Journal of Personality and Social Psychology Copyright 1982 by the American Psychological Association, Inc.

1982, Vol. 42, No. 4, 619-630 0022-3514/82/4204-0619$00.75

Affective and Semantic Components in

Political Person Perception
Robert P. Abelson, Donald R. Kinder, and Mark D. Peters
Yale University
Susan T. Fiske
Carnegie-Mellon University

According to the new conventional wisdom, social psychology has become cap-
tured by cognition; it should pay greater attention to affect. In that spirit, our
article explores comparisons between conventional semantic judgments and af-
fective reports. In two national surveys respondents were invited to ascribe per-
sonality traits to prominent national politicians as well as to report the feelings
that the politicians elicited. We find first that summary scores of good feelings
and bad feelings are nearly independent of each other, much more so than are
good and bad trait judgments. Affective registrations, in short, seem less se-
mantically filtered, less subject to consistency pressures. We also find that sum-
mary scores of affect strongly predict political preference. This effect is inde-
pendent of and more powerful than that for personality judgments. Thus,
affective registrations are not at all redundant with semantic judgments. Overall,
these results should encourage the quickening interest in human emotion.

Suddenly it is fashionable to write about available regarding the properties of affec-

emotion. According to a newly emerging tive response. Social psychology has made
conventional wisdom, social psychology has extensive use of semantic differential scales
become too exclusively cognitive, and it is (rating concepts on scales of good-bad,
time to reexamine the role of affect (Fiske, strong-weak, etc.) but has rarely employed
1981; Higgins, Kuiper, & Olson, 1981; questions of the general form, To what ex-
Tomkins, 1981; Zajonc, 1980a). Accompa- tent does this object or concept make you
nying this criticism is a proliferation of the- feel angry? . . . proud? . . . sad? (etc.).
ories of emotion (e.g., see Clark & Fiske, in Our purpose is to offer several sustained
press, for a sampling; DeRivera, 1977; Lev- comparisons between affective reports and
enthal, 1980; Mandler, 1975; Roseman, semantic judgments.
Note 1). Zajonc (1980b), meanwhile, has Social psychology research has tradition-
advanced his controversial thesis that two ally focused on various aspects of the judg-
independent systems of evaluation exist: a ments people make regarding other individ-
fast, crude, and, perhaps, predominant af- uals and groups. For example, work on
fective system and a slower, more detailed implicit personality theory (Rosenberg &
cognitive one. Sedlak, 1972) explores structural relation-
Although these theoretical developments ships among trait judgments. Information
are exciting, very little evidence is presently integration research (Anderson, 1974) ana-
lyzes the evaluative implications of trait
The research reported here was supported by a sub- combinations. Research on stereotyping has
contract to D. R. Kinder, S. T. Fiske, and R. P. Abelson investigated both the content (Katz & Braly,
from the University of Michigan Center for Political
Studies. The interpretations herein are of course solely 1933) and functions of categorization (All-
our own. We would like to thank Margaret Clark and port, 1954; Taylor, 1980). Finally, consis-
Richard Lau for their comments on an earlier draft of tency theories of attitudes have relied heavily
this manuscript. Donald Kinder is now at the University on the relations among cognitions (Abelson
of Michigan.
Requests for reprints should be sent to R. P. Abelson, et al., 1968). All of these cases reflect Za-
Box 11 A, Yale Station, New Haven, Connecticut jonc's (1980b) point that historically, social
06520. psychologists have viewed evaluation and
619
620 ABELSON, KINDER, PETERS, AND FISKE

preference as resulting from semantic judg- ment of citizens' appraisal of presidential

ments. candidates fit our interests precisely. In re-
The view explored here is that affective sponse to CPS's invitation, we developed
reports may differ qualitatively from the both affect and trait inventories (and other
semantic judgments as routinely employed measures as well; Kinder, Abelson, & Fiske.,
in social-psychological research. In the first Note 2), which have now been included in
place affect may be less "filtered. The struc- several national surveys.
ture of affective reports need not follow the This article reports the results from two
dictates of semantic relationships. In the sec- national surveys conducted by CPS early in
ond place, affective reports differ from se- the 1980 presidential election season. In
mantic judgments in their function, partic- each case we will compare the traits that the
ularly as motivators. Tomkins (1962) notes public ascribes to political candidates on the
that affect is the "motor" for behavior. Be- one hand and the feelings candidates elicit
cause affective reports reflect motivation on the other. Are reports of affective reac-
more directly than semantic judgments, they tions to candidates in fact structured differ-
may surpass semantic judgments in the pre- ently from trait ascriptions? And do affec-
diction of behavior. tive reports, more than trait judgments,
To conduct research on the structure and predict citizens' summary evaluations of
function of affects, we wanted to be certain candidates and their preferences between
of a domain in which real affective reactions candidates?
would occur over time, semantic judgments
could be made, and actual behavioral con- Method
sequences follow. The laboratory is ill suited
for these purposes. Hence, one seeks a real Sample and Procedure
world domain in which many people react Our analysis draws on two recent national surveys
affectively and behaviorally to the same tar- undertaken by CPS. In both, adult respondents were
selected according to a probability sampling procedure.
get persons; essentially one needs instances In terms of basic demographic and political character-
of mass person perception. Where could one istics, these samples represent reasonably well the na-
find a large sample of individuals who judge tional adult population, showing roughly proper distri-
the same set of target persons toward whom butions of age, sex, race, education, and political party
they might have a variety of feelings and to preference.
In Study 1, designed as a small-scale feasibility test
whom they could make trait attributions? for a variety of new. measures including our affect in-
What target persons can one use at large on ventory, respondents were interviewed in person during
a college campus? Possibly one or two very the first 2 weeks of March, 1979 (-« = 280) and rein-
famous professors and a handful of movie, terviewed approximately 3 weeks later (« = 236). In the
TV, or rocks stars would do; but it is not course of 2'A hours of interviewing, respondents were
asked about political parties, institutions, and topical
easy to find a set who are consensually in- issues. The interview also included a battery of measures
teresting and familiar enough for both feel- developed to assess trait-based impressions of four well-
ing and trait reports. In any case, it seems known political leaders (Carter, Ford, Kennedy, and
desirable to go beyond the usual college stu- Reagan), as well as the patterns of affects that these
leaders elicited.
dent sample. The reader will have antici- Study 2 was undertaken in early February 1980, as
pated us: Politicians are ideally suited to the presidential campaign was kicked off. Study 2's sam-
these demands. The most visible politicians ple was larger (N = 1,008), fewer questions were asked,
are familiar even to the inattentive public; and the average interview lasted about 1 hour. Short-
they project varying personalities, thereby ened trait and affect batteries were applied to six polit-
ical figures who at the time were in the running for
inviting semantic judgments, and as it turns presidential nomination: Carter, Kennedy, Reagan,
out, they evoke widespread reports of various Connally, Baker, and Bush.
feelings.
Just the opportunity we needed came Measures
about as the result of an invitation from the
Affect checklist. The interviewer introduced a sec-
Center for Political Studies (CPS) at the tion of the interview by saying, "Now I want to ask you
University of Michigan. CPS's interest in about (candidate). Think about your feelings when I
developing new approaches to the assess- mention (candidate). Now, has (candidate)—because
 AFFECTIVE AND SEMANTIC COMPONENTS 621

of the kind of person he is or because of something he trait set, three essentially behavioral items (all of a pos-
has done— ever made you feel: Angry? . . . Happy? itive nature) were included in the trait inventory. Thus
. . . Hopeful?.. . ." and so on through a list of affects. the final "trait" set included nine items (six positive and
For each affect term, the interviewer recorded a simple three negative).4
Yes or No response. Overall candidate evaluation and preference. One
In Study 1 half the sample was asked 12 affect ques- of our purposes was to compare how well affect and trait
tions (5 positive, 7 negative) about Carter and Ford, inventories predicted overall evaluation of each candi-
and the other half, the same questions about Kennedy date. In both studies summary evaluation was measured
and Reagan. In Study 2 the whole sample was asked in terms of a 100-point "thermometer scale," a device
seven affect questions (three positive, four negative) long used by CPS. Responses range from 0 degrees
about Carter, Kennedy, Reagan, Connally, Baker, and (extremely unfavorable) to 100 degrees (extremely fa-
Bush, except when the respondent had claimed not to vorable); the respondent is instructed to give a number
"know anything at all about" the particular candidate for each candidate corresponding to the relative degree
on a previous screening question. of favorability or unfavorability felt toward him. 5 It is
The affect checklist underwent several stages of se- noteworthy that such ratings are tantamount to vote
lection and refinement. An original set of affect terms intention. In CPS presidential election studies, citizens'
was selected predominantly to represent the basic pos- evaluations of the major party candidates assessed by
sibilities in Roseman's (Note 1) structural theory of means, of thermometer scales turn out to be essentially
emotion. His theory was particularly germane because equivalent to their votes (cf. Brody & Page, 1973); that
it is a cognitively informed analysis of affects as a func- is, the thermometer correlates extremely highly with
tion of goal outcomes. Given our interest in the real- actual vote, and the thermometer has exactly the same
tionship between semantic and affective responses, we patterns of relationship to other variables as does the
began with Roseman's terms. After a preliminary fea- vote.
sibility test on a small sample in New Haven, the list Finally, in Study 1 only, a measure of relative can-
of affects was pruned down to the 12 most informative didate preference was also available. Respondents were
terms, which then appeared in Study 1. Some that were presented with five would-be-candidates: Carter, Ford,
eliminated, such as "makes me feel guilty," were simply Kennedy, Reagan, and Brown, and were asked to rank
not common reactions to politicians. Seven of the 12 them in order of preference. In the analysis to come, we
terms then survived to Study 2.' (The complete lists are will make full use of both preference and overall eval-
shown in Table 1.) uation.
A final comment is in order regarding the use of self- Given our interest in comparing the predictive validity
report measures of affect. One might suggest that re- of trait attributions and affective reports, it is worth
spondents might prefer not to admit some affective re- noting that the summary evaluations are not themselves
sponses, although given the outspoken character of affective responses. The affect items ask for reports of
American politics, this seems doubtful. In any case the specific emotional events (e.g., "Has . . . Carter ever
measures are subjective reports about affective reactions made you feel. . . frustrated? . . . hopeful?") In con-
not measures of affect as facial expressions or as au- trast, the thermometer and preference ratings both ask
tonomic responses. It is arguable whether the phenom-
enology of affect is a less "direct" measure than are 1
There was a slight wording change for two of the
physiological measures (Fiske, 1981). Certainly self-re- affect questions in Study 2. "Afraid" became "afraid
ported affect is the most feasible alternative in a survey of him," and "sympathetic" became "sympathetic Jo-
context. ward him." These changes were designed to enforce our
Trait and behavioral descriptions. Respondents intention that the candicate be regarded as the object
were also asked to characterize the traits of the can- of the respondent's feelings not the mobilizer of feelings
didates. In Study 1, 16 traits (8 positive, 8 negative)2 directed toward other targets. Indications are that this
were used, and the whole sample rated all four candi- intention was clear to almost all respondents.
dates on all traits. The exact working of the question, 2
These 16 traits were courageous, warm, honest,
say for the trait honest, as applied to Carter, was, "Does smart, humble, knowledgeable, open-minded, inspiring,
the word honest describe Carter extremely well, very immoral, too political, selfish, weak, reckless, unstable,
well, pretty well, or not very well at all?"3 In Study 2, power hungry, prejudiced.
a subset of but six of these traits (three positive, three 3
In later analysis of the trait measure, these respec-
negative) was rated for each of six candidates by ev- tive responses were scored in the usual Likert scale fa-
eryone in the sample who was at all acquainted with the shion: 3, 2, 1, 0.
respective candidate. Originally, the traits had been se- 4
These nine items were moral, dishonest, weak,
lected from a much longer list winnowed after the New knowledgeable, power hungry, inspiring, solve our eco-
Haven feasibility study. nomic problems, provide strong leadership, develop good
Respondents were also asked to characterize candi- relations with other countries.
dates by miscellaneous behavioral descriptions such as, 'The exact wording was, "rate the person using the
"would appoint good advisors," "would develop good feeling thermometer. You may use any number from
relations with other countries," and so on. Results from 0 to 100 for rating. Ratings between SO degress and 100
Study 1 (Kinder et al., 1980) suggested that these be- degrees mean that you feel favorable and warm toward
havioral descriptions behaved very much like the trait the person. Ratings between 0 degrees and 50 degrees
items. In Study 2, therefore, they were eliminated as a mean that you don't feel favorable toward the person
separate category, but to compensate for a diminished and that you don't care too much for the person."
622 ABELSON, KINDER, PETERS, AND FISKE

for a summary judgment about the candidate (see Foot- jected to oblique rotation. Two factors were
note 5 for wording of thermometer). Such evaluations sufficient for each candidate except that a
are traditionally analyzed as a function of party, issues,
and candidate personality (e.g., Page & Jones, 1979). small third factor (omitted in Table 1) oc-
curred for Ford in Study 1.
The pattern in Table 1 is strikingly con-
Results
sistent. For every candidate in both studies,
Factor Structure of Affects one of the factors (labeled 1) is a Negative
Affect factor, the other (2) a Positive Affect
The first question to answer in analyzing factor. Differences in the factor coefficients
affective responses is how they cluster with for the different candidates are far less sa-
one another. Table 1 presents the results of lient than the massive overall pattern of
separate factor analyses for each of the pres- agreement.
idential front-runners in Study 1 and again There are two noteworthy properties of
in Study 2, with Ford absent, but Connally, this recurrent factor pattern. The first is the
Baker, and Bush added. Factors were ex- absence of factorial articulation among good
tracted from matrices of correlations over feelings or among bad feelings. One might
respondents of affect mentions and were sub- suppose that such articulation would occur,

Table 1
Affect Factor Structures
Affect Candidate

Study 1 (N = 110)
Carter Kennedy Reagan Ford

1 2 1 2 1 2 1 2

Afraid .49 -.26 .61 -.23 .76 -.13 .37 .07

Angry .67 -.31 .62 -.09 .75 -.04 .75 -.06
Digusted .64 -.42 .63 -.43 .75 -.19 .67 -.15
Disliking .57 -.30 .72 -.32 .71 -.34 .57 -.36
Frustrated .61 -.09 .60 -.12 .66 -.04 .39 .09
Sad .40 .01 .53 .10 .55 -.04 .61 .01
Uneasy .60 -.15 .62 -.30 .70 -.10 .51 -.11
Happy -.17 .55 -.15 .71 -.05 .72 .08 .65
Hopeful -.17 .71 -.24 .74 -.07 .70 .06 .62
Liking -.34 .63 -.37 .81 -.30 .70 .19 .65
Proud -.17 .63 -.12 .73 -.06 .80 .13 .75
Sympathetic -.14 .52 -.14 .68 -.12 .54 .09 .56
Factor r -,.30 -,.06 -,.24 -. 16

Study 2 (N = 915)'
Carter Kennedy Reagan Connally Baker Bush

1 2 1 2 1 2 1 2 1 2 1 2

Afraid .32 -.11 .63 -.33 .65 -.22 .71 -.18 .56 ,00 .55 -.10
Angry .67 -.19 .71 -.36 .63 -.33 .68 -.19 .70 -.04 .72 -.19
Disgusted .68 -.28 .71 -.43 .69 -.28 .73 -.18 .62 -.02 .66 -.16
Uneasy .47 -.19 .68 -.37 .70 -.27 .71 -.26 .71 -.03 .54 -.07
Hopeful -.20 .70 -.44 .77 -.28 .75 -.24 .78 -.06 .70 -.23 .68
Proud -.25 .59 -.41 .74 -.24 .72 -.17 .78 -.07 .77 -.15 .75
Sympathetic -.16 .41 -.28 .56 -.18 .47 -.18 .50 .03 .50 -.06 .64
Factor r -. 36 -. 54 -,.37 -,.29 -,.04 -,.20

Note. Factor 1 is the Negative Affect factor, Factor 2 is the Positive Affect factor.
" The N given is for Carter. Other Ns were smaller (See Table 2).
 AFFECTIVE AND SEMANTIC COMPONENTS 623

partly on theoretical grounds specifying Table 2

structural differences among some-valenced Correlations Between Positive and Negative
affects (Roseman, Note 1) and partly be- Indicators: Traits and Affects*
cause of the precedent established by factor
Candidate
analyses of trait ratings. With trait ratings and study » Trait r Affect r Difference
it is possible to discern different content def-
initions among the positive (or negative) Carter
traits of politicians: in particular, a Com- 1 105 -.44 -.34 .10
petence factor and an Integrity factor 2 975 -.53 -.23 .30
Kennedy
(Kinder, Fiske, & Wagner, Note 3). Related 1 118 -.53 -.30 .23
trait content differentiations occur in ordi- 2 937 -.60 -.42 .18
nary person perception (e.g., Rosenberg & Reagan
Sedlak, 1972). But the affective responses 1 111 -.53 -.16 .37
2 859 -.47 -.27 .20
in the present surveys do not ramify beyond Ford: 1 106 -.35 -.10 .25
the simple positive versus negative distinc- Connally: 2 631 -.54 -.23 .31
tion. Baker: 2 425 -.38 -.05 .33
A second and perhaps even more arresting Bush: 2 369 -.35 -.16 .19
feature of this pattern is that the Positive M -.50" -.26" .24
and Negative Affect factors are relatively
independent. The 10 factor intercorrelations * The positive and negative trait measures are equally
given in Table 1 have a median of -.26 and weighted additive scores based on replies to the Trait
Inventory. The positive and negative affect measures are
range from -.04 to -.54. (The latter value equally weighted scores based on replies to the Affect
attaches to Kennedy in Study 2). For the Checklist.
b
most part, then, feeling good things about The Ns for different candidates vary because calcu-
a political leader doe^ not imply the absence lations were limited to those individuals claiming to
know anything at all about the candidate.
of bad feelings. Citizen Smith, in whom c
Means for the first two columns are calculated as the
President Carter evoked anger and sadness, retransform of the weighted (by N) mean of the z-trans-
may very well have also expressed hope and formed correlation coefficients.
sympathy toward the president. Somehow
there is a weakness of consistency pressures It occurred to us that although positive
to bring these two polarities into natural op- and negative affects might be weakly linked
position with one another. in the overall sample, there might be strong
Although positive and negative affects (inverse) links within particular subsamples
elicited by candidates are typically only very in the two studies. We looked at several. The
weakly correlated, the comparative question clearest results emerged when each sample
remains: Are good and bad traits more was partitioned by party identification.6
sharply correlated with each other than are Three sections of each sample were defined:
good and bad affects? To explore this ques- strong partisans (from both parties), weak
tion, each respondent (in both studies) was partisans (from both parties and indepen-
assigned a positive affect score for each can- dents with leanings), and "true" indepen-
didate, as the simple count of the number dents. Table 3 shows what happened when
of positive affects mentioned; a negative af- positive and negative affects respectively
fect score was assigned correspondingly. were correlated within each subsample.
(This procedure was chosen in preference to The results are relatively easy to sum-
weighted factor scores because of its sim- marize: The negative correlation between
plicity and uniformity across candidates.)
The trait scores were calculated in a com- 6
The standard CPS measure of party identification
parable way. Table 2 shows the correlations was administered in both Study 1 and Study 2. It con-
among traits and among affects. For every sists of a lead-in question ("Do you think of yourself
candidate in both studies, the two affect as Democrat, a Republican, or an Independent?"), and
scores have a smaller (inverse) correlation a follow-up. The latter sorts both Republicans and De-
than do the trait scores. The average affect mocrats by whether or not they feel strongly, and In-
dependents by whether they lean toward the Democrats,
correlation is .24 smaller in magnitude than the Republicans, or neither. This procedure results in
the average trait correlation. a 7-point spectrum of party orientations,
624 ABELSON, KINDER, PETERS, AND FISKE

Table 3 Predicting Overall Evaluation and

Correlations Between Positive and Negative Preference
Indicators Within Levels of Partisanship
Dp the feelings candidates elicit affect
Candidate Whole Strong Weak Indepen- how favorably they are evaluated? Not sur-
and study sample partisans partisans dents prisingly, the answer is yes. What is sur-
Affects prising, perhaps, is how powerfully affect
patterns predict preference and evaluation.
Carter Table 4 presents data for the multiple regres-
1 -.34 -.69 -.18 -.04
2 -.23 -.35 -.19 -.07 sions of thermometer ratings on the two pre-
Kennedy dictors—positive and negative affect scores—
1 -.30 -.26 -.32 -.21 for both studies. Simple and multiple cor-
2 -.42 -.52 -.39 -.41 relations are given, along with unstandar-
Reagan
1 -.16 -.26 -.14 .01 dized and standardized regression weights.
2 -.27 -.34 -.25 -.22 The meaning of each unstandardized weight
Ford: 1 -.10 -.40 -.01 .06 (b) is the average number of points gained
Connally: 2 -.23 -.30 -.18 -.27 (lost) on the thermometer scale for each pos-
Baker: 2 -.05 -.08 -.08 .33 itive (negative) affect mention. In Study 2,
Bush: 2 -.16 -.26 -.11 -.14
for example, Carter was rated almost 10
M -.26 -.36 -.19 -.10 points better per good feeling mentioned and
Traits
8!/a points worse for each bad feeling men-
tioned.8 The median absolute value of these
Carter unstandardized regression weights in Table
1 -.44 -.57 -.42 -.60 4 is 6.79.
2 -.53 -.57 -.53 -.39
Kennedy The multiple correlations for the several
1 -.53 -.60 -.57 -.36 candidates are typically large, and in all
2 -.60 -.63 -.58 -.62 cases they significantly exceed (beyond the
Reagan .01 level) the largest of the two single-vari-
1 -.53 -.59 -.46 -.66
2 -.47 -.53 -.47 -.38 able correlations. Thus, for Reagan in Study
Ford: 1 -.35 -.29 -.34 -.58 1, the positive affect score correlates .51 with
Connally: 2 -.54 -.56 -.48 -.77 thermometer, the negative score correlates
Baker: 2 -.38 -.26 -.47 -.20 —.44, and the multiple correlation is .63. In
Bush: 2 -.35 -.48 -.31 -.03 Study 2 the corresponding figures are .56,
M -.50 -.52 -.47 -.49 —.47, and .65. The clear gain from having
two predictors accrues, of course, from the
positive and negative affect scores is typi- very low correlations between the two pre-
cally (8 times out of 10) more pronounced dictors.
among strong partisans than among weak
7
partisans. The comparison of strong parti- On affects and traits separately, an analysis of vari-
sans and independents shows an even clearer ance of the (3 Subsamples X 10 Target Cases) table of
pattern, for here the negative correlation z-transformed correlation coefficients was run, treating
target cases as a random factor. For affects, subsamples
between positive & negative affects was in- produced an F(2, 18) of 9.75, significant at the .01
variably more pronounced in the former level. For traits, the subsample F(2, 18) was .35, clearly
group (10 cases out of 10). In contrast, for not significant. A three-way analysis of Mea-
trait intercorrelations, there is hardly any sures X Subsamples X Target Cases yielded a Mea-
sures X Subsamples effect at the .10 significance level,
consistent gradient across intensity of par- with an F(2, 18) of 2.87.
tisanship.7 Putting the matter another way, 8
This analysis (as does any regression analysis) raises
the tendency for the two affect scores to be but- does not resolve a thorny causal issue. To some
less redundant than the two trait scores is extent the existence of positive (negative) feelings no
maximized among independents and weak doubt causes citizens to evaluate a candidate positively
partisans. The implications of these results (negatively). However, it is also possible that positive
(negative) evaluation spurs the evocation of positive
are examined in the final section of the ar- (negative) feelings. We postpone to another day at-
ticle. tempts to disentangle this simultaneity.
 AFFECTIVE AND SEMANTIC COMPONENTS 625

A general pattern evident throughout Ta- Table 5

ble 4 is the superiority of positive over neg- Predicting Preference Rank from Positive and
ative affect as a predictor of thermometer Negative Affect (Study 1)
ratings. Whether one looks at rs, bs, or betas, R
Candidate Simple r b Beta
the positive affect coefficient typically ex-
ceeds the negative affect coefficient. This Carter (N = 91)
result, although rich in potential implica- Positive affect .51 .30 .38 .63
tions, should be treated with caution. It no Negative affect -.53 -.26 -.40
doubt depends on which positive and nega- Kennedy (N = 102)
tive affects happen to be included in the re- Positive affect .72 .50 .61 .81
spective cluster. Negative affect -.56 -.28 -.38
Table 5 gives the comparable statistics Reagan (N = 96)
with preference score as the dependent vari- Positive affect .66 .51 .60 .76
able instead of thermometer rating for Study Negative affect -.47 -.29 -.38
1 only. Again the multiple correlations are Ford (N = 90)
large and significant, and there are notable Positive affect .48 .35 .46
-.23
.53
predictive gains in using both affect scores Negative affect -.27 -.19

Table 4 rather than one alone. And, once more, the

Predicting Thermometer Ratings From Positive regression coefficients are generally higher
and Negative Affect for positive than for negative affect.
Candidate Are Affect Measures Redundant With
and study N Simple r b Beta R
Semantic Information?
Carter: 1
Positive affect .55 6.80 .44 Evaluations of (say) Edward Kennedy
.63 partly reflect people's abiding loyalties to
Negative affect -.47 -4.14 -.33
Carter: 2 party, appraisals of his character, and judg-
Positive affect „,_ .52 9.74 .42 .67\ ments regarding how well he would perform
Negative affect -.53 -8.59 -.43 if elected president. Our final question is
Kennedy: 1 whether we can improve our prediction of
Positive affect .64 7.34 .51
Negative affect -.59 -5.64 -.44
.77 evaluation of Kennedy (and other political
Kennedy: 2
notables) by incorporating into our equa-
Positive affect „.„ .64 12.18 .45 tions measures of affective response.
Negative affect .76 In Study 1 we took the multiple regres-
-.64 -9.33 -.45
Reagan: 1
sions of thermometer (or preference) score
Positive affect .51 6.78 .46 on good and bad trait score, good and bad
Negative affect -.44 -4.90
.63 behavior rating scores, and good and bad
-.36
Reagan: 2 affect scores. Party identification was added
Positive affect „ .q .56 10.60 .47 as a seventh predictor variable to check the
Negative affect -.47 -6.77 -.34
.65
extent to which affect and trait scores might
Ford: 1 be simple reflections of partisan orientation
Positive affect ]f . fi .53 6.28 .50 .59 rather than substantive predictors in their
Negative affect -.31 -3.57 -.26
own right. In Study 2 a similar analysis was
Connally: 2 carried out with five predictors instead of
Positive affect ,-n .45 8.76 .36
.58 seven (recall that traits and behaviors were
Negative affect b2U -.46 -6.58 -.38
combined rather than separate measures).
Baker: 2 Table 6 gives the relevant details.
Positive affect ..» .42 6.76 .41 In Study 1, candidate by candidate, one
.50
Negative affect ' -.29 -5.70 -.28
or both affect variables made significant con-
Bush: 2 tributions in every case, and the signs of the
Positive affect .„ .49 7.96 .45 .56 regression coefficients were always in the
Negative affect -.34 -7.69 -.27
right direction, that is, positive for positive
626 ABELSON, KINDER, PETERS, AND FISKE

affects and negative for negative affects. For teresting issue; virtually all the betas were
Carter and Ford, positive affect score was significant.) These results indicate that the
the significant one; for Reagan, negative af- two polarities of affective reaction contribute
fect; for Kennedy, both. (Even stronger re- different and at least as consequential infor-
sults were obtained with preference as the mation about respondents' appraisals of can-
dependent variable. There, both affect scores didates as the pair of trait judgment scores.
made significant contributions for all cases
except Ford, where the positive affect regres- Discussion
sion weight was insignificant.) By contrast, In summarizing our major results on af-
regression weights for trait and behavior fective response to presidental candidates in
scores as predictors of the thermometer were two separate national surveys, we have found,
not always of appropriate sign, and even first, that good feelings and bad feelings are
when the signs were appropriate, the coef- nearly independent of each other, much
ficients were often not of significant mag- more so than are good traits and bad traits;
nitude. and second, that affect scores are highly pre-
In Study 2 the standardized regression dictive of overall candidate evaluations, add-
coefficients for affect scores were uniformly ing significant variance explanation over and
in the appropriate directions and of fairly above that due to trait scores and party iden-
consistent absolute size, ranging from .16 to tification.
.27. By contrast, the positive trait coeffi- Perhaps our most psychologically provoc-
cients were somewhat higher (.27 to .41), ative result is the near independence of neg-
but the negative trait coefficients were lower ative and positive affects—particularly when
(.03 to .13). Coefficients for party identifi- set against the findings for trait judgments,
cation were of lesser absolute magnitude where positive and negative trait attributions
(Mdn = .08). (Because of the large sample are substantially inversely related. How are
sizes, statistical significance was not an in- we to account for this difference? We see

Table 6
Predicting Thermometer Ratings from Positive and Negative Affects Along with Party
Identification and Trait (and Behavior) Scores (Beta weights for multiple regressions)
Predictor variable Candidate
Study 1
Carter Kennedy Reagan Ford
Positive affects .23* .34** .07 .32**
Negative affects -.13 -.35** -.20* -.13
Positive traits .18 .13 .35** -.08
Negative traits .13 -.16 -.01 -.08
Positive behaviors -.14 .10 .24* .43**
Negative behaviors -.02 .12 -.08 -.05
Party identification8 .18* .08 -.03 .08
Study 2
Carter Kennedy Reagan Connally Baker Bush
Positive affects .25 .23 .25 .23 .16 .21
Negative affects -.22 -.21 -.26 -.27 -.17 -.24
Positive traits .38 .40 .38 .27 .41 .31
Negative traits -.06 -.08 -.03" -.10 -.03" -.13
Party identification" .11 .05 -.07 -.04b -.21 -.09
" Party identification was coded from Strong Republican as -3 through Strong Democrat as +3.
b
These three coefficients are the only ones not significant at least at p < .05. The large Ns in Study 2 render
significance testing quite uninformative.
* p < .05. **p < .01.
 AFFECTIVE AND SEMANTIC COMPONENTS 627

three possible lines of explanation: one the possible ranges for the affect scores. Thus
methodological and boring, the other two (as the reader may already have noted) it
theoretical and provocative. The first iden- might be argued that the trait scores were
tifies potential mischief in the manner of more reliable than the affect scores, which,
measurement of affective and semantic re- in turn, enabled their more substantial (in-
sponses. The second argues for the greater verse) intercorrelations.
episodic character of affective as against se- Although this argument has surface plau-
mantic response. And the third elaborates sibility, a closer look at the data robs it of
on the possibility that the structural differ- its force. First of all we have seen (Table 6)
ences between affective response and seman- that affect scores have even greater predic-
tic judgments may be traced to differences tive validity than trait scores—an outcome
in psychological perspective associated with difficult to achieve were affect scores suffer-
each. ing from a serious deficiency in reliability
Methodological explanations, if applica- relative to trait scores. Second, low reliabil-
ble, are potentially damaging to the claim ity based on relatively more restricted range
that affective responses are in any funda- cannot readily explain why affect score in-
mental way different from semantic judg- tercorrelations are higher among strong par-
ments. The positive and negative affect tisans (Table 3), One would have to posit
scores were created by summing the number a reliability differential between trait and
of single mentions of particular experiences. affect scores for independents and weak par-
The possible range of positive affect scores tisans, which shrinks for strong partisans.
was 0 to 5 in Study 1, 0 to 3 in Study 2; for Although such an effect is conceivable, ar-
negative affect scores, the possible range was guing for it goes beyond the mere numerical
0 to 7 in Study 1 and 0 to 4 in Study 2. constraints imposed by the way trait and
These score ranges are small, especially in affect scores were calculated. A more sub-
contrast to the possible ranges of the positive stantive account of the differential result for
and negative trait scores. For those measures strong partisans is given below.
each item was scored for the degree to which The most direct way to address a potential
the trait described the candidate on a 4-point reliability artifact is to calculate reliability
scale (coded 0 to 3), and then sums were coefficients for both trait scores and affect
taken over all items. This Likert-type score scores. The present data set does not contain
had a possible range of 0 to 24 for both good test-retest information. The available alter-
and bad traits in Study 1 (with 8 items per native, although perhaps of arguable co-
scale). In Study 2 the possible range was Q gency here, is internal consistency reliability.
to 18 for positive trait score and 0 to 9 for Such a computation was performed on Study
negative trait score. Altogether the possible 2 data. Table 7 gives reliabilities computed
ranges of the trait scores considerably exceed as coefficient alpha (cf. Cronbach, 1970) for

Table 7
Reliabilities of Negative and Positive Affect and Trait Scores (Study 2)
Affects Traits

Candidate Negative Positive Mean" Negative Positive Mean'

Carter .616" .574 .595 .563 .828 .683

Reagan .763 .673 .717 .574 .855 .701
Kennedy .778 .723 .750 .682 .884 .776
Connally .799 .718 .757 .579 .862 -.706
Baker .744 .689 .716 .538 .848 .675
Bush .688 .727 .707 .502 .843 .651

* The geometric mean: the square root of the product of the two reliabilities. This quantity is in common use for
the "correction for attenuation" due to imperfect reliability.
b
Coefficients are Cronbach alphas.
628 ABELSON, KINDER, PETERS, AND FISKE

good and bad affect and trait scores for each "has (candidate) . . . ever made you feel
of the six candidates. (angry, etc.)?" In contrast, the trait items
The negative affect scores were more re- asked, "does the word (honest, etc.) describe
liable than the negative trait scores for all (candidate). . . ?" The former wording so-
six candidates. On the other hand the posi- licits past episodes, the latter wording pres-
tive affect scores were less reliable than the ent judgments. We deliberately introduced
positive trait scores for all six candidates. the unusual phrasing for the affect questions
These opposing tendencies play off about to reflect an apparent natural property of
equally against each other. Their net effect affects—they wax and wane in the course of
can be assessed from the geometric mean of particular experiences rather than being nec-
the reliabilities for positive and negative essarily present on demand at the moment
scores; Table 7 shows that for four of the six of questioning. If asked, "Are you angry at
candidates, the net affect score reliability is Jimmy Carter?" it might be reasonable to
actually greater than the net trait score re- say, "Not today." On the other hand, it
liability. What this means is that score re- seems a bit weird to give this same reply if
liability differences cannot account for the asked, "Do you think Jimmy Carter is
greater (inverse) correlation of good and bad weak?"
trait scores than good and bad affect scores. The episodic nature of affective experience
A second methodological argument goes can explain the frequent co-occurrence of
as follows: If some respondents respond good and bad affect mentions. Different ex-
"No" to all the affect items, then both the periences can each leave appropriately dif-
positive and negative affect scores will be ferent traces without any necessity to "rec-
zero, and this will work in the direction of oncile" them. A person can fairly say,
a positive intercorrelation between good and "Jimmy Carter made me feel hopeful after
bad affects. The number of such respondents Camp David, and made me feel angry when
might be high when people find the partic- he called for draft registration." The accre-
ular candidate unfamiliar—though we have tion in memory of individual episodes of
some protection against an "unfamiliarity varying affective significance seems a per-
effect" in that people who said that they did fectly natural process not only with political
not know anything at all about a candidate figures but with interpersonal experiences as
were not asked the affect questions and were well. It is a common observation that in
not included in any tables. married couples and other intimate relation-
In any case the correlation coefficient as ships mixed feelings develop. Systematic in-
a measure of association between good and terview data from married couples (Kelley,
bad affects does not indicate separately the Note 4) not surprisingly conform to this
relative occurrence of these (none, none) common observation.
cases as against (none, some), (some, none) Nevertheless, one might wonder what
and (some, some). However, inspection of would happen in response to altered phras-
these separation tallies reveals no big pileup ings of trait and affect questions. One could
of (none, none) cases. In Study 2 the greatest ask respondents how they presently feel,
such accumulation is for Howard Baker, for rather than how they ever felt. One could
whom 29.6% of the subsample claiming to also present trait items in episodic form, e.g.,
know something of him give neither a pos- "Has the word honest ever described Rich-
itive nor a negative affect mention. For more ard Nixon?," or perhaps, "Did you ever
familiar candidates, this pileup is much think Richard Nixon was honest?" It would
weaker. Jimmy Carter, for example, elicits be interesting if there were a tendency to
only 2.7% in the (none, none) cell. Thus, we update such retrospective judgments, much
are reassured that the relative independence as in the phenomenon of failure to remember
of positive and negative affect scores is not prior opinions (cf. Bern & McConnell, 1970).
due to an unfamiliarity or apathy effect. Unfortunately, there is nothing in our pres-
The second explanation of our results, the ent data that can bear on this speculation.
episodic hypothesis, is based on the differ- An appropriate experiment has not yet been
ence in the way the affect and trait questions carried out, but we plan to conduct research
are worded. The affect items were phrased, along these lines.
 AFFECTIVE AND SEMANTIC COMPONENTS 629

Thus we do v not know to what extent our partisans and independents. One function of
low intercorrelations between positive and strong partisanship is to organize feelings
negative affect scores might be the result of toward, political objects, thus depriving af-
the episodic nature of the affect measure fective responses of some of their naivete.
rather than (or in addition to) some other Thus, a strong partisan (whether Republican
property naturally attaching to affects. or Democrat) with a dislike for an incum-
An interesting alternative arises from the bent president or a well-publicized challen-
attributional perspective of affective reports ger might well burn with a purer flame, sup-
as opposed to trait judgments. Affective re- pressing any acknowledgment of possible
ports concern the internal state of the re- positive feelings. Respondents less emotion-
sponder, whereas trait judgments are fo- ally linked to particular candidates, on the
cused on an external stimulus. A well-known other hand, could incline more toward con-
analysis of self-report versus reports of oth- fessed ambivalence.
ers (Jones & Nisbett, 1971) can be applied Why would trait judgments not also be
here. Information about self is much more subject to partisan organization? Table 3
complex and realistic than information about shows no trend over partisanship in the size
others, that is, less subject to stereotyped or of the (inverse) intercorrelation between pos-
unidimensional structuring. Thus, it would itive and negative trait scores. A mean cor-
not be surprising if affective responses (which relation of -.49 is manifest even among in-
concern the self) were more "contradictory" dependents. Our intuition is that this is
than trait judgments (which concern the because trait judgments in politics are sub-
candidate).9 ject to consistency pressures borrowed from
Here again, there is nothing in our data the domain of ordinary interpersonal judg-
that would either support or subvert such an ments. People in general are reluctant to rate
account. It is conceivable, however, that an someone both open-minded and selfish or
experiment could be designed to disentangle both knowledgeable and prejudiced, and so
the self-other distinction from the affect- forth. These pressures apply independently
trait distinction. One could, for example, ask of political partisanship. With affects, how-
for feelings evoked by candidates in other ever, we hypothesize a general lack of con-
people: "Has your mother ever felt angry at sistency pressure except under conditions of
Jimmy Carter?" "Does your best friend feel emotional commitment such as is provided
afraid of Ronald Reagan?" by partisanship.
Pending the arrival of new data pertinent The preceding discussion is of course spec-
to various explanations of our correlational ulative. We do not have sufficient evidence
result, we content ourselves with noting that to favor one explanation over another, but
both the "episodic" and "perspective" ex- we have demonstrated that affective re-
planations claim that affective responses are sponses are structured differently from trait
less subject to consistency pressures than
are trait judgments. One might say that trait 9
We are indebted to Joseph Jaspars of Oxford for
judgments are "semantically filtered" to a suggesting this line of explanation of our results.
10
greater extent than are affective responses, Having thus contrasted affective reports from trait
that is, considered more heavily in relation judgments, we hasten to add the caveat that we should
not expect good (bad) feelings to be totally uncorrelated
to an overall conception the respondent is with good (bad) trait judgments. Not even Zajonc's two-
attempting to convey. By contrast, affective -process model asserts that the results of the two pro-
responses are a more "naive," direct reflec- cesses are necessarily unconnected. In our survey con-
tion of experience, in the spirit of Zajonc's text many factors dispose somewhat toward like-signed
(1980b) proposal of an affect registration responding across trait and affect inventories, including
that old rating bugaboo, the halo effect. Over the 10
system remote from conscious verbal pro- candidate cases of the two studies, good feelings cor-
10
cessing. relate on average .55 with good traits and bad feelings,
This line of argument is consistent with .42 with bad traits. Our claims about affects versus traits
the differential results (Table 3) according refer to the differential properties of the two types of
to level of partisanship. Strong partisans, it measure over and above what they may have in common.
The multiple regression analysis of Table 6 makes clear
will be recalled, did not show the low affect that their differential properties are predictively con-
score intercorrelations characteristic of weak sequential.
630 ABELSON, KINDER, PETERS, AND FISKE

judgments. Affective responses add predic- salience of premanipulation attitudes. Journal of Per-
tive validity in the political domain above sonality and Social Psychology, 1970, 14, 23-31.
Bradburn, N. M., & Caplovitz, D. Reports on happi-
and beyond that available from standard se- ness. Chicago: Aldine, 1965.
mantic judgments of candidates, and they Brody, R. A., & Page, B. I. Indifference, alienation, and
have the very interesting property of near rational decisions. Public Choice, 1973, /5, 1-17.
independence of positive and negative scores Clark, M. S., & Fiske, S. T. (Eds.), Affect and cognition:
The 17th annual Carnegie Symposium. Hillsdale,
(except among strong partisans). In this re- N.J.: Erlbaum, in press.
gard it is interesting that a comparable in- Cronbach, L. J. Essentials of psychological testing. (3rd
dependence of positive and negative affects ed.), New York: Harper & Row, 1970.
was also uncovered in surveys of life satis- De Rivera, J. A structural theory of the emotions. Psy-
faction (Bradburn & Caplovitz, 1965). Those chological issues. Monograph #40. N.Y.: Interna-
tional Universities Press, 1977.
authors found that reports of not having felt Fiske, S. T. Social cognition and affect. In J. Harvey
negative feelings (restless, lonely, etc.) within (Ed.), Cognition, social behavior, and the environ-
the previous week and of having felt positive ment. Hillsdale, N.J.: Erlbaum, 1981.
feelings (proud, excited, etc.) were each pre- Higgins, E. T., Kuiper, N. A., & Olson, J. M. Social
cognition: A need to get personal. In E. T. Higgins,
dictive of general reported happiness. How- C. P. Herman, & M. P. Zanna (Eds.), Social cog-
ever, positive and negative feeling reports nition: The Ontario Symposium. Hillsdale, N.J.: Erl-
had different personal correlates and corre- baum, 1981.
lated a mere -.07 with each other. Jones, E. E., & Nisbett, R. E. The actor and the ob-
If political figures, perceived at a distance, server: Divergent perceptions of the causes of behav-
ior. In E. E. Jones et al. (Eds.), Attribution: Per-
evoke such frequent and theoretically pro- ceiving the causes of behavior. Morristown, N.J.:
vocative affective reports, close perceptions General Learning Press, 1971.
of ordinary people ought to evoke even more Katz, D., & Braly, K. W. Racial stereotypes of 100
interesting affective reports. Exploration of college students. Journal of Abnormal and Social
Psychology, 1933, 28, 280-290.
the structure and functions of such inter- Kinder, D. R., Peters, M. D., Abelson, R. P., & Fiske,
personal affective reactions is empirically S. T. Presidential prototypes. Political Behavior,
tractable and promising. 1980, 2, 315-338.
Leventhal, H. Toward a comprehensive theory of emo-
• tion. In L. Berkowtiz (Ed.), Advances in experimental
Reference Notes social psychology (Vol. 13). New York: Academic
1. Roseman, I. Cognitive aspects of emotion and emo- Press, 1980.
tional behavior. Paper presented at the meeting of Mandler, G. Mind and emotion, New York: Wiley,
the American Psychological Association, New York, 1975.
September 1979. Page, B. I., & Jones, C. C. Reciprocal effects of policy
2. Kinder, D. R., Abelson, R. P., & Fiske, S. T. De- preferences, party loyalty, and the vote. American
velopmental research on candidate instrumentation. Political Science Review, 1979, 73, 1071-1089.
(Report available from the Center for Political Stud- Rosenberg, S., & Sedlak, A. Structural representations
ies, University of Michigan, Ann Arbor, MI, 48106.) of implicit personality theory. In L. Berkowitz (Ed.),
3. Kinder, D. R., Fiske, S. T., & Wagner, R. G. Pres- Advances in experimental social psychology (Vol. 6).
idents in the public mind. Unpublished manuscript, New York: Academic Press, 1972.
Yale University, 1980. Taylor, S. E. A categorization approach to stereotyping.
4. Kelley, H. H. Personal communication, September, In D. L. Hamilton (Ed.), Cognitive processes in ste-
1979. reotyping and intergroup behavior. Hillsdale, N.J.:
Erlbaum, 1980.
References Tomkins, S. S. The quest for primary motives: Biog-
raphy and autobiography of an idea. Journal of Per-
Abelson, R. P., Aronson, E., McGuire, W. J., Newcomb, sonality and Social Psychology, 1981, 41, 306-329.
T. M., Rosenberg, M. J., & Tannenbaum, P. H. Tomkins, S. S. Affect, imagery, and consciousness (Vol.
(Eds.). Theories of cognitive consistency: A source- 1). New York: Springer, 1962.
book. Chicago: Rand McNally, 1968. Zajonc, R. B, Cognition and social cognition: A histor-
Allport, G. W. The nature of prejudice. Reading, Mass: ical perspective. In L. Festinger (Ed.), Four decades
Addison-Wesley, 1954. of social psychology. Oxford University Press, 1980.
Anderson, N. H. Information integration: A brief sur- (a)
vey. In D. H. Krantz, R. C. Atkinson, R. D. Luce, Zajonc, R. B. Feeling and thinking: Preferences need
& P. Suppes (Eds.), Contemporary developments in no inferences. American Psychologist, 1980, 35, 151-
mathematical psychology (Vol. 2). San Francisco: 175. (b)
Freeman, 1974.
Bern, D. J., & McConnell, H. K. Testing the self-per- Received December 15, 1980
ception explanation of dissonance phenomena: On the Revision received August 26, 1981 •