluative fashion.
For example, in the political domain, Bizer and his colleagues
(2003) have shown that individuals with a high
need to evaluate are more likely to evaluate candidates
with regard to how well the candidates' issue positions
accord with their own. Moreover, Federico (2003) has
shown that individuals with highly developed political
schemas (i.e., political experts) are more likely to align
their positions on specific political issues with their overall
ideological orientation if they have a higher need to
evaluate.
These results suggest that a high need to evaluate may
play an important role in emergence of attitude extremity.
In particular, it may do so by determining when
schema development will be associated with the evaluative
integration necessary for the expression of extreme
attitudes. This general claim leads to a number of specific
hypotheses. At the simplest level, it suggests that the
need to evaluate may be associated with increased extremity
and evaluative integration because the need to
evaluate has been linked to both a propensity for evaluation
in general and evaluatively integrated response
patterns.
However, it also suggests a number of more complex
relationships. Most important, it implies that the association
between the possession of well-developed schemas
and increased attitude extremity in a given domain may
be strongest among those with a high need to evaluate.
Moreover, if the need to evaluate strengthens the relationship
between schema development and extremity by
encouraging univalence in the relevant attitude domain,
then this moderating effect of the need to evaluate
should itself be mediated by evaluative integration. In
turn, the perspective developed here suggests that
evaluative integration may mediate the key interaction
between schema development and the need to evaluate
in two ways (Baron & Kenny, 1986; Lance, 1988;
Wegener & Fabrigar, 2000). On one hand, schema development
and the need to evaluate may have an interactive
effect on evaluative integration, which in turn has a direct
effect on extremity. This suggests that the interactive
effect of schema development and the need to evaluate
on extremity is due to the interactive effect of schema development
and the need to evaluate on a key antecedent
Federico / THE NEED TO EVALUATE AND ATTITUDE EXTREMITY 1283
of extremity, that is, evaluative integration. This hypothesis
is tested by looking at whether the addition of
evaluative integration to a model already containing
schema development, the need to evaluate, and the interaction
between the two reduces the predictive power
of the interaction between schema development and the
need to evaluate. On the other hand, the need to evaluate
may be directly associated with increased evaluative
integration, which in turn moderates the relationship
between schema development and extremity. This suggests
that the interactive effect of schema development
and the need to evaluate on extremity is actually accounted
for by the interactive effect of schema development
and evaluative integration on extremity. This hypothesis
is tested by looking at whether the addition of
the interaction between schema development and
evaluative integration to a model already containing
schema development, the need to evaluate, the interaction
between schema development and the need to
evaluate, and the main effect of integration reduces the
predictive power of the interaction between schema
development and the need to evaluate. In the analyses
that follow, both of these "mediated moderation" hypotheses
are examined.
OVERVIEW OF THE ANALYSES
The purpose of this study was to examine the preceding
hypotheses in the context of political attitudes. To do
this, data from a large, nationally representative sample
of adults were analyzed. Consistent with the aforementioned
hypotheses, four specific predictions were examined:
(a) that the need to evaluate would be associated
with more extreme attitudes toward objects in the political
domain, even after other antecedents of extremity
were considered; (b) that political expertise--perhaps
the best indicator of political schema development (e.g.,
Fiske et al., 1990)--would be more strongly associated
with extremity among those with a high need to evaluate;
(c) that expertise would be more strongly associated
with evaluatively integrated responses toward features of
the political attitude domain (i.e., responses to liberal
and conservative political figures) among those with a
high need to evaluate; and (d) that this evaluative integration
would mediate the moderating effects of the
need to evaluate vis--vis the relationship between expertise
and extremity.
METHOD
Data
The data were taken from the 2000 National Election
Study (NES), conducted by the Center for Political Studies
at the University of Michigan (N = 1,807). This survey
contained measures of attitudes, political expertise, and
most important, the short form of the Need to Evaluate
scale (see Bizer et al., 2003). The 2000 NES interviewed
respondents both before and after the 2000 election
using a nationally representative sample. Respondents
were randomly assigned to be interviewed either face to
face or via telephone. The preelection interviews were
conducted between September 5 and November 6, producing
a response rate of 64.3% in the face-to-face mode
and 56.5% via telephone. In the postelection panel,
1,555 of the preelection respondents were interviewed
again between November 8 and December 21. The response
rate for this wave was 86% in the face-to-face
mode and 85.9% via telephone.
Independent Variables
Political expertise. In this study, expertise was used as
a proxy for the possession of well-developed political
schemas. It was measured using several factualknowledge
items, consistent with work suggesting that
the latter are the most valid indicators of the differences
in political cognition that should go along with varying
levels of schema development (see Delli Carpini &
Keeter, 1996). In this vein, Fiske and her colleagues
(1990) looked at the relationship between five expertise
measures (i.e., political knowledge, political activity,
electronic media use, print media use, and political selfschematicity)
and a series of cognitive measures and
found that political knowledge was the most useful in
predicting information-processing benefits suggested by
theoretical models of political schematicity. These
included reduced reading time for political materials,
quicker decision making, and improved recall for relevant
stimulus items. Other analyses indicate that knowledge
may be the best available index of the degree to
which survey respondents are able to perceive implicational
links among related political ideas and use ideological
considerations to structure their policy attitudes.
For example, a number of studies have indicated that
factual political knowledge is a stronger predictor of attitude
constraint and response stability than education
(Judd, Krosnick, & Milburn 1981; Zaller 1990, 1992).2
Eight items were included in the actual scale. These
asked respondents to indicate (a) which party controlled
the House of Representatives prior to the 2000 election,
(b) which party controlled the Senate prior to the election,
(c) the office held by Trent Lott, (d) the office held
by William Rehnquist, (e) the office held by Tony Blair,
(f) the office held by Janet Reno, (g) which state George
W. Bush lived in at the time of the 2000 election, and (h)
which state Al Gore was from. These items formed a reliable
scale (with a KR-20 coefficient of .88, M = .40, SD =
.29).
1284 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
Need to evaluate. This was measured using a shorter version
of the original scale developed by Jarvis and Petty
(1996). The 2000 NES asked two questions about the
degree to which a person was prone to evaluative
thought. The first question asked, "Some people have
opinions about almost everything; other people have
opinions about just some things; and still other people
have very few opinions. What about you?" Responses
included almost everything, about many things, about some
things, or about very few things. The second question
asked, "Compared to the average person, do you have
fewer opinions about whether things are good or bad,
about the same number of opinions, or more opinions?"
Those who responded that they had fewer or more opinions
than average were asked a follow-up question where
they were asked if they had "a lot" or "somewhat" fewer
or more opinions. Respondents' scores in both samples
were recoded on a 0 to 1 scale, where 1 indicated a high
need to evaluate and a 0 indicated a low need to evaluate. The
scores on the individual items were averaged to form the
need to evaluate scale. Together, the items formed a reliable
scale ( =.70, M = .56, SD = .23).
Need for cognition. To provide a control for respondents'
general tendency to exert cognitive effort--
as opposed to cognitive effort aimed specifically at
evaluation--we also included a short-form measure of
the need for cognition (Bizer et al., 2003; Cacioppo
et al., 1996). Because a general tendency to engage in
cognitive elaboration--as opposed to specifically
evaluative forms of elaboration--also may be associated
with attitude extremity (Cacioppo et al., 1996), both the
main effect of the need for cognition and its interaction
with expertise also were considered in the analyses that
follow. The measure was based on two NES questions.
The first question asked, "Some people like to have
responsibility for handling situations that require a lot of
thinking, and other people don't like to have responsibility
for situations like that. What about you?" Respondents
were given the option to say that they neither liked
nor disliked thinking situations. However, those who
responded that they did like or dislike thinking situations
were asked a follow-up question where they were asked if
they liked or disliked thinking situations a lot or somewhat.
The second question asked, "Some people prefer
to solve simple problems instead of complex ones,
whereas other people prefer to solve more complex
problems. Which type of problem do you prefer to solve:
simple or complex?" Again, responses to these questions
were recoded on a 0 to 1 scale, where a 1 denotes a high
need for cognition and 0 denotes a low need for cognition.
These items also formed a reliable scale ( =.67, M = .60,
SD = .35).3
Evaluative integration. To assess evaluative integration,
some index of the degree to which respondents' assessment
of different features of the political attitude
domain had evaluatively consistent implications was
needed (see Judd & Brauer, 1995; Tesser et al., 1995).
Previous studies have typically measured this characteristic
in terms of within-subject correlations between ratings
of domain features on different attributes (e.g.,
Judd & Lusk, 1984) or the level of ambivalence in people's
evaluative responses to various domain features
(e.g., Liberman & Chaiken, 1991). For the present study,
the latter strategy was chosen: integration was measured
in terms of the absence of ambivalence in people's
responses to differently valenced political figures, that is,
a tendency to evaluate liberal and conservative figures
differently and oppositely (Priester & Petty, 1996;
Thompson et al., 1995; Zaller, 1992). To this end, the following
modified index based on Thompson et al.'s
(1995) similarity-intensity model of ambivalence was
used:
Integration = ­ [[(C + L) / 2] ­ |C ­ L|],
where C is the respondent's rating of conservative figures
and L is the respondent's rating of liberal figures.
Scores on this index increase as respondents' ratings of
liberal and conservative attitude objects become less ambivalent
and more evaluatively integrated (Lavine et al.,
2000; Thompson et al., 1995).4
This choice of rating targets
and indices had a number of advantages. First, although
the ratings used to construct the index clearly
tap into the same overall domain as the extremity measures,
they do not rely on responses to the same stimuli
(i.e., the left-right continuum and specific issues; see below).
This reduced the risk of obtaining estimates of the
relationship between integration and extremity that
were inflated by shared method variance. Second, unlike
correlation-based measures, which look at relationships
among dimensions that place oppositely valenced domain
features at different ends of the same continuum,
the measure used here allows oppositely valenced features
to be evaluated separately. This makes it easier to
clearly assess the tendency to evaluate oppositely
valenced features differently, which is central to the concept
of evaluative integration (see Judd & Brauer, 1995;
Liberman & Chaiken, 1991; Priester & Petty, 1996).5
Finally, more so than correlation-based measures, the
ambivalence-based measure used here assesses similarity
in the intensity as well as the valence of evaluations (see
Thompson et al., 1995; see also Craig, Kane, & Martinez,
2002; Lavine et al., 2000), thereby avoiding underestimation
of the amount of integration present in a
given domain.
Federico / THE NEED TO EVALUATE AND ATTITUDE EXTREMITY 1285
Raw data for this index was obtained by looking at respondents'
ratings of eight political figures. Four of the
figures were associated with relative liberalism (i.e., Bill
Clinton, Al Gore, Joe Lieberman, and Hillary Clinton)
and the other four were associated with relative conservatism
(i.e., George W. Bush, Pat Buchanan, John
McCain, and Dick Cheney). Respondents rated each figure
on a 100-point feeling thermometer, with 0 indicating
negative affect and 100 indicating positive affect. These
scores were recoded to run from 0 to 1. Composite measures
of respondents' feelings about liberal and conservative
features of the attitude domain were then created
by averaging ratings of each set of four figures ( =.87, M
= .56, SD = .23, for the liberal figures;  =.66, M = .53, SD =
.17, for the conservative figures). These indices were
then used as the conservative (C) and liberal (L) terms in
the above formula. For each subject, the resulting scores
were recoded to run from 0 to 1; higher scores indicated
greater evaluative integration (or a relative lack of ambivalence;
M = .48, SD = .17).
Demographics and education. Several demographics
also were considered: age (in years), income (in thousands
of dollars per year), race (0 = non-White, 1 =
White), and gender (0 = female, 1 = male). Finally,
because both expertise (Delli Carpini & Keeter, 1996)
and the need to evaluate (Bizer et al., 2003) correlate
with higher levels of educational attainment, an index of
educational attainment also was included in the analyses.
Because earlier work on the role of education has
focused on the completion of a college degree as the critical
experience responsible for the development of complex
attitude structures in the domain of politics (e.g.,
Judd & Milburn, 1980; Sniderman, Brody, & Tetlock,
1991), a dummy variable indicating whether the respondents
had completed a college degree (0 = no, n = 1,251;
1 = yes, n = 556) was used.6
The Dependent Variable: Attitude Extremity
The overall extremity of respondents' attitudes toward
objects in the political domain served as the dependent
variable in this study. This was operationalized
in two ways. First, an index of ideological extremity was
constructed by looking at respondents' deviation from
the center of ideological space on the left-right continuum.
This continuum is typically conceptualized as the
most general evaluative dimension in the political domain,
and it is thought to organize responses to particular
objects within that domain (e.g., specific policies; see
Converse, 1964). As such, this index served as a useful
summary measure of respondents' tendency to hold relatively
extreme political opinions (for a similar use of
this measure, see Sidanius, 1988; Sidanius & Lau, 1989).
Respondents' self-placements on the 7-point NES ideology
scale were used as the raw data for this index. Responses
were recoded to run from 0 to 1, with higher
scores indicating greater conservatism (M = .55, SD =
.27). The measure of extremity was created by taking the
absolute value of the difference between each respondent's
scale score and the midpoint of .5, indicating
moderate self-placement (M = .49, SD = .27).7
However, for the purposes of the present study, an extremity
index based on responses to a variety of specific
objects in the relevant domain also was desirable (see
Judd & Lusk, 1984; Millar & Tesser, 1986). Thus, an
index of issue extremity was constructed by looking at
respondents' deviation from the midpoint across 13
issue-attitude items. The items were government services
and spending, defense spending, national health insurance,
government-guaranteed jobs, aid to Blacks, affirmative
action, women's rights, abortion, gays in the military,
school vouchers, environmental protection versus
job preservation, support for laws protecting gays from
job discrimination, and gun control. Respondents answered
each item on a 5-point scale. Responses to these
items scaled reliably ( = .70) so they were averaged.
Scores on this composite were recoded to run from 0 to 1
(with higher scores indicating greater conservatism).
The scale was created by taking the absolute value of the
difference between each respondent's scale score and .5
(M = .14, SD = .11).
RESULTS
Intercorrelations between the five key variables are
presented in Table 1. As these coefficients indicate, expertise,
the extremity indices, and the proposed moderators
and mediators of the relationship between them
were all reliably associated with one another (all ps at
least < .05).
Schema Development and the Need to
Evaluate as Predictors of Attitude Extremity
The hypotheses that the need to evaluate would predict
attitude extremity on its own and moderate the rela-
1286 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
TABLE 1: Intercorrelations for Study Variables (2000 NES)
Variable 1 2 3 4 5
1. Political expertise --
2. Need to evaluate .23*** --
3. Need for cognition .23*** .33*** --
4. Evaluative integration .17*** .14*** .07** --
.5. Ideological extremity .14** .18*** .05* .25*** --
6. Issue extremity .04 .08*** .03 .19*** .17***
NOTE: All coefficients are Pearson correlations. 2000 NES = 2000 National
Election Study.
*p < .05. **p < .01. ***p < .001.
tionship between schema development and extremity
were tested using a series of hierarchical ordinary leastsquares
regression models. In these models, each extremity
index was regressed on expertise (i.e., as a proxy
for schema development), the need to evaluate, need for
cognition, and the three two-way interactions between
these variables. Age, income, race, gender, and the
college-degree indicator were included in each model as
well. Because education is typically associated with
expertise (e.g., Delli Carpini & Keeter, 1996) and may
function in similar ways, the interaction between education
and the need to evaluate was included as well. To
guard against possible effects of heteroskedasticity,
HC3 robust standard errors were used in these analyses
(as recommended by Long & Ervin, 2000). Finally, in
accordance with Aiken and West's (1991) suggestions
for interactive models, all predictors were centered
prior to this analysis.8
The results of this analysis are shown in Table 2. For
each extremity index, Model 1 simply examined the
main effects of expertise, need to evaluate, and need for
cognition in a multivariate context. As expected, the estimates
for this step indicated that the need to evaluate
was independently related to ideological extremity (b =
.19, p < .001) and issue extremity (b = .03, p < .01), even
after the effects of expertise, a generalized propensity
for cognitive elaboration (i.e., the need for cognition),
the college-degree variable, and the demographics were
considered. Moreover, expertise was related to both
ideological extremity (b = .09, p < .001) and issue extremity
(b = .03, p < .01).
Thus, individuals who were more highly motivated to
engage in evaluative thought were also more likely to display
extreme attitudes. However, the potential moderating
effects of this variable were of somewhat greater interest
here. To test this critical interactive hypothesis for
each extremity index, Model 2 added the critical two-way
interaction between expertise and the need to evaluate
as well as the Expertise × Need for Cognition, Need to
Evaluate × Need for Cognition, and College Degree ×
Need to Evaluate interactions. For ideological extremity,
this step revealed two significant interactions. The key
interaction between expertise and the need to evaluate
was significant and in the predicted positive direction
(b = .27, p < .05), suggesting that the relationship between
expertise and extremity was stronger among those
high in the need to evaluate. There was also a significant
negative interaction between expertise and the need for
cognition (b = ­.16, p < .05), suggesting that a general
tendency to engage in cognitive elaboration may actually
weaken the expertise-extremity relationship. The
College Degree × Need to Evaluate interaction failed to
reach significance (p > .10). For issue extremity, Model 2
revealed only the predicted significant interaction between
expertise and the need to evaluate, which was
again significant and positive (b = .14, p < .01).
To probe these interactions, simple slopes for the relationship
between expertise and extremity were comFederico
/ THE NEED TO EVALUATE AND ATTITUDE EXTREMITY 1287
TABLE 2: Expertise and Need to Evaluate as Predictors of Attitude Extremity (2000 NES)
Ideological Extremity Issue Extremity
Model 1 Model 2 Model 1 Model 2
Predictor b SE b b SE b b SE b B SE b
Age .0004 (.0004) .0003 (.0004) ­.0004 (.0002) ­.0004 (.0002)
Income ­.002 (.002) ­.002 (.002) .001 (.001) .001 (.001)
Race .03 (.02) .03 (.02) ­.04*** (.01) ­.03*** (.01)
Gender .02 (.02) .02 (.02) ­.02*** (.01) ­.02*** (.01)
College degree .003 (.01) .01 (.01) .001 (.003) .0004 (.003)
Political expertise .09*** (.03) .10*** (.03) .03** (.01) .03** (.01)
Need to evaluate .19*** (.03) .18*** (.04) .03* (.01) .03 (.01)
Need for cognition ­.01 (.02) ­.01 (.02) ­.003 (.01) ­.002 (.01)
Expertise × Need to Evaluate -- -- .27* (.12) -- -- .14** (.05)
Expertise × Need for Cognition -- -- ­.16* (.09) -- -- .03 (.03)
Need to Evaluate × Need for Cognition -- -- ­.17 (.10) -- -- .02 (.04)
College Degree × Need to Evaluate -- -- ­.05 (.04) -- -- ­.02 (.01)
Constant .45*** (.03) .46*** (.03) .19*** (.01) .19*** (.01)
F (degrees of freedom) 9.06 (8, 1,431)*** 7.48 (12, 1,427)*** 7.01 (8, 1,583)*** 5.93 (12, 1,579)***
R
2
.048 .057 .037 .046
N 1,440 1,440 1,592 1,592
NOTE: Entries are unstandardized ordinary least squares (OLS) regression coefficients and HC3 robust standard errors. 2000 NES = 2000 National
Election Study.
p < .10. *p < .05. **p < .01. ***p < .001.
puted at need-to-evaluate levels 1 standard deviation
above and below the variable's mean, using Aiken and
West's (1991) method. The relationship between expertise
and ideological extremity was positive and significant
at high levels of the need to evaluate (b = .16, SE b =
.04, p < .001) but nonsignificant at low levels of the need
to evaluate (b = .03, SE b = .04, p > .10). Similarly, the relationship
between expertise and issue extremity was positive
and significant at high levels of the need to evaluate
(b = .06, SE b = .02, p < .001) but nonsignificant at low levels
of the need to evaluate (b = ­.01, SE b = .02, p > .10).
Thus, as expected, the relationship between the possession
of well-developed schemas--in the form of domainrelevant
expertise--and attitude extremity was
significantly stronger among those with a high need to
evaluate.
Schema Development and the Need to
Evaluate as Predictors of Evaluative Integration
The argument offered here also suggests that schema
development in a given domain and the need to evaluate
may be interactively related to attitude extremity via
evaluatively integrated responses to features of the
evaluative domain tapped by the attitudes in question.
More concretely, this implies (a) that the need to evaluate
may be associated with increased integration and (b)
that it may moderate the relationship between schema
development (i.e., expertise) and integration, such that
schema development is positively related to integration
primarily among those with a high need to evaluate. This
question was examined by regressing integration on the
same set of predictors used in the analysis summarized in
Table 2. Again, all predictors were centered prior to the
analysis. The results of this analysis are summarized in
Table 3. Model 1 simply examined the key main effects,
revealing that both expertise (b = .07, p < .001) and the
need to evaluate (b = .09, p < .001) were associated with
increased integration, as predicted. Model 2, on the
other hand, examined the critical interaction. The estimates
from this model revealed the predicted interaction
between expertise and the need to evaluate (b = .15,
p < .05); none of the other interactions were significant
(all ps > .10). The significant interaction was probed
by examining the simple slopes for the expertiseintegration
relationship 1 standard deviation below and
above the mean for the need to evaluate. This analysis indicated
that expertise was not significantly related to integration
at low levels of the need to evaluate (b = .03, p >
.10) but significantly associated with increased integration
at high levels of the need to evaluate (b = .10, p <
.001). Thus, as expected, the need to evaluate was
associated with higher levels of evaluative integration on
its own and also moderated with the relationship
between schema development and integration in the
predicted fashion.
Does Evaluative Integration Mediate the
Moderating Effect of the Need to Evaluate?
The final part of the argument offered here is that the
tendency for schema development to be more strongly
associated with extremity among those high in the need
to evaluate should be mediated by evaluative integration
in the domain of interest. According to Baron and
Kenny's (1986) well-known criteria, mediation is demonstrated
when (a) the dependent variable is reliably
associated with the independent variable, (b) the independent
variable is significantly associated with the hypothesized
mediator of the relationship between the
dependent and independent variables, and (c) the net
association between the dependent variable and independent
variable is significantly reduced in a regression
containing both the independent variable and the mediator,
with the mediator remaining significant. In the
present analysis, this basic procedure was modified
slightly to test for the two forms of "mediated moderation"
described earlier (cf. Baron & Kenny, 1986; Lance,
1988; Wegener & Fabrigar, 2000).
1288 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
TABLE 3: Expertise and Need to Evaluate as Predictors of
Evaluative Integration (2000 NES)
Evaluative Integration
Model 1 Model 2
Predictor b SE b B SE b
Age .0004 (.0003) .0004 (.0003)
Income .003 (.002) .003 (.002)
Race .02 (.01) .02 (.01)
Gender .00003 (.01) -.001 (.01)
College degree -.003 (.01) -.003 (.01)
Political expertise .07*** (.02) .07*** (.02)
Need to evaluate .09*** (.02) .09*** (.02)
Need for cognition .01 (.01) .01 (.01)
Expertise × Need to
Evaluate -- -- .15* (.07)
Expertise × Need for
Cognition -- -- ­.01 (.04)
Need to Evaluate × Need
for Cognition -- -- .01 (.06)
College Degree × Need
to Evaluate -- -- ­.001 (.03)
Constant .57*** (.02) .57*** (.02)
F (degrees of freedom) 10.32 (8, 1,561)*** 7.23 (12, 1,557)***
R
2
.051 .055
N 1,570 1,570
NOTE: Entries are unstandardized ordinary least squares (OLS) regression
coefficients and HC3 robust standard errors. 2000 NES = 2000
National Election Study.
p < .10. *p < .05. **p < .01. ***p < .001.
The findings presented in Table 2 satisfy Baron and
Kenny's first criterion with regard to both forms of mediated
moderation: the key interaction between expertise
and the need to evaluate was significantly associated with
each extremity index, net of the effects of all constituent
lower-order terms. However, the tests needed to satisfy
modified versions of Baron and Kenny's second and
third criteria were slightly different for each form of mediated
moderation. As noted earlier, the first mediated
moderation hypothesis suggests that the interactive effect
of schema development and the need to evaluate on
extremity is due to the interactive effect of schema development
and the need to evaluate on evaluative integration.
In this case, satisfaction of Baron and Kenny's second
criterion would require that the interaction
between expertise and the need to evaluate significantly
predict scores on the proposed mediator, that is,
evaluative integration. This is demonstrated by the findings
summarized in Table 3, which indicate that each of
these terms had a significant net relationship with integration
(b = .15, p < .05). In turn, the second mediatedmoderation
hypothesis suggests that the interactive effect
of schema development and the need to evaluate on
extremity is actually transmitted by the more proximal
interactive effect of schema development and evaluative
integration. In this case, satisfaction of Baron and
Kenny's second criterion would require that the interaction
between expertise and the need to evaluate significantly
predict scores on the product term for the interaction
between expertise and evaluative integration
(Lance, 1988). Testing for the presence of this effect required
a number of steps. First, the product term for this
interaction was regressed on its two constituent main effects,
that is, expertise and evaluative integration. The
residuals for this regression were then computed to generate
a product term representing only the unique variance
associated with interaction itself (see Lance, 1988).
This corrected term was then regressed on the same set
of predictors used in the analysis summarized in Table 3.
This analysis indicated that the Expertise × Need to Evaluate
significantly predicted the corrected product term,
b = .10, p < .001, satisfying the second criterion.
Finally, a three-step hierarchical regression procedure
was used to test Baron and Kenny's third criterion
for each form of mediated moderation. This procedure
was carried out separately for each extremity measure.
These analyses are summarized in Table 4. For each extremity
measure, the first step of this procedure used
only the demographics, the college-degree indicator, expertise,
the need to evaluate, the need for cognition, the
Expertise × Need to Evaluate interaction, the Expertise ×
Need for Cognition interaction, the Need to Evaluate ×
Need for Cognition interaction, and the College Degree
× Need to Evaluate interaction as predictors (Model 1).
This simply shows the total effect of the Expertise × Need
to Evaluate interaction.
To examine Baron and Kenny's final criterion with regard
to the first form of mediated moderation, Model 2
added evaluative integration to the predictors included
in Model 1. In this case, significant mediation requires
that the statistical effect of the interaction between expertise
and the need to evaluate be significantly reduced
once this is done. For ideological extremity, Model 2 indicated
a highly significant effect of integration (b = .35,
p < .001). In turn, the actual mediated effect--corresponding
to the reduction in the magnitude of the interaction
between expertise and the need to evaluate--is
given by the product of this coefficient and the coefficient
for the interactive effect of expertise and the need
to evaluate on evaluative integration (b = .15; see Table
3), which yields an estimate of .050. A Sobel test indicated
that this mediated effect was significant, z = 2.01,
p < .05. For issue extremity, Model 2 also indicated a significant
effect of integration (b = .11, p < .001). Multiplied
by the coefficient for the interactive effect of expertise
and the need to evaluate on evaluative integration
(b = .15, again), this yielded a modest but significant
mediated effect of .02, z = 2.00, p < .05.
To examine Baron and Kenny's final criterion with regard
to the second form of mediated moderation, Model
3 added the mediating interaction between expertise
and evaluative integration to the predictors included in
Model 2. In this case, significant mediation requires that
the statistical effect of the interaction between expertise
and the need to evaluate be significantly reduced once
this is done. For ideological extremity, Model 3 indicated
a highly significant interaction between integration and
expertise (b = .85, p < .001). The actual mediated effect is
given by the product of this coefficient and the coefficient
for the interactive effect of expertise and the need
to evaluate on the corrected Expertise × Evaluative Integration
product term (i.e., b = .10), yielding an estimate
of .085. A Sobel test indicated that this mediated effect
was significant, z = 3.00, p < .01. For issue extremity,
Model 3 also indicated a significant effect of the interaction
between integration and expertise (b = .23, p < .001).
Multiplied by the coefficient for the interactive effect of
expertise and the need to evaluate on the corrected Expertise
× Evaluative Integration term (b = .10, again),
this indicated a mediated effect of .023, z = 2.65, p = .01.9
Thus, for both dependent measures, both "mediated
moderation" hypotheses were supported. Consistent
with the pattern suggested by these individual effects,
further analyses indicated that the total indirect effect
(MacKinnon, 2000) of the interaction between expertise
and the need to evaluate via integration and the Expertise
× Integration interaction was significant for both
ideological extremity (IE = .135, z = 3.46, p < .001) and
Federico / THE NEED TO EVALUATE AND ATTITUDE EXTREMITY 1289
issue extremity (IE = .043, z = 3.58, p < .001). Taken together,
these results suggest that the tendency for
schema development in a given domain to be more
strongly associated with attitude extremity among those
with a high need to evaluate may indeed be mediated by
increased integration in one's evaluative responses to
objects in that domain.
DISCUSSION
The question of when the possession of welldeveloped
schemas should be associated with extreme
attitudes has occupied researchers for some time (see
Judd & Brauer, 1995; Tesser et al., 1995, for reviews). As
noted earlier, a number of well-known studies have indicated
that this relationship may be stronger when
perceivers' responses to various features of an attitude
domain are evaluatively integrated or lacking in ambivalence
(e.g., Judd & Lusk, 1984; Millar & Tesser, 1986; see
also Tesser et al., 1995). However, relatively little attention
has been devoted to the issue of when schema development
and integration should co-occur, producing
a positive relationship between the possession of
information-rich schemas and attitude extremity. Drawing
on recent studies of a relatively new construct--the
need to evaluate (Jarvis & Petty, 1996; see also Bizer et al.,
2003; Federico, 2003)--the perspective developed here
suggests that individual differences in evaluative motivation
may shed some light on this issue. More precisely, it
was hypothesized (a) that the need to evaluate, that is, a
strong motivation to engage in evaluative thought,
would be associated with both attitude extremity and increased
integration; (b) that the possession of schemas
containing information about a greater number of
domain features would be more strongly associated with
extreme attitudes toward objects in that domain among
individuals with a high need to evaluate; (c) that the possession
of well-developed schemas would be more
strongly associated with higher levels of evaluative integration
among those with a high need to evaluate; and
(d) that this increased integration would mediate the
moderating effects of the need to evaluate.
Data from a nationally representative sample of adults
provided a clear pattern of support for these hypotheses
in the context of people's political attitudes. At the most
basic level, the data indicated that the need to evaluate
was associated with increased extremity across two
aspects of respondents' attitudes toward objects in the
political domain (i.e., ideology and issues) and with
higher levels of integration in responses to oppositely
valenced features of the political domain (i.e., conservative
and liberal political figures). However, of more
interest, they also indicated that the possession of welldeveloped
political schemas--measured in terms of
1290 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
TABLE 4: Mediated Moderation Analyses (2000 NES)
Ideological Extremity
Model 1 Model 2 Model 3
Predictor b SE b b SE b b SE b
Expertise × Need to Evaluate .27* (.12) .22 (.12) .13 (.12)
Integration -- -- .35*** (.05) .34*** (.05)
Expertise × Integration -- -- -- -- .85*** (.15)
F (degrees of freedom) 7.48 (12, 1,427)*** 10.98 (13, 1,414)*** 13.60 (14, 1,413)***
R 2 .057 .099 .120
N 1,440 1,428 1,428
Issue Extremity
Model 1 Model 2 Model 3
Predictor b SE b b SE b b SE b
Expertise × Need to Evaluate .14** (.05) .11* (.05) .09 (.05)
Integration -- -- .11*** (.02) .11*** (.02)
Expertise × Integration -- -- -- -- .23*** (.06)
F (degrees of freedom) 5.93 (12, 1,579)*** 9.00 (13, 1,556)*** 9.51 (14, 1,555)***
R 2 .046 .076 .086
N 1,592 1,570 1,570
NOTE: Entries are unstandardized ordinary least squares (OLS) regression coefficients and HC3 robust standard errors. All models also include
age, income, race, gender, collegedegree,expertise, need to evaluate, need for cognition, the Expertise × Need for Cognition interaction, the Need
to Evaluate × Need for Cognition interaction, and the College Degree × Need to Evaluate interaction; estimates for these terms and the regression
constant are not shown. For both dependent measures, both Step 2 and Step 3 produced a significant increase in R2 (all ps < .001). 2000 NES = 2000
National Election Study.
p < .10. *p < .05. **p < .01. ***p < .001.
political expertise--was more strongly associated with
both attitude extremity and integration among those
with a high need to evaluate. Putting the last piece of the
puzzle in place, a final set of analyses provided evidence
for both of the "mediated moderation" hypotheses outlined
earlier, indicating that the critical interaction
between expertise and the need to evaluate was indeed
mediated by integration.
Implications
So, what do these results tell us? In addition to providing
evidence for a main effect of the need to evaluate on
attitude extremity, they suggest that the need to evaluate
may help determine when attitude-relevant schemas develop
features essential for the emergence of extremity.
Previous work on the relationship between schema development
and extremity has focused largely on how this
relationship might be moderated by structural characteristics
of the schemas themselves, such as evaluative integration
(Judd & Brauer, 1995; Judd & Lusk, 1984;
Tesser et al., 1995). However, this body of research has
had little to say about why attitude-relevant schemas may
develop these structural characteristics in the first place.
The perspective developed here attempts to address this
question by situating the critical interface between
knowledge structures and attitude extremity in the context
of a broader motivational model anchored by the
need to evaluate. In particular, it builds on previous work
by suggesting that one of the structural moderators highlighted
by earlier work--integration--may in fact mediate
the moderating effects of a more distal moderator of
the relationship between schema development and extremity,
namely, the motivation to engage in evaluative
thought. According to this perspective, the motivational
impact of the need to evaluate plays the lead role, serving
as the ultimate moderator of the relationship between
schema development and attitude extremity. Nevertheless,
as the notion of mediated moderation suggests, integration
plays an essential role in the unfolding of this
effect. More precisely, the need to evaluate may moderate
the relationship between schema development and
extremity by determining when schema development is
associated with the integrated, univalent response pattern
necessary for the emergence of extreme attitudes
toward objects in a domain. In other words, the need to
evaluate may motivate perceivers to use the conceptual
information embedded in well-developed schemas to
evaluatively integrate their responses, giving their
representations of a given attitude domain the structure
necessary for the expression of extreme attitudes.
More broadly, however, the findings reported here
provide evidence for the importance of general motivational
constructs in our understanding of attitude extremity.
As noted in the introduction, work on motivated
social cognition increasingly suggests that informational
factors--including the number of domain features represented
by relevant schemas and structural characteristics
of these schemas, such as the degree to which they
are evaluatively integrated--may interact with various
motivational factors to determine how preexisting
knowledge is used to construct beliefs and preferences
(Jarvis & Petty, 1996; Kruglanski, 1996; Lavine, 2002).
Although this general perspective has been useful in
making sense of the cognitive processes behind stereotyping
(e.g., Kruglanski, 1996), impression formation
(Fiske & Neuberg, 1990), and the development of interattitudinal
structure (Federico, 2003), its relevance to
the social psychology of extreme attitudes has not been
considered in any great detail. Instead, most analyses
have focused on informational bases of extremity,
namely, the possession of attitude-relevant knowledge
structures, the degree to which these structures are integratively
organized, and so on. Moving away from this
somewhat one-dimensional mode of analysis, the perspective
developed here strongly suggests that attitude
extremity is not merely a matter of having certain kinds
of attitude-relevant cognitive structures. Rather, it is also
a matter of having the motivation to use preexisting
knowledge in a broadly evaluative fashion.
However, in addition to highlighting the general relevance
of motivational factors in the expression of
extreme attitudes, these findings also provide further
evidence for the particular motivational significance of
the need to evaluate. As noted earlier, previous research
on the need to evaluate suggests that it is reliably associated
with evaluatively directed thought and opinion formation,
as well as a stronger tendency to engage in behaviors
indicative of well-formed preferences (Bizer
et al., 2003; Jarvis & Petty, 1996). The present study extends
these findings in a number of ways. At the simplest
level, it provides further evidence for an association between
the need to evaluate and the expression of definite,
well-formed opinions. Although previous studies
have suggested thatindividuals with a high need to evaluate
are likely to have opinions about a greater number of
objects (Jarvis & Petty, 1996), none of these analyses have
provided evidence of a relationship between the need
to evaluate and the extremity of these opinions, even
though the conceptual definition of the construct would
appear to strongly predict a relationship of this sort. The
results reported here fill this gap, suggesting that the
need to evaluate is reliably associated with extremity,
even after other predictors are considered.
At a more complex level, though, the results presented
here suggest that the need to evaluate also may
condition the evaluatively oriented use of information
made available by the possession of certain knowledge
structures. More precisely, these findings suggest that
Federico / THE NEED TO EVALUATE AND ATTITUDE EXTREMITY 1291
the possession of well-developed schemas about various
attitude domains may not be associated with extreme
evaluations of objects in those domains unless perceivers
are particularly motivated to form opinions about the
social world. In this regard, these results suggest that
the need to evaluate--similar to other cognitivemotivational
variables--may have some of its most important
effects in interaction with other knowledgeconstitutive
variables (Cacioppo et al., 1996; Federico,
2003; Kruglanski, 1996). Previous work has provided
clear evidence for a "main effect" of the need to evaluate
with respect to attitude formation and expression in a
number of domains (Jarvis & Petty, 1996). In contrast, researchers
have paid little or no attention to how the need
to evaluate may interact with other factors to affect the
construction of preferences and perceptions. The findings
presented here provide a step in this direction and
point toward a fruitful avenue for future work.
Conclusions and Caveats
In sum, the findings reported here contribute both to
our understanding of the conditions under which attituderelevant
schema development is associated with attitude
extremity and to our understanding of the need-to-evaluate
construct. Nevertheless, certain limitations of the
data are worth considering. Most obviously, the
correlational nature of the data used in these analyses
makes it difficult to draw causal conclusions. Although
these findings are clearly consistent with the hypothesis
that the need to evaluate--in conjunction with schema
development--has a causal impact on the expression of
extreme attitudes, we cannot firmly draw this conclusion
in the absence of experimental control. Moreover, the
fixed format of the NES data precluded the consideration
of an issue central to many analyses of attitude extremity,
namely, the effects of attitude-relevant thought.
As noted in the Introduction, a number of studies have
shown that mere thought can lead to extremity, especially
if it is guided by well-developed, evaluatively integrated
schemas (Tesser et al., 1995). Although the effects
of thought could not be considered in the NES data,
future studies may want to examine its effects in the context
of the model developed here. For example, if it is
true that the need to evaluate determines when attituderelevant
schemas are likely to acquire the high level of
integration necessary for the emergence of extreme attitudes,
then the tendency for schema-directed thought to
be associated with increased extremity may be moderated
by the need to evaluate. Future work should consider
this possibility.
Finally, some readers may be concerned about the
present study's focus on extremity in a single domain,
that is, politics. Despite this choice of contexts, there is
no reason to believe the model of attitude extremity developed
here is any less applicable to other domains; it
simply suggests that the need to evaluate may motivate
perceivers to structure the information embedded in
well-developed schemas in ways that facilitate expression
of extreme attitudes. In the present study, this information
came from political schemas, but other types of attitude-relevant
information should do so in the same fashion.
Put another way, regardless of the context, the key
argument is that attitude extremity depends on both the
possession of relevant knowledge structures and the motivation
to use the information embedded in them in an
evaluative fashion. Although the model developed here
may be of particular interest to political psychologists--
who have long been interested in the psychology of
extreme attitudes (see Sidanius, 1988; Sidanius & Lau,
1989)--it should be of equal relevance to researchers interested
in other domains. Hopefully, future research
will conceptually replicate the present study's findings in
these other domains.
NOTES
1. Throughout this article, the terms schema development and welldeveloped
schemas will be used interchangeably to refer to the same
thing, namely, the possession of attitude-relevant schemas containing
information about a larger number of object features or features of a
given attitude domain.
2. Consistent with these earlier findings, an ordinary least squares
(OLS) regression analysis in which knowledge, need to evaluate, need
for cognition, interest in the interview, interest in politics, political participation,
and a dummy variable indicating completion of a college
degree were used to predict attitude constraint indicated that knowledge
was by far the strongest predictor of the latter ( = .13, p < .0001).
The next most powerful predictor, completion of a college degree, had
only half the predictive power ( = .06, p < .05). The two other "political"
predictors, interest in politics and political participation, did not
have significant net relationships with constraint (both ps > .15).
3. Further information on the validation and psychometric properties
of the National Election Study (NES) Need to Evaluate and Need
for Cognition scales can be found in Bizer et al. (2003). Confirmatory
factor analyses using the data examined here have shown that the need
to evaluate and need for cognition items do in fact measure two distinct
dimensions (Bizer et al., 2003). In these analyses, covariance
among the full set of items was best explained by an oblique two-factor
structure.
4. This index simply reverses the Thompson, Zanna, and Griffin
(1995) ambivalence index by taking its negative. Its use can be illustrate
by imagining two hypothetical respondents who have rated a liberal
figure and a conservative figure on a scale ranging from 0 (very negative)
to 100 (very positive). Say one of them has provided integrated
responses as defined here, giving the conservative figure at 100 and the
liberal figure a 0, whereas the other has provided poorly integrated
responses, giving both figures a 100. Using the integration formula
given above, the first respondent would get a higher raw score (i.e., 50)
than the second `poorly integrated' respondent (i.e., ­100). Note that
these raw scores were normalized to run from 0 to 1 for ease of analysis,
as noted below.
5. This is one of the reasons why evaluative responses to issues were
not used to construct the integration measure: the NES issue items
force respondents to choosebetweenoptions with oppositeideological
valence rather than rating them separately. As noted above, the other is
that this would contribute to an overlap in method variance with the
dependent measures. In principle, however, there is no reason why
1292 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN
integration has to be indexed in terms of evaluations of political figures:
this is simply what is most convenient given the available
measures.
6. Although dichotomous indices of this sort are commonly used to
assess (and control for) education in studies of political attitude structure
(see Federico & Sidanius, 2002; Judd, Krosnick, & Milburn, 1981;
Judd & Milburn, 1980; Sniderman, Brody, & Tetlock, 1991), the analyses
reported below also were repeated using the full seven-category
NES education variable in place of the college-degree indicator. These
analyses producedresults that werevirtually identical to those reported
below. The need to evaluate had a significant main-effect relationship
with ideological extremity (b = .19, p < .001), issue extremity (b = .03, p <
.05), and integration (b = .09, p < .001) and the critical Expertise× Need
to Evaluate interaction was significant vis--vis ideological extremity (b
= .29, p < .05), issue extremity (b = .12, p < .01), and integration (b = .14,
p < .06). Moreover, the indirect effects corresponding to each form of
"mediated moderation" remained significant for ideological extremity
(IE = .049, p = .06; IE = .085, p < .01) and issue extremity (IE = .015, p =
.06; IE = .023, p < .01). The total indirect effect of the interaction
between expertise and the need to evaluate via both forms of mediated
moderation also was significant for both dependent variables (total IE
= .134, p < .01, for ideological extremity; total IE = .038, p < .01, for issue
extremity). Finally, none of the effects involving the full education
measure were significant in any of the models (all p s> .10).
7. This is only one of the ways in which ideological extremity might
be operationalized (see Sidanius, 1988). Therefore, the analyses were
repeated using a variety of extremity indices, including (a) deviation
from the midpoint of .5 on a composite measure of ideology (based on
the 7-point scale and the difference between their thermometer ratings
of conservatives and liberals), (b) deviation from the sample mean
on the ideology composite, (c) deviation from the sample median on
the ideology composite, (d) a dummy variable indicating whether they
classified themselvesas conservatives or liberalson the 7-point ideology
scale rather than classifying themselves as moderates, and (e) the
extent to which the difference between their thermometer ratings of
conservatives and liberals deviated from zero. Although some of these
analyses required forms of estimation other than OLS (e.g., logit regression
for the dummy indicator), all of them produced virtually the
same pattern of results reported below.
8. Note that the ideological extremity index has only four categories.
Given this relative lack of continuity, OLS may not produce efficient
estimates (Borooah, 2002). As such, the Table 2 analyses using
this measure as a dependent variable were repeated using ordered
probit instead of OLS. This analysis produced identical results. As
such, only the original OLS estimates for this dependent measure are
reported to maintain consistency across models.
9. To break this result down even further, simple slopes for the relationship
between expertise and each extremity index at low and high
levels of integration also were computed (based on the terms included
in Model 3 for each dependent variable). Consistent with earlier work
on the role of evaluative integration (Judd & Lusk, 1984; Tesser,
Martin, & Mendolia, 1995), these analyses indicated that expertise was
associated with increased ideological extremity among those high in
integration (b = .21, p < .001) and decreased ideological extremity
among those low in integration (b = ­.08, p < .05). Similarly, expertise
was associated with increased issue extremity among those high in integration
(b = .05, p < .001) and decreased issue extremity among those
low in integration (b = ­.02, p < .10).
REFERENCES
Abelson, R. P. (1995). Attitude extremity. In R. E. Petty & J. A.
Krosnick (Eds.), Attitude strength: Antecedents and consequences
(pp. 25-42). Mahwah, NJ: Lawrence Erlbaum.
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting
interactions. Newbury Park, CA: Sage.
Baron, R. M., & Kenny, D. G. (1986). The moderator-mediator variable
distinction in social-psychological research: Conceptual, strategic,
and statistical considerations. Journal of Personality and Social
Psychology, 51, 1173-1182.
Bizer, G. Y., Krosnick, J. A., Holbrook, A. L., Petty, R. E., Rucker, D. E.,
& Wheeler, S. C. (2003, May). The impact of personality on political
beliefs, attitudes, and behavior: Need for cognition and need to evaluate.
Paper presented at the annual meeting of the American Psychological
Society, Atlanta, GA.
Borooah, V. K. (2002). Logit and probit: Ordered and multinomial models.
Thousand Oaks, CA: Sage.
Cacioppo, J. T., Petty, R. E., Feinstein, J. A., & Jarvis, W. B. G. (1996).
Dispositional differences in cognitive motivation: The life and
times of individuals varying in need for cognition. Psychological
Bulletin, 119, 197-253.
Converse, P. (1964). The nature of belief systems in mass publics. In
D. Apter (Ed.), Ideology and discontent. New York: Free Press.
Craig, S. C., Kane, J. G., & Martinez, M. D. (2002). Sometimes you feel
like a nut, sometimes you don't: Citizens' ambivalence about abortion.
Political Psychology, 23, 285-303.
Delli Carpini, M. X., & Keeter, S. (1996). What Americans know about
politics and why it matters. New Haven, CT: Yale University Press.
Federico, C. M. (2003, April). When does schematicity imply structure? The
interactive effects of expertise and the need to evaluate on interattitudinal
structure. Paper presented at the annual meeting of the Midwest
Political Science Association, Chicago.
Federico, C. M., & Sidanius, J. (2002). Racism, ideology, and affirmative
action, revisited: The antecedents and consequences of "principled
objections" to affirmative action. Journal of Personality and
Social Psychology, 82, 488-502.
Fiske, S. T., Lau, R. R., Smith, R. A. (1990). On the varieties and utilities
of political expertise. Social Cognition, 8, 31-48.
Fiske, S. T., & Neuberg, S. L. (1990). A continuum model of impression
formation from category-based to individuating processes:
Influences of information and motivation on attention and interpretation.
In M. P. Zanna (Ed.), Advances in experimental social psychology
(Vol. 23, pp. 1-74). San Diego, CA: Academic Press.
Goren, P. (2001). Core principles and policy reasoning in mass
publics: A test of two theories. British Journal of Political Science, 31,
159-177.
Hamill, R., Lodge, M., & Blake, F. (1985). The breadth, depth, and
utility of partisan, class, and ideological schemas. American Journal
of Political Science, 29, 850-870.
Jarvis, W. B., & Petty, R. E. (1996). The need to evaluate. Journal of Personality
and Social Psychology, 70, 172-194.
Judd, C. M., & Brauer, M. (1995). Repetition and evaluative extremity.
In R. E. Petty & J. A. Krosnick (Eds.), Attitude strength: Antecedents
and consequences (pp. 43-72). Mahwah, NJ: Lawrence Erlbaum.
Judd, C. M., & Krosnick, J. A. (1989). The structural bases of consistency
among political attitudes: Effects of expertise and attitude
importance. In A. R. Pratkanis, S. J. Breckler, & A. G. Greenwald
(Eds.), Attitude structure and function (pp. 99-128). Hillsdale, NJ:
Lawrence Erlbaum.
Judd, C. M., Krosnick, J. A., & Milburn, M. A. (1981). Political involvement
and attitude structure in the general public. American Sociological
Review, 46, 660-669.
Judd, C. M., & Lusk, C. M. (1984). Knowledge structure and evaluative
judgments: Effect of structural variables on judgmental
extremity. Journal of Personality and Social Psychology, 46, 1193-1207.
Judd, C. M., & Milburn, M. A. (1980). The structure of attitude systems
in the general public: Comparisons of a structural equation
model. American Sociological Review, 45, 627-643.
Kruglanski, A. W. (1996). Motivated social cognition: Principles of
the interface. In E. T. Higgins & A. W. Kruglanski (Eds.), Social
psychology: A handbook of basic principles (pp. 493-522). New York:
Guilford.
Lance, C. E. (1988). Residual centering, exploratory and confirmatory
moderator analysis, and decomposition of effects in path
models containing interaction effects. Applied Psychological Measurement,
12, 163-175.
Lavine, H. (2002). Online versus memory-based process models of
political evaluation. In K. R. Monroe (Ed.), Political psychology
(pp. 225-247). Mahwah, NJ: Lawrence Erlbaum.
Lavine, H., Borgida, E., & Sullivan, J. L. (2000). On the relationship
between attitude involvement and attitude accessibility: Toward a
Federico / THE NEED TO EVALUATE AND ATTITUDE EXTREMITY 1293
cognitive-motivational model of political information processing.
Political Psychology, 21, 81-106.
Liberman, A., & Chaiken, S. (1991). Value conflict and thought
induced attitude change. Journal of Experimental Social Psychology,
27, 203-216.
Linville, P. W. (1982). The complexity-extremity effect and age-based
stereotyping. Journal of Personality and Social Psychology, 42, 193-
211.
Linville, P. W., & Jones, E. E. (1980). Polarized appraisals of outgroup
members. Journal of Personality and Social Psychology, 38, 689-703.
Lodge, M., & Hamill, R. (1986). A partisan schema for political information
processing. American Political Science Review, 80, 505-519.
Long, J. S., & Ervin, L. H. (2000). Using heteroskedasticity consistent
standard errors in the linear regression model. American Statistician,
54, 217-234.
Lusk, C. M., & Judd, C. M. (1988). Political expertise and the structural
mediators of candidate evaluations. Journal of Experimental
Social Psychology, 24, 105-126.
MacKinnon, D. P. (2000). Contrasts in multiple mediator models. In
J. Rose, L. Chassin, C. C. Presson, & S. J. Sherman (Eds.), Multivariate
applications in substance use research: New methods for new questions
(pp. 141-160). Mahwah, NJ: Lawrence Erlbaum.
Millar, M. G., & Tesser, A. (1986). Thought-induced attitude change:
The effects of schema structure and commitment. Journal of Personality
and Social Psychology, 51, 259-269.
Myers, D. G., & Lamm, H. (1976). The group polarization phenomenon.
Psychological Bulletin, 83, 602-627.
Priester, J. R., & Petty, R. E. (1996). The gradual threshold model of
ambivalence: Relating the positive and negative bases of attitudes
to subjective ambivalence. Journal of Personality and Social Psychology,
71, 431-449.
Sidanius, J. (1988). Political sophistication and political deviance: A
structural equation examination of context theory. Journal of Personality
and Social Psychology, 55, 37-51.
Sidanius, J., & Lau, R. R. (1989). Political sophistication and political
deviance: A matter of context. Political Psychology, 10, 85-109.
Sniderman, P. M., Brody, R. A., & Tetlock, P. E. (1991). Reasoning and
choice. New York: Cambridge University Press.
Tesser, A., Martin, L., & Mendolia, M. (1995). The impact of thought
on attitude extremity and attitude-behavior consistency. In R. E.
Petty & J. A. Krosnick (Eds.), Attitude strength: Antecedents and consequences
(pp. 73-92). Mahwah, NJ: Lawrence Erlbaum.
Thompson, M. M., Zanna, M. P., Griffin, D. W. (1995). Let's not be
indifferent about (attitudinal) ambivalence. In R. E. Petty & J. A.
Krosnick (Eds.), Attitude strength: Antecedents and consequences
(pp. 361-386). Mahwah, NJ: Lawrence Erlbaum.
Tormala, Z. L., & Petty, R. E.. (2001). On-line versus memory-based
processing: The role of "need to evaluate" in person perception.
Personality and Social Psychology Bulletin, 27, 1599-1613.
Wegener, D. T., & Fabrigar, L. R. (2000). Analysis and design for nonexperimental
data: Addressing causal and non-causal hypotheses.
In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in
social and personality psychology (pp. 412-450). New York: Cambridge
University Press.
Zaller, J. (1990). Political awareness, elite opinion leadership, and the
mass survey response. Social Cognition, 8, 125-153.
Zaller, J. (1992). The nature and origins of mass opinion. New York: Cambridge
University Press.
Received August 29, 2003
Revision accepted December 23, 2003
1294 PERSONALITY AND SOCIAL PSYCHOLOGY BULLETIN