You are on page 1of 12

Psychological Bulletin

1972, Vol. 77, No. 5, 361-372

ON THE AMBIVALENCE-INDIFFERENCE PROBLEM IN


ATTITUDE THEORY AND MEASUREMENT:
A SUGGESTED MODIFICATION OF THE SEMANTIC
DIFFERENTIAL TECHNIQUE 1
KALMAN J. KAPLAN 2
Wayne Stale University

This article explores the alternate meanings of attitudinal neutrality in the context
of the bipolarity-reciprocal antagonism issue. Specifically, it proposes a modification
of the semantic differential technique wherein the "liking" and "disliking" com-
ponents of attitude can be separately measured. A geometrical model is developed
in which three nondirectional attitude variables ("total affect," "ambivalence,"
and "polarization") are distinguished from the usual attitude variable. Reliability
and validity data are presented, and an application of the model is discussed.

In his 1935 review of the general area of must be considered in addition to evaluations
attitude theory and research, Gordon Allport for a workable definition of attitude (cf.
concluded that most investigators basically Fishbein, 1963, 1965, 1967; Rosenberg, 1956,
agreed that "attitude is a learned predisposition 1960). The preponderance of current research,
to respond to an object in a consistently favor- however, has emerged from the second camp
able or unfavorable manner [p. 818J." Further- (e.g., Komorita & Bass, 1967; Smith, 1961;
more, he suggested that this bipolarity in the Wiggins & Fishbein, 1969). Additionally, a
direction of an attitude (i.e., the favorable number of studies have specifically been inter-
versus unfavorable) was viewed as its most ested in the implications of the bipolarity-
distinctive feature. Thus, he felt that attitude reciprocal antagonism issue for the interpre-
had been conceptualized as a simple uni- tation of attitudinal neutrality (e.g., Bass &
dimensional concept—the evaluative (or affec- Rosen, 1969; Green & Goldfried, 1965;
tive) dimension. Herzberg, Mausner, & Snyderman, 1959). It
A number of attitude theorists and re- is the attempt to conceptually and methodo-
searchers, including Allport himself, have been logically distinguish ambivalence from indif-
unhappy with this unidimensional view and ference that represents the central concern of
attacked it as oversimplified. Basically, the the present study.
critics fall into two major camps: (a) multi-
component attitude theorists, who feel the Some General Requirements for the Measurement
evaluative dimension or component is not a of Ambivalence
sufficient definition of attitude; and (b) uni-
component attitude theorists, who feel that the Despite its obvious theoretical importance,
evaluative dimension, though a sufficient defi- there exists a dearth of measurement pro-
nition of attitude, is not itself unidimensional. cedures specifically designed to tease out ambiv-
Much of the early controversy emanated alence from other attitudinal variables. To a
from the first camp, focusing on the question large extent this has been due to the prevailing
of whether cognitive and conative aspects view of attitudes as a directional bipolar evalu-
1
ative response representing, on the one hand,
This paper was adapted in part from the author's
master's thesis conducted at the University of Illinois,
positive feelings, appraisals, and tendencies;
1966, under the direction of Martin Fishbein and and on the other hand, negative feelings,
Ledyard Tucker. appraisals, and tendencies (again, see the
2
The author would like to thank Joel Ager, Alan Bass, review article of Allport, 1935). Yet despite
Reuben Baron, Ira Firestone, and Gary Lichtman for this view, researchers have often found the
reading subsequent drafts of this paper.
Requests for reprints should be sent to Kalman J. need to postulate nondirectional attitude vari-
Kaplan, Department of Psychology, Wayne State ables (e.g., intensity, affective involvement).
University, Detroit, Michigan 48202. As Scott's recent review of attitude measure-
361
362 KALMAN J. KAPLAN

ment pointed out: flexible, and easy to use of current measure-


The "directional" property raises, but does not auto- ment techniques. It offers to researchers a
matically answer, the question as to how neutral alti- do-it-yourself attitude-measurement kit, easily
tudes are to be conceived. Are they midway between transferable across situations. Second, the SD
positive and negative attitudes? Should they be sub- technique has been buttressed more than other
divided into indifferent and ambivalent attitudes? The
conception of favorable and unfavorable as "opposites" current measurement techniques by a fully
implies that persons will not be found with attitudes developed theoretical model—a mediational
simultaneously at both ends of the dimension. Yet an learning approach viewing semantic space as
alternative formulation might treat degree of favorable- bipolar and denning attitude as the bipolar
ness and degree of unfavorableness as conceptually evaluative dimension.
distinct (although no doubt empirically correlated)
components, on which persons may make, simultane- A concept is rated on the SD as follows:
ously, a variety of position combinations. In other
words, it is only by convention that the direction of an
attitude is conceptualized as a single bipolar attribute (0) (1) (2) (3)
[Scott, 1968, p. 206].
in which the scale positions have already been
In other words, the measurement of ambiv- defined for the subject in the instructions as:
alence seems to require a situation in which
( — 3) extremely X (3) extremely Y
an individual has the opportunity to simul- (-2) quite X (2) quite Y
taneously indicate both a favorable and an (-1) slightly X (1) slightly Y
unfavorable attitude toward a given stimulus (0) neither X nor Y; equally X and Y
object. Such a technique represents a marked
contrast to the typical measurement procedures When an individual is asked to rate a con-
discussed above which allow a respondent to cept on a SD scale, he is instructed to make
make one and only one evaluative response to only one check mark. He can respond with:
a given object. This overall response is then V
taken as an indication of either a favorable, X
"("-3) (-2) (-1) (0) (1) (2) (3)
a neutral, or an unfavorable attitude, and we or
are left with ambiguity in interpretation of neu-
trality (i.e., indifference versus ambivalence). V - Y
X
(-3) (-2) (-1) (0)' (1) (2) (3)
The argument might be raised that measures
of ambivalence are actually obtainable from He cannot, however, attempt to indicate
the usual directional procedure through the ambivalence toward a concept by checking
separate consideration of items eliciting re- both sides of the continuum:
sponses that indicate favorable and unfavor-
able attitudes (e.g., adjective checklist). This X V _y_
(-3) (-2) (-1) (0) (1) (2)
does not seem entirely satisfactory, however,
since it is possible that individuals may in fact For Osgood's major purposes this represents
be ambivalent to each item and still indicate a no problem. He bases his theoretical under-
favorable, neutral, or unfavorable response to pinnings for the SD technique on the assump-
it. Thus, the ambivalence problem is only re- tion of reciprocally antagonistic XY pairs (i.e.,
moved one level, but not dealt with directly. an object cannot be both X and Y), thus,
A solution seems to demand a basic modifica- making the question of ambivalence irrelevant.
tion in the measurement techniqvie itself. In Osgood's own terms:
First, since [italics added] the polar terms X and Y are
Ambivalence and the Semantic Differential meaningful opposites, we assume that the rm pattern
characteristic of X will be reciprocally antagonistic to
The technique to be offered in this article that characteristic of Y (i.e., wherever a component of
represents a modification of the semantic dif- X is rm, the same component of Y will be fm, and con-
ferential (SD) scale (Osgood, Suci, & Tannen- versely). Second, since [italics added] as will be seen
in subsequent chapters, scales are chosen which maxi-
baum, 1957). The SD scale is used as the point mize one factor or component and minimize all others,
of departure for several reasons. For one, the the rm pattern elicited by an X-Y set will tend to have
SD technique is probably the most popular, one dominant component [Osgood et al., 1957, p. 29],
AMBIVALENCE VERSUS INDIFFERENCE 363

Obviously, the assumption of bipolarity is marking SD instructions often force the sub-
critical to the question of ambivalence. Un- ject to perform a psychologically impossible
fortunately, the restrictions of the single- task—to directionally evaluate a concept in a
marking SD instructions have made it im- manner eliminating ambivalence when he is
possible for this assumption to be adequately faced with adjective pairs he does not judge to
tested. Rather, because of the forced bipolar be bipolar.
structure of the SD scale, polar adjectives are The matter, however, is more complicated
arbitrarily required to be reciprocally antagon- than has been suggested. Though Osgood et al.
istic leaving no room for disconfirmation. proposed a semantic theory in which opposite
Green and Goldfried (1965) have pointed to pairs are reciprocally antagonistic, they, in
this problem and advocated the use of scales practice, allowed the subject a response through
that do not force the reciprocal antagonism as a which he can try to indicate that he believes
means of determining whether or not semantic both sides of the scale are appropriate for his
space and individual semantic differential response to the concept. Consider the middle
scales actually are bipolar. (neutral) category: it is alternatively defined
The technique Green and Goldfried have as "neither X nor Y" or "equally X and Y."
used involves the intercorrelation of sets of In other words, a check in this "neutral" cate-
single-adjective unipolar scales each con- gory can indicate either (a) neither X nor Y
structed from the same bipolar adjectival SD is an appropriate response to the concept (i.e.,
scale. The specific assumptions and instructions feelings of indifference exist toward the con-
of this single-adjective technique are discussed cept) ; or (b) both X and Y are equally appro-
shortly as a way of contrast with the technique priate responses to the concept (i.e., feelings
proposed in the present article. Basically, of ambivalence exist toward the concept).
however, negative intercorrelations between Through their dual definition of the neutral
any two single-adjective scales are interpreted category, Osgood et al. have built a safeguard
as indicating the bipolarity of these adjectives. into their instructions to allow their usage
Specifically, they calculated 9,750 correlation whether or not XY pairs are perceived as
coefficients and found that certain adjective reciprocally antagonistic. Their safeguard, how-
pairs (such as good-bad) tended to be highly ever, has been exacted at the price of clarity of
negatively correlated with each other for interpretation of results not only at the neutral
certain concepts. However, as an overall pat- category but at every category on a SD scale.
tern, Green and Goldfried were forced to Allowing an ambivalent "neutral" response in
conclude: the context of single-marking instructions
There were wide differences in occurrence of bipolar
produces a fundamental change on the inter-
factors on concepts. There were also wide differences pretation of the SD instructions. Specifically,
in the occurrence of semantic differential paired op- it serves to transfer the quality of reciprocal
posites across scales. Any tendency for reciprocally antagonism from the XY pairs to the instruc-
antagonistic adjectives to form the two poles of a tions themselves.3 In effect, subjects are not
single scale was clearly dependent on the concept or set
of concepts involved and therefore was not a generalized rating the concept on the XY scale but are
phenomenon [Green & Goldfried, 1965, p. 30]. instead rating the "net" (Y minus X) qualities
of the concept. If the Y qualities are dominant,
Green and Goldfried's technique has been the Y half of the scale is checked. If the X
recently criticized by Heise (1969) on the qualities are dominant, the X half of the scale
grounds that unipolar scale ratings are more is checked.
affected by the unintended denotative, periph- The implications of this analysis for attitude
eral, and fleeting aspects of adjectives and measurement are critical. That a subject can
thus have more sources of variance than the indicate ambivalence through a single mark
double-anchored bipolar scale ratings. Never- allows him to rate concepts on scales whether
theless, Heise concluded that a number of SD or not the defining XY adjective pairs are
scales exist that do not meet the assumption judged as evaluatively bipolar. Consider, for
of true bipolarity. This conclusion has direct 3
The author would especially like to thank Ledyard
implications for the present article: single- Tucker for his patience in clarifying this problem.
364 KALMAN J. KAPLAN

example, the adjectives "kind" and "wise," a ment of the Y and X components. Such a pro-
pair evaluatively similar (both positive) and cedure has markedly different demands than
not reciprocally antagonistic (i.e., many wise that required to simply test for the bipolarity or
people are also kind). Allowing ambivalent reciprocal antagonism of XY pairs. The latter
"neutral" responses, however, makes the procedure is exemplified by the previously cited
scale usable by forcing subjects to transfer single-adjective technique (Green & Goldfried,
their rating from people to the net "kind minus 1965) and involves only the gross test for the
wise" qualities of people. On these "net" presence or absence of ambivalence without
qualities, of course, it is true but trivial to say any attempt at its quantification.
that "kind" and "wise" are reciprocally antag-
onistic pairs (i.e., the Y minus X characteristics Single-Adjective Scale Technique
of a concept cannot be both Y and X directed,
though the concept itself can be). Specifically, Instead of using the standard SD scale that
the consideration of single-marking instructions is comprised of adjectival opposites (e.g.,
and the denning of the neutral category as good-bad), Green and Goldfried measured the
"equally X and Y" allow (force?) an individual conative meaning of concepts by means of
to cognitively redefine each of the scale single-adjective scales (e.g., good and bad).
positions as follows: They did this by having individuals rate a
concept according to whether the scale was
(—3) extremely more X (3) extremely more Y positively or negatively related to it. However,
thanY than X
for the usual SD scale, only one check mark was
(—2) quite more X (2) quite more Y
thanY than X allowed. Specifically, the instructions have
(—1) slightly more X (1) slightly more Y been described in detail:
thanY than X
(0) equally X and Y (of which "neither X nor Y" If you feel that the descriptive adjective (e.g., good) is
very positively related to the word at the top of the
is a special case)
page, you should place your check mark as follows:
Thus, the SD instructions, while intended to : : : : : :V
sidestep the concept of ambivalence, seem in
actuality to have allowed it to creep in, un- Or, if it is very negatively related, then:
measured, at every scale point and perhaps V : : : : :
affect directional attitudinal markings as well.
What is needed is a modification of the SD If the descriptive adjective is quite positively (but not
technique that will allow the isolation of the extremely) related to the word, then:
ambivalence factor in attitude. : : : : : V :
As a rough definition, Roger Brown's (1965)
view of ambivalence as "mixed feelings," Or, if it is quite negatively related, then:
positive and negative sentiments concentrated : V : : : :
on the same object will suffice. Qualification of
ambivalence, however, requires a more precise If the descriptive adjective seems only slightly posi-
working definition. A lead to this effect is pro- tively related (but is not really neutral) to the word,
vided by J. Brown and Farber (1951) in their then:
discussion of conflict theory. Basically, they : : : : V : :
suggested that both the absolute and relative
strengths of the competing tendencies (i.e., Or, if it is only slightly negatively related, then:
approach and avoidance tendencies) are im- : : V : : : :
portant. Applying their reasoning to the area of
attitudes, Scott (1966) argued that the greater If you consider the word to be neutral on the adjective
and the more equal the opposite tendencies scale, or not at all related to the word, then you should
place your check mark in the middle space [Green &
(i.e., the favorable and unfavorable com- Goldfried, 1965, p. 5].
ponents) the higher is the degree of ambiv-
alence. This working definition demands a As reported earlier, the criteria Green and
technique that allows the separate measure- Goldfried used for bipolarity of XY adjectival
AMBIVALENCE VERSUS INDIFFERENCE 365

pairs are predicated on high negativity in corre- Attitudinal Component Technique: A Modifica-
lation between the X and Y single-adjective tion of the Semantic Differential
responses. Green and Goldfried couch their
criteria in more factor-analytic language: To briefly summarize up to this point, we
have argued that the combination of single-
If the semantic differential scales, and therefore also marking instructions and double definition of
semantic space, are actually bipolar, several correla- the neutral category allow (force?) an indi-
tional and factorial results should follow. For example, vidual to cognitively redefine each of the
factor analyses of sets of single-adjective scales, each single SD scale positions into net (Y minus X)
constructed from the same bipolar adjectival scale,
should have equal but opposite loadings on the same judgments. We have further argued that the
factor. To illustrate, consider the good-bad scale, measurement of ambivalence requires a tech-
which has been found to have a .88 loading on the nique that will separate the X and Y compo-
evaluative factor (Osgood et al., 1957). If the bipolarity nents inherent in any overall (net) attitudinal
assumption is valid, one would expect the good scale
to have a loading of approximately .88 on the evalu- response. Green and Goldfried's single-adjec-
ative dimension and the bad scale to have a mirror- tive scale technique for testing bipolarity has
image loading—that is, —.88 [Green & Goldfried, been rejected as an ambivalence measurement
1965, p. 4]. on the grounds that it fails to clearly isolate
X and Y components. Allowing double mark-
The rationale behind these criteria seems to ing on the traditional semantic differential, an
be as follows: to the extent that XY adjectival alternative not previously discussed, can be
pairs are perceived as bipolar, individuals will rejected on these same grounds. However, it
tend to implicitly insert Y at the open end of does suggest a modification of the SD tech-
the X single-adjective scale and X at the open nique designed specifically to separate out the
end of the Y scale. For example, if an individual positive (Y) and negative (X) components
perceives good and bad as reciprocally antago- inherent in any bipolar attitudinal response.
nistic adjectives vis-a-vis a concept, he would The usual bipolar evaluative SD scales are
tend to treat as a bipolar good-bad scale both presented with typical instructions (i.e., single-
the good and bad single-adjective scales (the marking limitations plus dual definition of the
good and bad responses should be negatively neutral category).4 In conjunction with these
correlated). To the extent that good and bad scales, however, subjects can be independently
are not perceived as reciprocally antagonistic, presented with unipolar positive and negative
an individual's single-adjective good and bad component scales.
responses should tend to be relatively inde- That the component technique is not simply
pendent of one another (i.e., they should be intended to generate single-adjective scales in
uncorrelated). Two explanations seem reason- a slightly different guise (i.e., is devoted to
able. Either the individual might treat both actually measuring X and Y components) is
single-adjective scales as unipolar or implicitly evidenced in the instructions that emphasize
define their open ends with other adjectives he the ignoring of opposing characteristics in
might view as reciprocally antagonistic (e.g., each unipolar response. Specifically, the com-
denning evil as the opposite of good). ponent instructions are as follows: on the
In no sense, however, can this above tech- positive component (Y) scale (i.e., the "liking"
nique be viewed as capable of separately scale), subjects are asked to make the following
diagnosing the individual's X and Y com- judgment. Considering only the positive (Y)
ponents making up any overall (net) XY rating qualities of a concept and ignoring its nega-
and thus the amount of this ambivalence. This tive (X) ones, evaluate how positive (Y) its
is true because the single adjectival X and Y positive (Y) qualities are on a 4-point unipolar
ratings are hopelessly confounded with the positive (Y) scale:
question of XY reciprocal antagonism. What is
: : : Y
needed is a technique that allows separate
diagnosis of the X and Y components in- 0 1
herent in any overall XY rating independent 4
To give face meaning to the notion of positive and
of the relation of the members of the XY pairs negative components, the bipolar scales always run
to one another. from —3 to +3 treating 0 as the neutral point.
366 KALMAN J. KAPLAN

where the categories are defined as follows: The question of the interrelationships be-
tween A, Ap, and An is more complex. Let us
(0) not at all Y (zero Y)
examine the Ap~An relationship first. Though
(1) slightly Y
these two component scores are independently
(2) quite Y
measured, it is by no means certain that sub-
(3) extremely Y
jects are capable of making component rather
On the negative component (X) scale (i.e., the than single-adjective judgments (i.e., are
"disliking" scale), subjects are asked to make capable of ignoring negative qualities while
the following judgment. Considering only a making positive judgments, and vice versa).
concept's negative (X) qualities and ignoring In their single-adjective technique, Green and
its positive (Y) ones, evaluate how negative Goldfried (1965) offered a negative inter-
(X) its negative (X) qualities are on a 4-point correlation as the test for reciprocal antagonism
unipolar negative (X) scale: between XY pairs (i.e., for no ambivalence in
X : : : the case of evaluatively loaded pairs). To the
extent that subjects treat A p and A n as single-
-3 -2 -1 0
adjective scores, they too should be negatively
where the categories are defined as follows: intercorrelated. Zero correlations, on the other
(0) not at all X (zero X) hand, would indicate that A p and A n are truly
(-1) slightly X independent component scores (i.e., that the
( — 2) quite X potential exists for ambivalence).
(—3) extremely X The results (see Table 3) are quite gratifying
in this regard. For the same samples and atti-
Three independent measures are thus gener- tude objects used in the calculations of reli-
ated by this technique: first, the usual bipolar abilities (again objects with naturally exist-
attitude (.4), second, its positive or "liking" ing cognitive structures), the correlations
component (Ap), and third, its negative or between Ap and An ranged from —.13 to .28
"disliking" component (An). Before we turn (f — .05).6 In other words, subjects seem to be
to the question of estimating ambivalence, it is capable of treating their component judgments
necessary to examine the stability of these independently even when the component
measures and their interrelationships. Let us scales are presented in close physical proximity
turn to the first consideration. Reliability to each other. Informal self-reports from sub-
coefficients for A, A,,, and An have been jects also indicate that they are, with little
obtained over a variety of "real attitude difficulty, able to concentrate on one pole while
objects" (i.e., objects with naturally existent ignoring the other. That zero correlations are
cognitive supports) for six independent samples obtained through the components technique
ranging in size from 67 to 236. For each sample, while negative correlations are obtained by the
correlations are calculated across subjects and single-adjective technique suggests the former
objects. Test-retest reliabilities range from .81 may represent a more appropriate test for even
to .93 for each of the three scores (regardless of the gross question as to the existence of ambiv-
the exact nature of the underlying evaluative alence independent of its quantitative
scale)6 significant at well beyond the .001 level properties.
and indicative of a high degree of stability. The contention that Ap and An actually
6
While this technique seems applicable to any seman- measure independent components is further
tic dimension, it is the evaluative dimension, of course, supported through examination of the under-
which is relevant to the question of attitudinal ambiv- lying cognitive structure each of the measures
alence. Various evaluative scales generated all the reflects. Table 1 presents correlations between
data presented in this article—for example, Fishbein the directly obtained liking (Ap) and disliking
and Raven's A scales (containing the scales good-bad,
clean-dirty, harmful-beneficial, healthy-sick, and wise- (An) components toward the attitude object
foolish), the good-bad scale alone, and the like-dislike and the estimates of these components based,
scale. As the intercorrelations between each pair of respectively, on the simple number of positively
scales consistently exceeded .80, data deriving from
0
any or all of these scales will be treated equivalently— The average correlations are estimated through r to
that is, simply labeled as A, A p, or A „. z transformation.
AMBIVALENCE VERSUS INDIFFERENCE 367

and negatively loaded beliefs (i.e., amount of TABLE 1


positive and negative information) elicited by AVERAGE CORRELATIONS BETWEEN ABSOLUTE
subjects about the object. The results indicate ATTITUDE COMPONENTS AND NUMBER OF
that while the number of positive (negative) SUPPORTING BELIEFS
beliefs provides a moderately good average Number of
estimate (f = .44, p < .01, in both cases) of beliefs AP An

absolute liking (disliking), it provides only a


.44** -.18**
very weak average inverse estimate (f = — .16 Positive (.31* to .52**) (-.11 to -.23)
and —.18, respectively) of absolute disliking -.16** .44**
(liking). Negative (-.02 to -.37**) (.32*^.51**)
A second question of concern involves the
Note.—Ranges are in parentheses. These correlations were
relative influence of A p and A „ in the determi- calculated across subjects and objects for each of three inde-
pendently measured samples ranging in size from 67 to 172
nation of A and the extent of the correspon- (total N = 320). A significance test for each of these correla-
dence between A and its best component tions is based on the respective sample size it is calculated on.
The average correlations, in contrast, are calculated through
prediction. A noted attitude theorist, Jordan r to z transformations and based on 311 degrees of freedom
(see McNemar, 1962, p. 140).
(1965), has criticized an oversimplified "push- * p < .05.
**p <.01.
pull" model of attitudes and suggested that
positivity and negativity may work in different
tation of the bipolar SD scale, of course,
ways. In like manner, several generations of
becomes superfluous as it is possible to diagnose
linguists (e.g., Bierwisch, 1967; Greenberg,
it from the component scores. In the remainder
1966; Lyons, 1963, 1968; Sapir, 1944) have
of this article, in any case, no distinction is
observed that word pairs like good and bad are
made between A and A' = Ap + An.
not symmetrical because of differential lexical
markings. They argue that good, the unmarked
member of the pair, can be neutralized in some Attitude Component Model: The Development of
contexts as in "How good was the movie?" N ondirectional Attitude Indexes
whereas bad, the marked term, cannot (see The foundation of our component technique
Clark, 1969). The implication for the present laid, let us turn to the problem of developing
problem, of course, is differential weighting an index of ambivalence and a model wherein
for the A p and A n components in the determi- such an index can be directly interpreted. To
nation of A. The evidence collected up to this begin with, the demonstration of the relative
point on a variety of attitude objects for each independence of the Ap and An components
of the six samples described above, however, and the high correspondence between their
suggests a large range in the influence of both unweighted sum and A, the overall attitude,
Ap (r = - .15 to .59) and An (r = - .04 to allow us to draw Ap and An as orthogonal
.56) on A. Thus, either component seems dimensions. Thus, in Figure 1, Ap is repre-
capable of being either the dominant determi- sented by the ordinate and Anby the abscissa
nant of overall attitude or, at times, even a mild allowing the representation of every attitude
suppressant variable. In line with these results, object by a point in the 2-space (An,Ap). If
a multiple regression analysis suggests both the two components are scaled with equal
equal weighting of Av and An and a high units, the 45-degree line corresponds to a zero
correspondence between their unweighted sum marking on the bipolar SD scale and can be
(A' = Ap + An) and A (r = .89 to .97, labeled the line of ambivalence (A = Ap
p < .001). + An = 0). Three nondirectional attitude
Thus, the independently derived A score measures become directly interpretable around
can be tentatively interpreted as the simple this line: total affect (TA), polarization (POL),
unweighted net summary of the Ap and An and ambivalence (AMB).
components (A «* ^4')' A positive A indicates Of these three indexes, TA is the most
that, overall, the concept is more liked than it inclusive, referring simply to the total amount
is disliked, a negative A indicates the opposite, of affect directed toward the object regardless
while a zero A indicates that the liking and of sign. It is calculated through summing the
disliking components are equal. The presen- absolute values of the liking and disliking
368 KALMAN J. KAPLAN

10,11

10,21 £.
in

1-3,0) 1-2,01 •10,01

\n (Disliking Component) / 'V

p
b
TA=Ap+|Anl
I* '
POL=|A|
AMB-TA-POL
FIG. 1. A geometrical representation of the attitude component model.

components (TA = Ap + \An ) 7 and is easily some line (A = a) parallel to the line of ambiv-
interpretable via the components model pre- alence—a equals zero for the line of ambiv-
sented in Figure 1. The origin (0,0) is the only alence. When the positive component of an
point on the line of ambivalence wherein an attitude is greater than its negative component,
interpretation of "no affect" or indifference is A is positive and the point representing that
justified (TA=Ap+\An = 0 ) . All other attitude will fall above the line of ambivalence
points on this line, in contrast, indicate that (the 45-degree line). When the opposite is
an individual does possess both positive and true, A is negative and the point will fall below
negative affect—the farther out along the line the line of ambivalence. Perpendicular to the
the greater the affect—which happen to cancel family of A lines (A = a) lies the family of TA
out in net terms (A =0). lines (TA = f). Every attitude score can be
For points on the line of ambivalence (i.e., described as lying on the intersection of a line
whenever ^ 4 = 0 ) , TA serves as a good indi- from the TA family with a line from the A
cator of ambivalence (i.e., the total amounts of family. TA does not itself provide a good
liking and disliking completely cancel out). index of ambivalence for points not falling on
Clearly, however, the component model is not the line of ambivalence. Though the individual
limited to the line of ambivalence. It is in- may possess both positive and negative affect,
tended to describe all points in the 2-space, the the nonzero net attitude indicates that the two
projection of every attitude object falling on components do not completely cancel out.
' As A p > 0 and A „ < 0, placing an absolute value Thus, TA contains two distinct segments—
around A? is superfluous. AMB or the amount of exactly counterbalanc-
AMBIVALENCE VERSUS INDIFFERENCE 369

ing positive and negative affect, and POL or TABLE 2


the absolute value of the directional residual. HYPOTHETICAL Af-An COMBINATIONS AND THEIK
For the line of ambivalence, POL is zero and EFFECTS ON OTHER ATTITUDINAL INDEXES
AMB = TA. For other points, however,
A MB = TA - POL. An (Disliking) A, (Liking)
0 i 2 3
In other words, for a neutral SD response,
AMB represents the total underlying affect 0 A" 0 1 2 3
(TA). For nonneutral responses, however, TAh 0 1 2 3
POL" 0 1 2 3
AMB represents only one segment of TA; an AMB* 0 0 0 0
independent contribution is provided by the
absolute value of the directional score itself -1 A -1 0 1 2
(POL). TA 1 2 3 4
Further examination of Figure 1 provides a POL 1 0 1 2
AMB 0 2 2 2
geometrical interpretation of each of these
indexes. A perpendicular may be dropped from _2 A -2 -1 0 1
any attitude point in the 2-space to the axis TA 2 3 4 5
(ordinate or abscissa) on the opposite side of POL 2 1 0 1
AMB 0 2 4 4
the line of ambivalence. The length of this
perpendicular from the attitude point to its -3 A -3 -2 -1 0
intersection with the line of ambivalence TA 3 4 5 6
represents an index of the POL of this point POL 3 2 1 0
AMB 0 2 4 6
while the length of the line of ambivalence
itself from the intersection to the origin repre- -A ~A'
sents a rough index of its AMB derivative. TA, >>TA ~A, + \A,\.
"POL = \A\.
of course, is denned as the sum of the direct line o- AMB = TA - POL.
between the attitude point and the origin.8
The relationships between these indexes may and AMB (its two contributors), while they
be formally summarized as follows : themselves must be negatively related. Under
A~A' = Ap + An [1] high POL, little AMB can occur. Under high
AMB, little POL can occur. Under moderate
TA=Av+\An\ [2] ranges, however, both AMB and POL can
occur. The empirical question, of course, is the
POL= \A\ [3] relative occurrence of A -A combinations as
p n
AMB = TA - POL [4] naturally existing states and their effects on the
interrelationships of the other indexes (e.g.,
It can be seen that POL = A f + A „ and whether POL or AMB is the more important
An\ - A An. Thus, if contributor to TA).
A j , = - An, A and POL = 0 and TA = AMB. As previously reported, Table 3 indicates
If either Ap or An = 0, AMB = 0 and the relative independence of the A,, and An
TA = POL. A more complete delineation of components. As expected, TA, POL, and AMB
the effects of hypothetical Ap-An combina- are all relatively independent of directional
tions on TA, POL, and AMB is presented in attitude (A) (f = - .12, -.07, and -.10,
Table 2. Theoretically, at least, A can vary respectively; df — 737 in all cases), yet mod-
somewhat independently of TA, POL, and erately related to the absolute values of both
AMB. The latter three indexes, because of the liking (Ap) (f = .71, .45, and .54, respectively;
dependent nature of their derivations, must be 'p < .01 in all cases) and disliking (An)
somewhat interrelated— specifically, TA must (f = .60, 54, and .35, respectively; df = 737,
be somewhat positively related to both POL p < .01 in all cases). It should be noted,
8 however, that a considerable range in each of
The length of these latter two lines, as any student
of geometry is already aware, represents underestimates these correlations obtains across the six
of A MB and TA . These estimates are used for simplicity samples. The variability seems primarily a
in presentation. function of the relative strengths of the Ap and
370 KALMAN J. KAPLAN

TABLE 3
AVERAGE INTEKCORRELATIONS OF ATTITUDE COMPONENT INDEXES

Measure At An TA AMB POL

A .28** -.32** -.12** -.07 -.10*


(-.15, .71**) (-.67**, .04) (-.47**, .22) (-.36**, .18) (-.48**, .24)
A, -.05 .71** .45** .54**
(-.28*, .13) (.60**, .90**) (.29*, .62**) (.44**, .70**)
An .60** .54** .35**
(.41**, .80**) (.38**, .71**) (.08, .42**)
TA .55** .54**
(.35, .75**) (.38, .64**)
AMB -.40**
(.32*, -.49**)
POL
Note. — Ranges are in parentheses. These correlations were calculated across subjects and objects for each of six independently
measured samples ranging in size from 67 to 236 (total N = 755). A significance test for each of these correlations is based on the
respective sample size it is calculated on. The average correlations, in contrast, are calculated through r to a transformations and
, 140).
* p < .05.
** p < .01.

An components across objects. The objects differ markedly across categories in a manner
themselves consist of ethnic categories, insti- consistent with this pattern. In line with
tutions, current events, and specific persons. these results are the relative contributions of
Internal analysis suggests some general though POL and AMB to TA (ranging from .35 to
by no means exceptionless tendencies. For .75 for AMB and from .38 to .64 for POL). It is
specific people and ethnic categories, "warm is true but trivial to point out that with objects
the norm" (little An); for institutions, in eliciting high polarization and low ambivalence,
contrast, the opposite pattern seems to hold POL becomes the more important contributor
(little Ap)—this is hardly surprising as our to TA. For highly ambivalence-producing
subjects are predominantly students in their objects, AMB becomes the more important
late teens or early twenties. The remaining contributor.
category, current events, seems to be treated As the reader will no doubt recall, the
more evenhandedly (eliciting both A p and A „). validity of the A p and A „ scores as components
Though no one sample contains a monopoly was buttressed by examination of their re-
on any one type of attitude object, the rela- spective cognitive supports (i.e., number of
tive occurrence of the various types (specifically positive and negative beliefs). In like manner,
ethnic categories versus institutions) does such analyses were performed on A, TA, POL,
and AMB. The theoretical point of departure
TABLE 4
derives from an observation by Brehm and
Cohen (1962).
AVERAGE CORRELATIONS BETWEEN ATTITUDINAL
INDEXES AND NUMBER OF SUPPORTING BELIEFS It is generally assumed that as a person's attitude varies
from neutral in either direction, he is likely to have
Measure Total number of beliefs Range increasing amounts of relevant information [greater
number of beliefs] about the issue. We hasten to point
A .06 .04-.07 out, however, that this correlation does not necessarily
TA .35* .12-51* hold, since it is perfectly possible for a person to have a
POL .07 .01-.14 "neutral" position because he possesses conflicting
AMB .30* .02 -.45* information. [Brehm & Cohen, 1962, p. 14].

Note.—These correlations were calculated across subjects The implication of this suggestion for the
and objects for each of three independently measured samples components model is straightforward. While
ranging in size from 67 to 172 (total N = 320). A significance
test for each of these correlations is based on the respective number of beliefs should be directly related to
sample size it is calculated on. The average correlations, in
contrast, are calculated through Y to z transformations and TA and AMB indexes, which do not cancel out
based on 311 degrees of freedom (see McNemar, 1962, p. 140).
* <.01. conflicting attribute evaluations, it should be
AMBIVALENCE VERSUS INDIFFERENCE 371

relatively unrelated to A and POL. The re- informal object (F = 12.95, df = 2/627,
sults in Table 4 indicate that this indeed seems p < .001).
to be the case for naturally existing objects, In the context of this particular study, the
though the results again are variable across results tend to support a little-noted suggestion
samples, highly polarized objects tending to that neither positive nor negative feeling is
raise the correlations of A and POL with likely to decrease with increased contact (Park
number of beliefs and lower its correlation & Burgess, 1927, p. 283). Of more general
with TA and A MB. This trend is even more importance, however, is the evidence that the
pronounced when the number of beliefs and the component technique allows us to distinguish
extremity of their evaluative loadings are between increasing liking (disliking) and de-
under experimental control, initial results creasing disliking (liking) as the source of
suggesting that the presentation of sets of positive (negative) attitude change.
highly polarized but directionally balanced
attributes (varying in size from 6 to 12) gener- REFERENCES
ates greater TA and AMB toward an abstract
cipher than equally sized, directionally balanced ALLPORT, G. Attitudes. In C. Murchison (Ed.), Hand-
book of social psychology, Worcester, Mass.: Clark
attribute sets of lesser polarization (p < .01 University Press, 1935.
in both cases). Both sets of attributes, however, AMIR, Y. Contact hypothesis in ethnic relations. Psy-
tend to produce near neutral directional atti- chological Bulletin, 1969, 71, 319-342.
tudes (^4) and polarizations (POL}. BASS, A. R., & ROSEN, H. Some potential moderator
One application of the component technique variables in attitude research. l<Aucational and
should be mentioned in closing. In an investi- Psychological Measurement, 1969, 29, 331-348.
BIERWISCH, M. Some semantic universals of German
gation of the contact hypothesis (see Amir, adjectivals. Foundations of Language, 1967, 3, 1-36.
1969; Heider, 1958; Newcomb, 1961), the BREHM, J. W., & COHEN, A. R. Explorations in cogni-
author, along with several colleagues9 ob- tive dissonance. New York: Wiley, 1962.
tained A p and A „ as well as A measures. Their BROWN, J., & FARBER, I. Emotions conceptualized as
results supported a predicted interaction on intervening variables with suggestions toward a
theory of frustration. Psychological Bulletin, 1951,
overall attitude (A) between the formality- 48, 465-495.
informality of an attitude object and the degree BROWN, R. Social psychology. New York: The Free
of expected contact with it. Specifically, Press, 1965.
subjects' attitudes became less favorable over- CLARK, H. H. Influence of language on solving three-
all toward the formal object and more favorable term series problems. Journal of Experimental Psy-
chology, 1969, 82, 205-215.
overall toward the informal object with in- FISHBEIN, M. An investigation of the relations between
creasing potential for contact (decreasing beliefs about an object and the attitude toward that
distance) (F = 29.39, df = 3/627, p < .001). object. Human Relations, 1963, 16, 233-279.
FISHBEIN, M. A consideration of beliefs, attitudes, and
Analysis of the liking (Ap) and disliking (An) their relationships. In I. D. Steiner & M. Fishbein
gradients enabled us to pinpoint the locus of (Eds.), Current studies in social psychology. New
these effects. Subjects' liking remained rela- York: Holt, Rinehart & Winston, 1965.
FISHBEIN, M. A consideration of beliefs and their role
tively constant with decreasing distance to the in attitude measurement. In M. Fishbein (Ed.),
formal object but tended to increase with Readings in attitude theory and measurement. New
nearness to the informal object (/*' = 12.68, York: Wiley, 1967.
df = 2/627, p < .001). On the other hand, GREEN, R. F., & GOUMRIED, M. R. On the bipolarity
of semantic space. Psychological Monographs, 1965,
subjects' disliking increased with nearness to 79 (6, Whole No. 599).
the formal object, while remaining relatively GREENBERG, J. H. Language universals. The Hague:
constant with decreasing distance from the Mouton, 1966.
HEIDER, F. The psychology of interpersonal relations,
9
The study originally reporting these results, "Atti- New York: Wiley, 1958.
tude Object Formality-Informality and Resultant HEISE, D. R. Some methodological issues in semantic
Attitude Gradient Slopes," was prepared by the author differential research. Psychological Bulletin, 1969, 72,
in conjunction with Ira Firestone, Melvin Kimmel, and 406-422.
Richard Segan for presentation at the 1970 meeting of HERZBERG, F., MAUSNER, B., & SNYDERMAN, B. B. The
the Eastern Psychological Association. motivation to work. New York: Wiley, 1959.
372 KALMAN J. KAPLAN

JORDAN, N. The "asymmetry" of "liking" and "dis- ROSENBERG, M. J. Cognitive structure and attitudinal
liking" : A phenomenon meriting further reflection affect. Journal of Abnormal and Social Psychology,
and research. Public Opinion Quarterly, 1965, 29, 1956, 53, 367-372.
315-322. ROSENBERG, M. J. A structural theory of altitude dy-
K.OMORITA, S. S., & BASS, R. R. Attitudes differenti- namics. Public Opinion Quarterly, 1960, 24, 319-340.
ation and evaluative scales of the semantic differ- SAPIR, E. Grading: A study in semantics. Philosophy of
ential. Journal of Personality and Social Psychology, Science, 1944, 11, 93-116.
1967, 6, 241-244. SCOTT, W. A. Brief report: Measures of cognitive
LYONS, J. Structural semantics. Oxford: Blackwell, 1963. structure. Multivariale Behavior Research, 1, 1966,
LYONS, J. Introduction to theoretical linguistics. Cam- 391-395. (Abstract)
bridge: Cambridge University Press, 1968. SCOTT, W. A. Attitude measurement. Tn G. Lindzey &
McNEMAR, Q. Psychological statistics. New York: Wiley, E. Aronson (Eds.), Handbook of social psychology.
1962. Vol. 2. Reading, Mass: Addison Wesley, 1968.
NEWCOMB, T. M. The acquaintance process. New York: SMITH, R. G. A semantic differential for theatre con-
Holt, Rinehart & Winston, 1961. cepts. Speech Monographs, 1961, 28, 1-8.
OSGOOD, C. E., Sucr, G. J., & TANNENBAUM, P. 11. WIGGINS, N., & FISHBEIN, M. Dimensions of semantic
The measurement of meaning. Urbana: University of space'. A problem of individual difference. In J. G.
Illinois Press, 1957. Snider & C. E. Osgood (Eds.), Semantic differential
PARK, R. E., & BURGESS, E. W. Introduction to the technique: A sourcebook. Chicago: Aldine, 1969.
science of sociology. Chicago: University of Chicago
Press, 1927. (Received August 11, 1970)

NOTICE TO AUTHORS
In response to a request from the APA Publications Board, the selected reviewers
of Psychological Bulletin have agreed to participate in a blind reviewing system.
An effort will be made to insure that the individuals who review manuscripts will
have no information regarding the identity of the author or of the institution with
which the author is affiliated.
Authors submitting manuscripts to Psychological Bulletin are requested to include
with each copy of the manuscript a cover sheet, which contains the title of the manu-
script, the name of the author or authors, and the author's institutional affiliation.
The date the manuscript is submitted also should appear on the cover sheet. The
first page of the manuscript should not contain the author's name or institutional
affiliation and should contain the title of the manuscript and the date it is submitted.
Footnotes containing information pertaining to the identity of the author or his
institutional affiliation should be on separate pages. Every effort should be made by
the author to see that the manuscript itself contains no clues as to his identity.
The selected reviewers are not aware that knowledge of the author's identity pre-
viously has influenced their judgments, but they recognize this as a possibility, and
the blind review system therefore is being initiated.
Authors are requested to use the proposed format immediately, and manuscripts
received after July 1, 1972, that do not conform to these suggestions will be returned
to the author.

You might also like