You are on page 1of 17

JOURNAL OF EXPERIMENTAL SOCIAL PSYCHOLOGY 5, 400416 (1969)

The Prediction of Behavioral Intentions


in a Choice Situation

ICEK AJZEN AND MARTIN FISHBEIN


University of Illinois

Two complementary hypotheses, one derived from decision theory


and the other from Fishbein’s (1967) theoretical model, were
tested with respect to the prediction of behavioral intentions. One
hundred subjects completed questionnaires including measures of
attitudes and normative beliefs toward single behaviors and toward
dichotomous and multiple behavioral choices. Consistent with
decision theory notions it was found that behavioral intentions in a
choice situation could be predicted with higher accuracy by con-
sidering attitudes toward all behavioral alternatives than by using
the attitude toward only one of the possible actions. In support
of the prediction based on Fishbein’s model it was found that be-
havioral intentions for single acts as well as for acts in dichotomous
and multiple choice situations were a function not only of attitudes
toward the acts but also of normative beliefs with respect to these
behaviors.

Ever since LaPiere (1934) failed to find the predicted relationship hr-
tween a paper and pencil measure of attitude and observed behavior, in-
vestigators have sought to account for the discrepancy by postulating a
number of possible explanations. One reaction has been to question th(.
validity of our measuring instruments; a second has been to attack om
definition of attitude, expanding it to include cognitive, affective, and
conative components (cf. e.g., Fishbein, 1967). A widely accepted posi-
tion regards attitudes as merely one of a variety of factors which enter
into the determination of behavior (e.g., Cook and Selltiz, 1964 and
Triandis, 1967). Some of the proposed additional factors are social norms.
expected consequences of the behavior, situational variables, and per-
sonality characteristics of the subjects.
In a recent article, Fishbein (1967) presents a theoretical model for
the prediction of behavioral intentions and corresponding behaviors. The
model is essentially an adaptation of Dulany’s (1967) ‘theory of propo-
sitional control” to social behavior. Its immediate concern, like that of
the original formulation, is the prediction of behavioral intentions which
are assumed to mediate overt behavior. The model’s greatest merit is its
400
PREDICTION OF BEHAVIORAL INTENTIONS 401

striking simplicity. According to the adapted model, behavioral intentions


are a joint function of the attitude toward performing a particular be-
havior in a given situation and of the norms perceived to govern that be-
havior multiplied by the motivation to comply with those norms. The
normative component includes both the individual’s personal beliefs
about what he should do in a given situation (NB,) as well as his per-
ception of others’ expectations about his behavior in that situation, i.e.,
social normative beliefs (NB, >. These relations have been expressed in a
symbolic form which may help to clarify the components involved in
the model:

where B-Overt behavior


BI-Behavioral intentions
A-act-Attitude toward the behavior in a given situation
NB,-Personal normative beliefs
NBS-Social normative beliefs, i.e., perceived expectations of
others
MC,-Motivation to comply with social normative beliefs
w,,, u,, and tu,-Empirically determined weights.

Any additional variable is held to influence BZ if, and only if, it affects
one or more of the model’s predictors. Thus, situational variables, per-
sonality characteristics, etc., will influence a person’s behavioral inten-
tions only if they are related to A-act, to NB,l, or to NB,( IMc,), or if
they influence the relative weights of the three components. The cri-
terion to be predicted is BI, the behavioral intention. BI serves as an
intervening variable between the three predictors and overt behavior. It
is expected to account for most of the behavior variance, i.e., there
should be a high correlation between BZ and B and thus, according to
the theory, if one can predict BZ, one can also predict B with only a
slight attenuation. This high correspondence between BZ and B will, of
course, only obtain if the BZ selected by the experimenter is appropriate
for the particular behavior under study. The more general the behavioral
intention, or the longer the time interval between the statement of in-
tention and the actual behavioral performance, the lower the BZ-B cor-
relation will tend to be. It is therefore the experimenter’s responsibility

I Originally, NB, was multiplied by the subject’s motivation to comply with his
personal normative beliefs ( McP). However, previous experiments have shown that
an individual usually is motivated to do what he himself thinks he should do and
multiplying NB, by MC, failed to yield significant improvements of the correlation
with BI. MC, was therefore eliminated from the model in the present paper.
402 AJZEN AND FISHBEIN

to obtain or select Bl’s that will correlate highly with the behavior he in-
tends to predict3
A-act, the attitude toward a particular behavioral act, deserves special
comment. Unlike traditional approaches to attitude measurement, the
attitude under consideration here is the individual’s attitude toward per-
forming a particular act in a given situation with respect to a given object,
rather than his attitude toward the object or class of objects per se. That
is, we are here interested in the attitude toward performing a specific
behavior and not in the attitude toward a given person, object, or
situation.
It should be noted, however, that just like any other external variable,
the attitude toward an object (A,) may, under certain conditions, be re-
lated to A-act and to the normative components of the model. As indi-
cated above, if A, does affect the model’s predictors it will also be re-
lated to behavioral intentions through them. We shall return to this point
below.
In line with Rosenberg ( 1956)) Zajonc (1954)) and others, Fishbein
(1963) has demonstrated that an individual’s attitude toward any object
can be predicted with a high degree of accuracy from a knowledge of
the individual’s beliefs about the attitude object and the evaluative as-
pects of those beliefs. More specifically, the attitude is conceived of as
the sum of the beliefs multiplied by their respective evaluative aspects.
Based on Dulany’s (1967) theorizing,3 A-act is seen as the sum of the be-
liefs about the consequences of performing a given act (Bi) times the
evaluation of these consequences ( ai) and thus, algebraically
71

A-act = Bia;
c
i=l

It is interesting to note that this conceptualization of attitude toward


an act corresponds quite literally with a number of formulations pro-
posed by other theorists. For example, Rotter’s (1954) social learning
theory maintains that the probability of the occurrence of a given be-
havior in a particular situation is determined by two variables: the sub-
jectively held probability (or expectance) that the behavior in question
will be reinforced and the value of the reinforcer to the subject. Simi-

’ Clearly, the experimenter should therefore independently validate the predicted


BI-B correlation whenever possible. Several studies (e.g., Carlson, 1968; Ajzen, 1967;
Fishbein, 1968; D&my, 1961, 1964, 1967) have demonstrated extremely high
(r=:. 95) BI-B relationships that are obtained when appropriate BI’s are selected.
’ In Dulany’s theory RHd (hypothesis of the distribution of reinforcement) is mul-
tiplied by RSv (subjective value of the reinforcer). This product is A-act in Fish-
bein’s adaptation.
PBEDIClTON OF BEHAVIORAL INTENTIONS 403

larly, both Peak ( 1955) and Rosenberg ( 1956, 1965) view attitudes as a
function of beliefs about the instrumentality of the object in obtaining
goals and the value importance of those goals.,
Of particular interest in this context is the formulation proposed by
the “Decision Theory” model.* Very briefly, decision theory first estimates
the subjective utilities of alternative actions. Different strategies for de-
cision making may then be employed. The most generally useful strategy
is one that leads to the choice of the alternative which maximizes average
gain or minimizes average loss. The subjective average or expected utility
(SEU) of a given alternative is a function of the subjective probability
that certain outcomes will follow the particular act (SPi) multiplied by
the respective subjective values, i.e., utilities, attached to the outcomes
(vi):

SEU = 2 Spill<
i-1

It can be seen that this formula is essentially equivalent to Fishbein’s


( 1967) definition of the attitude toward an act:
n
A-act = B,ui
c
i=l

Applying the strategy of maximizing the subjective expected utility we


may predict that the behavioral alternative for which SEU or A-act is
maximal will occur.

Aims of This Study and Hypotheses


The genera1 purpose of the present research was to combine the no-
tions of decision theory and of Fishbein’s model for the prediction of
behavioral intentions in choice situations.
The following were the major hypotheses of the study:
1. Fishbein’s model was first employed in the prediction of behavioral
intentions for a single behavioral act, i.e., not in a choice situation. Ac-
cording to the model, adding NB, and NB,( MC,) to A-act in a multiple
regression equation will greatly improve the prediction of BI above the
level obtained when only A-act, the attitude toward the act, is used.
2. In the present study A,,, the attitude toward an object, was expected
to be highly related to the model’s predictors. A,, will therefore also be
expected to wrrelate with BI. However, holding A-act, NB,, and NB,
(MC,) constant in computing the partial correlation between A, and BI
should greatly reduce the A,-BI correlation,
‘A mathematical presentation of decision theory may be found in Blackwell ~ncl
Girshik ( 1954). For a review directed toward psychologists cf. Edwards ( 1954).
404 AJZEN AND FISHBEIN

3. When an individual has to choosy from among two or more mu-


tually exclusive behavioral acts, consideration of the attitiide toward each
of the alternatives will yield better predictions of behavioral intentions
in the choice situation than using the attitude toward one of the be-
havioral acts alone.
This prediction is based directly on decision theory, using the strategy
of maximizing the subjective expected utility. Attitude research often
neglects the fact that an individual usually has a choice between alterna-
tive acts, including the choice of not doing anything. If we want to pre-
dict his behavior we need to estimate his attitudes toward each of the
alternatives and take all scores into consideration. Specifically, if a sub-
ject has to choose between two mutually exclusive behaviors it is pre-
dicted that the algebraic difference between his two attitude scores on
the respective behaviors will provide a higher correlation with the be-
havioral choice than the attitude toward any one of the two acts alone.
Similarly, if the subject has to rank order his choices between N
mutually exclusive behaviors, the prediction based on the rank of his
attitude score for behavior zi relative to the attitude scores for the alter-
native behaviors will be closer to the rank order he gives to behavior
xi per se. That is, we propose that the rank order of the attitude scores is
a better predictor of behavioral intentions in a multiple choice situation
than is the attitude score itself. Using the rank order we again take into
account the attitudes toward all behaviors involved.
4. Based on Fishbein’s model it is proposed that, as for single acts,
the prediction of behavioral intentions in choice situations will be further
improved by employing normative beliefs times the subject’s motivation
to comply, in addition to attitudes.
Analogous to the case for A-act it is again predicted that taking the
various normative beliefs for the alternative behaviors into account will
provide better correlations with BI in the appropriate choice situations
than consideration of normative beliefs toward only one of the alternatives.
We predict, therefore, that using the decision theory measures of A-act,
NB,, and NB,( MC,) in a multiple regression equation will predict BI in
choice situations better than A-act alone.
METHOD AND PROCEDURE
One hundred undergraduate students, 57 males and 43 females, participated in the
study. In order to obtain the necessary data, all the subjects were asked to fill out a
seven-part questionnaire.
The Questionnaire
Eight behavioral acts were selected by the experimenter as representing a wide
range of activities students may be expected to engage in. These behaviors were:

1. Going to a party on a Friday night.


2, Visiting an exhibition of modem art on a Friday night.
PRJZDICTION OF BEHAVIORAL INTENTIONS 405

3. Watching a western on TV on a Friday night.


4. Going to a concert on a Friday night.
5. Playing a game of poker on a Friday night.
6. Going to a French movie on a Friday night.
7. Participating in a discussion on a Friday night.
8. Reading a mystery novel on a Friday night.

Each of the eight behaviors was followed by “on a Friday night” in order to make
the situation somewhat more concrete. The questionnaires included the following
measures with respect to each of the eight activities.
A,-attitude toward the object of the activity. Consistent with the previous work
of Osgood, Suci, and Tannenbaum ( 1957) and of Fishbein and Raven ( 1962), these
attitudes were measured by asking the respondent to rate each of the attitude objects
(e.g., “a party, ” “an exhibition of modern art,” “a western on TV,” “a concert,” etc.)
c:n a form of the Semantic Differential. More specifically, the attitude score was
ubtnined by summing the subject’s ratings on four 7-point semantic differential scales
\j+th high loadings on the evaluative factor.
Exarnpk:
Parties are
punishing . ... .. ... . .. : . .... ... .. .... : ....-....-..... : . .. ... ... .... : .. .. .. . .... : .. .... ...-..... : .... ... .... .... rewarding
interesting ..................: ..................: ..................: .....-...._.....: ..................: .....“. ..._..... : ....._....“..... boring
good .... ... ... .... . . .... ... .. .... . ... .. ..... ... . . ..... .. ..^.... . ...” ... ... ... . .....”....-...... ....._....* ..... bad
unp?eaqant .........._I.....:... ... .... ... . . ..... ... ... ..... . ... .... ... . : ...........I.....:....._....“...... .....I...._..... pleasant
A-act--attitude toward a certain act. A-act was assessed by taking the sum over
the same four semantic differential scales as those used for A,. The concepts rated,
however, were the activities themselves as listed above. For example, “going to a
psrty on a Friday night” was rated on the four semantic differential scales, and the
sum taken as an index of A-act with respect to this activity.
NB,-personal normative beliefs with regard to a certain act. Measured on a single
7-point semantic differential scale.
Example:
I personally think I should go to a party
on a Friday night
probable ... ... .... ... . . ..... ... .... .... .. .... ... .... . . ....._...._....... .... ...._..........._.................“...._..... improbable
NBa5-the subject’s perception of his friends’ expectations with regard to a certain
act. It was assumed that the subject’s friends were the most relevant significant others
for the behaviors under consideration. NBS was measured similarly to NB,.
Exampkz:
My friends expect me to go to a party
on a Friday night
probable . ...-............ ... . . .... .... . . ..._...._.......... ...._.............. .... .. .... . .. ... .. .... ... . ....“........... improbable
BI-behavioral intentions with respect to a certain act. BI was measured by a
behavioral differential scale (Triandis, 1964).
Example:
On a Friday night

a In addition to NB, a measure of MC,, the motivation to comply with the expec-
tations of friends, waq taken. But the results showed that multiplying NB, by MC,
in most cases reduced the correlations with BI. This attenuation is probably due to
difficulties in the operational definition of MC, in the present study. It was therefore
decided not to utilize MC. in this study.
406 AJZEN AND FISHBEIN

I would ....._...._.........._...I.I...._:....._............_...“...._.“.: ..“..“..._.....: ....._._.I.....: ._..,_...._..... I would not


go to a party
BI?-behauioral intentions in a dichotomous choice situation. All pos&le p&rings
of the eight behavioral stimuli were presented at opposite ends of a behavioral dif-
ferential scale and the subject had to indicate the degree of his intention to perform
one or the other alternative. The 28 possible pairs were arranged and presented in
random order.
Example:
On a Friday night
I would go to I would go to
a French ....._................. ....._..........._...._..........._...._........ ... ....“...... ....._............. ...“....._.... a concert
movie
can’t
decide
Bl,behavioral intentions in a multiple choice situation with eight different alter-
natives. The subject was asked to rank order the eight alternative behaviors according
to the likelihood of his engaging in them on a Friday night.
On the basis of these measures the following derived scores were computed:
A-acta, NB,a, and NB .,-the algebraic diferences between each of the 28 possible
combinations of A-act scores, NB, scores, and NB, scores, respectively.
A-a&, NBP8, and NB .-the rank order of each of the eight A-act’s, the eight
NB,‘s, and the eight NB.% for each individual subject.
For the purpose of checking the potential influence of the order in which the
different measures were presented in the questionnaire, four different arrangements
were chosen at random from all possible arrangements. That is, the order in which
the A-act questions appeared relative to NB,, to A,, etc., was varied. Twenty-five
subjects completed questionnaires in each of the four arrangements.
The standard instructions (cf. Osgood, Suci, and Tannenbaum, 1957) were given
for all semantic differential scale items in the questionnaire. Use of the dichotomous
and multiple choice questions was explained to the subjects by presenting relevant
examples, unrelated to the experimental topics.
RESULTS AND DISCUSSION

Before turning to the major hypotheses we shall discuss briefly the


effects of the different orders of variables in the questionnaire and some
of the differences between male and female subjects.
The questionnaires included six sets of questions, each containing eight
behaviors: A,, A-act, NB,, NB,, BI, and BZ,. In addition, there were 28
pairs of BI,, the behavioral intentions in the dichotomous choice situation.
The total number of variables in the questionnaire, therefore, was 76.
Two-way analyses of variance were computed for each of the 76
variables. The four arrangements were treated as a random factor while
the two sexes were treated as the second, fixed factor.

Arrangement of Variables in the Questionnaire


As may be expected by chance alone, two of the 76 differences between
the four arrangements were statistically significant at the .05 level (3 and
92 df). These results indicate that the order in which the variables ap-
PFlEDICI’ION OF BEHAVIORAL INTENTIONS 407

peared in the questionnaire did not significantly affect the subjects’


responses.

Some Sex Differences


The differences between males and females reflect their differential
interest in the eight activities. Most of the significant differences were
related to “playing a game of poker on a Friday night.” As might be ex-
pected, males had more favorable attitudes with respect to performing
this activity than had females. They also expressed more willingness to
engage in the activity on a Friday (BI). In six of the seven pairings in
which “playing a game of poker on a Friday night” appeared in BZ,,
males tended more toward this activity than did females (p < .05)
although in most cases even the males did prefer the alternative activity
( e.g., going to a movie or to a concert).
It is interesting to note that NB, and NB, were not rated significantly
different by males and females for any of the eight activities. That is,
male and female subjects tended to have similar expectations with respect
to their own behavior and tended to perceive their friends to hold similar
expectations as well, for all eight activities.

Prediction of a Single Behavioral Zntention (BZ j


As is done in most attitudinal research we may first investigate the
extent to which an individual’s behavioral intentions with respect to a
single act can be predicted from our independent variables.
Table 1 presents the correlations of each of the three predictors (A-act,
NB,, and NB,) with BI, their beta weights in the multiple regression

TABLE 1
CORRELATIONS OF BI WITH A-act, NB,, AND XB,
(N = 100)

A-act-BI NB,-BI NBS-BI


-
T P T B T B R

Going to a party .523 ,077 ,815 ,714 ,587 ,083 ,819


Visiting an exhibition .670 ,440 ,630 .275 .437 ,128 ,724
Watching a western on TV .567 ,255 ,662 .423 ,439 ,180 .709
Going to a concert ,665 ,303 ,713 ,376 ,598 .249 .787
Playing poker ,668 ,227 .767 .502 591 .158 .794
Going to a French movie ,640 ,190 ,782 ,649 ,499 .191 .794
Participating in a discussion ,669 ,252 .702 ,335 .678 .300 .779
Reading a mystery novel ,538 ,292 .543 ,268 ,513 .323 .684

Note.-All correlations are significant at Q < .Ol.


408 AJZEN AND FISHBEIN

equation, and the multiple correlation coefficients for prediction of BZ


from all three variables simultaneously.
The first thing to note in Table I are the high multiple correlation
coefficients in the last column of the table. These results provide strong
evidence for the power of Fishbein’s model to predict behavioral inten-
tions. Moreover, the reader will note that for most behaviors NB, carries
the greatest weight in predicting BZ. This finding supports the model’s
claim that normative beliefs are an important predictor in addition to
attitudes. Indeed, a comparison of columns one and seven in Table 1
indicates the increment gained by utilizing normative beliefs, as well as
attitudes, as factors influencing an individual’s behavioral intentions. All
the beta weights of the normative beliefs are statistically significant
( p < .05), indicating that the contributions made by NB, and by NB, in
predicting BZ are above chance levels. The beta weights of A-act are also
significant (p < .05) except for the first behavior, i.e., “going to a party
on a Friday night,” where the explained variance is attributable primarily
to normative beliefs.
As predicted, A, was found to be highly correlated with A-act, with
NB,, with NB,, and with BZ (Table 2).
These high correlations were expected since in the present study the
attitude object and the specific acts were very similar. However, as has
been pointed out by Fishbein (1967)) A, will not always be related to
the variables of the model, and previous studies (e.g., Ajzen, 1967) have
found negligible correlations, supporting this argument. Furthermore,

TABLE 2
CORRELATIONS OF A, 'IVITH A-act, NB,, NB,, AND BI
AND PARTIAL CORRELATIOXS BETWEEN A, AND BI
(N = 100‘)

A,,-BZ
with A-act,
NB,, and NB.
A,,-A-act A,-NB, A,NB, A,--BI constant

Going to a party .770 ,570 ,398 ,561 .172


Visiting an exhibition .827 .560 .352 ,659 ,247*
Watching a western on TV .872 ,657 .307 ,599 ,133
Going to a concert .827 .775 .579 ,750 ,227*
Playing poker ,911 ,726 .497 t643 - ,031
Going to a French movie ,875 ,725 .461 ,653 ,051
Participating in a discussion .851 ,588 ,561 .668 267**
Reading a mystery novel 742 ,542 ,358 ,441 - ,077

Note.-All product-moment correlat,ions are significant at (Y < .Ol.


* p < .05 (partial correlation).
** p < .Ol (partial correlation).
PREDICTION OF BEHAVIORAL INTENTIONS 409

as can be seen in Table 2, when A-act, NB,, and NB, are held constant
the correlations between A, and BI are greatly reduced. However, in the
present experiment it should be noted that because of the high A,,-A-act
correlations, essentially the same results are obtained if A, is held con-
stant. That is, when A,,, NB,, and NB, are partialled out, the A-act-BZ
correlations are also greatly attenuated. Thus, while the finding that the
A,BZ correlations are reduced when the model’s predictors are held
constant lends support to the hypothesis that the effect of external vari-
ables such as A, are mediated by these predictors, the data from the
present study cannot be taken as conclusive evidence.

Prediction of Behavioral Zntentions in a Dichotomous


Choice Situation (BZ,)
(a) A-act vs. A-act+ We shall consider next the prediction based on
decision theory that A-actd, the algebraic difference between the attitudes
toward two behaviors, is more highly correlated with BZz, the intentional
choice between the same two behaviors, than A-act toward any one of
the two acts alone. It will be recalled that 28 dichotomous choices (BZ,)
had to be predicted from the corresponding A-act’s and A-actis.
In Table 3 it can be seen that in all but two cases A-act, is a better
predictor of BZ, than either of the relevant two single A-act’s.
Table 3 presents the correlations of each single A-act with each of the
seven choice situations of which the behavior the A-act refers to is a
part.6 Thus A-act toward “playing poker” is correlated with seven choice
situations in which the subjects had to choose between “playing poker”
and one of the other seven activities. The last column in the Table pre-
sents the correlations between A-a& and BZ?.
Although only 13 of the A-ac&-BZ, correlations are significantly greater
(p < .05) than the highest corresponding A-act-BZ, correlation, 26 of
the 28 differences are in the predicted direction (p < .OOl, by sign test).
The two exceptions to the prediction are in choice situations 11 and 15
and both involve high correlations between A-act toward “going to a
party” and BZ,. “Going to a party” was rated extremely positively by most
subjects and therefore, in these two cases, it was itself a sufficiently good
estimate of the choice that would be taken between this activity and
some alternative.
In general, then, it can be concluded that the decision theory hypoth-
esis is strongly supported by the results. Table 3 presents some dramatic
increases in correlations when BZ, is predicted by the difference between

‘The signs of the correlations depend on the scoring of BL. The behavior to the
left was arbitrarily assigned to the positive side of the scale.
T$BLE 3
CORRELATIONS OF Ble WLTH A-act AND WITH A-m&
(N = 100)

A-act A-act A-& A-act A-act A-act A-act A-act A-a&


novel discussion movie poker concert western exhibition party

1. Movie-concert ,349 -.401 .587


2. Party-poker .241 .385 .436
3. Novel-discussion .223 - .606 .727
4. Western-concert - .496 .459 .653
5. Novel-exhibition .386 - .500 .676
6. Party-concert - .238 .447 .487
7. Exhibition-movie - ,347 ,365 ,638
8. Poker-concert .551 - .511 .691
9. Movie-novel -.119 .591 .622
10. Discussion-party .357 - ,312 .544
11. Party-western -.130 .578 .437
12. Novel-western .186 - .607 .653
13. Exhibition-concert -.175 .317 ,518
14. Discussion-concert .421 .294 .616
15. Party-novel - .019 ,470 .337
16. Poker-western .518 - ,214 .645
17. Poker-novel - .255 ,561 ,592
18. Movie-western ,534 - .457 ,714
19. Novel-concert .252 - ,542 ,584
20. Exhibition-party ,226 -.385 .434
21. Discussion-poker .628 -.504 ,738
22. Movie-party .309 - .453 ,494
23. Discussion-movie .470 --.509 ,741
24. Poker-exhibition .514 - ,632 ,747
25. Discussion-western .637 - .541 ,770
26. Poker-movie - .589 .404 .715
27. Discussion-exhibition .374 - .374 .611
28. Western-exhibition .502 - ,661 ,792

Note.-p.05 = .195, p.01 = .254.


PEUGDICTION OF BFJIAVIOBAL INTENTIONS 411

the attitudes toward each of the appropriate two behaviors as compared


with the correlations of each of the single A-act’s with BI,.
(b) NB, vs. NB,a and NB, vs. NB,d. Similar to the results with respect
to A-act, most algebraic differences between the two personal normative
beliefs correlate more highly with the appropriate BX, than do the two
single NB,‘s. There are four cases in which the prediction does not hold,
two of which again involve “going to a party.”
Even stronger results are found with respect to NB’s. Here all NB,d’s
are more highly correlated with BI, than are the two individual NB,‘s.
Lack of space prevents detailed presentation of the results comparing
NB, with NB,, and NB, with NB,d.7

Prediction of Behavioral Intentions in a Multiple Choice Situation ( BIs)


As will be recalled, the subjects were instructed to rank order the eight
alternative behaviors with respect to the relative likelihood of their en-
gaging in these behaviors on a Friday night.
The rank order of each of the eight behaviors ( B18) was correlated
with A-act, with NB,, and with NB, toward the appropriate behavior.
That is, the rank assigned by the subject to behavior 1 has been cor-
related with A-act toward behavior 1, with NB, toward behavior 1,
and with NB, toward behavior 1. The same procedure was followed
for all eight behaviors. In addition, correlations were computed between
B18 and the ranks of each of the corresponding A-act’s, NB,‘s, and NB,‘s
as assessed for each subject separately. For example, the rank order as-
signed by the subject to activity 1 has been correlated with the rank
of the A-act score for activity 1, ranked with respect to all eight activi-
ties of this subject (A-act,). And similarly for NB, and for NB,.
It is predicted that the rank orders (A-act,, NBpB, and NB,,) will yield
higher correlations with BIs than the original A-act, NB,, and NB, scores.
Table 4 presents results supporting this hypothesis. The predictions
based on all eight behaviors (on their rank) are always better than those
based on the appropriate original score, except again for the activity of
“going to a party on a Friday night.” These results hold for all three
predictors, A-act, NB,, and NB,.
In summary, taking into account the alternative acts open to an in-
dividual in a given situation, as suggested by decision theory, has by and
large improved the prediction of behavioral intentions in a dichotomous
choice situation as well as in a situation involving the choice between
eight different behaviors.

‘Interested readers can obtain these tables by writing to the senior author. The
correlations of N&a and of NB.a with BL can be inspected in Table 5 below.
412 AJZEN AND FISHBEIN

TABLE 4
CORRELATIONS BETWEEN BIs AND A-act, NB,, AND NB,
AND BETWEEN BIB AND A-a&, NB,g, AND NB,8
(N = 100)

B18 B18

BZs exhibi- B18 BI8 BZs B18 discus- BZa


party tion western concert poker movie sion novel

A-act ,539 ,616 ,615 ,603 ,607 ,567 ,615 ,429


NB, ,698 ,623 ,694 ,670 ,739 ,657 ,589 ,418
NB, ,623 ,507 .455 ,603 ,630 ,496 ,664 ,372
A-ads ,420 ,649 ,674 .720 ,753 ,640 ,765 ,648
N&s ,579 ,713 ,788 ,752 ,845 ,775 ,664 ,555
NBsa .621 622 .541 .713 ,758 ,624 ,675 520

Note.-All correlations are significant at (Y < .Ol.

Predictions of BI, and BI, by Multiple Regression (Fishbein’s Mod&)


It will be recalled that decision theory is concerned only with what
corresponds to A-act in Fishbein’s ( 1967) model. That is, decision theory
defines the utility of an act as the sum of the probabilities of the possible
outcomes of that act multipled by their respective values for the individ-
ual. Essentially the same definition has been proposed by Fishbein for
the attitude toward an act. In addition, Fishbein’s model asserts that
normative beliefs, as well as attitudes, toward the act have to be taken
into consideration for the prediction of behavioral intentions and corre-
sponding overt behavior. In the results presented above we have ex-
tended the decision theory notion in relation to choice situations to the
two additional variables which, according to Fishbein’s model, enter into
the determination of BI, viz. NB, and NB,.
We may now ask whether the addition of NBpd and NBSa and of NBDs
and NB,,, as proposed by Fishbein’s model, will indeed increase the
correlations with BI, and BIB, respectively. Table 5 presents the results
for the prediction of B12 and Table 6 those for BI,. The correlations in
columns 1, 3, and 5 of both tables have already been discussed before
and are here presented merely to facilitate interpretation of the results.
In these three columns the correlations of each of the three predictors
with the criterion are given. In the columns to the right of these correla-
tions, each predictor’s respective weight in the multiple regression equa-
tion is given. The last column presents the multiple correlation coeffi-
cients of A-act,], NBDd, and NBSd on BI, in Table 5 and of A-act,, NB,s,
and NB,, on BI, in Table 6.
The following conclusions can be drawn from Tables 5 and 6:
(1) It is quite obvious that the three independent variables A-act, NB,,
PREDICTION OF BEHAVIORAL INTENTIONS 413

TABLE 5
CORRELATIONS AND BETA- WEIGHTS FOR PREDICTION OF
Blz FROM A-a&, A’B,a, AND N&d
(N = 100)

1. Movie-concert .587 ,129 ,677 ,339 .677 ,374 ,747


2. Party-poker ,436 - ,029 ,624 .522 ,528 .172 .635
3. Novel-discussion ,727 .378 ,714 ,341 ,621 ,195 ,800
4. Western-concert ,653 ,293 ,609 ,237 ,626 ,307 .727
5. Novel-exhibition ,676 ,381 ,619 ,189 ,622 ,300 ,749
6. Party-concert ,487 ,075 ,616 ,326 ,629 ,410 .704
7. Exhibition-movie ,638 .323 ,622 ,227 .598 .288 721
8. Poker-concert ,691 ,202 ,716 .267 ,751 ,451 ,818
9. Movie-novel ,622 ,332 .563 ,173 ,614 .356 .773
10. Discussion-part,y ,544 ,144 ,640 ,292 ,668 ,400 ,730
11. Party-western .437 .099 .562 .371 ,475 .200 t589
12. Novel-western ,653 ,386 .596 ,304 ,447 .209 ,726
13. Exhibition-concert ,518 .277 ,518 ,250 ,482 ,212 .612
14. Discussion-concert ,616 .219 ,643 ,230 ,704 ,437 769
15. Party-novel .337 ,167 ,627 ,496 ,501 ,231 .6d7
16. Poker-western ,645 .170 ,749 ,504 ,601 .216 .786
17. Poker-novel ,592 .167 .664 .391 ,595 .327 .743
18. Movie-western ,714 ,320 ,709 .325 ,649 ,257 ,791
19. Novel-concert .584 ,259 ,586 ,269 ,576 ,306 .696
20. Exhibition-party .434 ,129 ,548 ,218 ,617 .452 ,674
21. Discussion-poker ,738 .318 ,743 .339 ,701 ,261 ,818
22. Movie-party .494 ,071 ,632 .413 ,560 ,293 ,678
23. Discussion-movie ,741 ,433 ,689 ,209 ,681 .245 ,795
24. Poker-exhibition .747 ,285 ,744 .338 ,686 .297 ,817
25. Discussion-western .770 ,446 ,701 .228 ,685 ,262 ,826
26. Poker-movie ,715 ,221 ,775 ,442 .680 .217 ,805
27. Discussion-exhibition ,611 ,151 ,667 .330 ,704 .429 .784
28. Western-exhibition .792 .559 .687 ,183 .583 .159 ,813

Note.-All correlations are significant at 01 < .Ol.

and NB, are relevant for, and are highly correlated with, the criterion of
behavioral intentions in each of the two different choice situations. That is,
A-a&, NBpd, and NBsd are each highIy correIated with B12, and A-act,,
NBp8, and NBS8 are each highly correlated with BI,. These findings point
to the importance of normative beliefs for the prediction of behavorial
intentions in different choice situations. Attitudes toward the alternative
acts-the only variables considered by decision theory-provide only part
of the information needed for accurate prediction.
(2) The relative importance of each predictor varies with the par-
414 AJZENANDFISHBFZN

TABLE 6
'CORRELATIONS OF BIs WITH A-a&, NB,,, AND NBS8
(N = 100)

A-acts-BIs NB,g-BIs h’B,s-BIs


-
r P r P r P R

Going to a party .420 .074 ,579 ,299 ,621 .434 ,688


Visiting an exhibition ,649 .209 ,713 .434 622 ,276 ,785
Watching a western on TV ,674 ,160 ,788 .603 541 .135 ,810
Going to a concert ,720 .318 ,752 ,350 713 .317 ,847
Playing poker ,753 .171 .s45 .530 758 ,287 ,891
Going to a French movie ,640 ,135 ,775 ,606 624 ,101 .787
Participating in a discussion .765 .487 ,664 ,236 654 ,213 ,818
Reading a mystery novel ,648 .373 ,555 ,287 520 ,239 725

Note.-All correlations are significant, at 01 < .Ol.

titular choice that has to be made. With respect to the particular be-
haviors considered in the present study, the personal normative beliefs
tend to carry more weight in the prediction of behavioral intentions than
either the attitudes toward the acts or the normative beliefs of friends.
(3) The multiple correlations of the three predictors with the criteria
are high and provide strong evidence for the usefulness of Fishbein’s
theoretical model and its predictive power.

CONCLUSIONS
The results of the present study draw attention to a number of impor-
tant issues in attitude research, First, the results indicate that behavorial
intentions to perform specific behavioral acts can best be predicted by
considering the attitudes as well as normative beliefs toward these acts.
By combining these variables in a multiple regression equation highly
accurate predictions of specific behavioral intentions can be made. Pre-
vious research (footnote 2) has indicated that these types of behavioral
intentions are very highly correlated with corresponding behaviors.
Second, it was found that, as suggested by decision theory, alternative
behaviors open to the individual have to be considered in predicting be-
havioral intentions and thus in predicting behavior. By taking into ac-
count A-act toward alternative behaviors, better predictions of the sub-
ject’s choice behavior is obtained. Further, in these choice situations, it
was again found that normative beliefs account for much of the variance,
indicating the superiority of Fishbein’s model over the relatively simple
decision theory model. Extending the decision theory notion to normative
beliefs and their incl&n in the predictive equation greatly improved the
PREDICTION OF BEHAVIORAL INTENTIONS 415

correlations with behavioral intentions in dichotomous and multipIe


choice situations.
As encouraging as these results appear, a few words of caution arc
necessary. As one reviewer pointed out, “there may be strong demand
characteristics in operation in this study. No matter which of the four
orders of questionnaire presentation was used the force upon the subject
would be to make the choices consistent with the earlier ones he had
made.” While this concern may be somewhat vahd in the context of the
present experiment, it should be noted that, as the reviewer himself
pointed out, the argument would lose much of its impact if a “true” be-
havioral criterion had been included in the study. While this was not
done in the present study, other tests of the model have included a be-
havioral criterion and, as was mentioned above (footnote 2)) these stud&
have demonstrated that extremely high BZ-B correlations can be, and are,
obtained when appropriate BZ’s are selected. Further, while a complete
discussion of this problem is beyond the scope of the present paper, both
Dulany ( 1967) and Carlson (1968) have demonstrated that a “demand
for consistency” explanation cannot account for the obtained relationships
between the model’s predictors. Among other things, such an explanation
does not allow one to account for the finding that the model’s predictors
take on different relative weights for different behavioral intentions and
different choice situations. Thus, although there is no way to clearly
demonstrate the absence of demand characteristics in the present experi-
ment, several factors do seem to indicate that this problem is not as
serious as it may initially appear.
Finally, and somewhat related to the above, it must be recalled that
Fishbein’s model, like Dulany’s original formulation, is a model for the
prediction of behavioral intentions and not a model for the prediction of
behavior per se. Although the model can, and in fact does, lead to ex-
tremely accurate behavioral predictions, it will only do so when, and if,
the experimenter selects an appropriate behavior intention. While some
of the factors that influence the size of the BZ-B correlation have been
discussed above, additional research on this problem is clearly necessary.

REFERENCES
AJZEN, I. Attitudes, normative beliefs, and the prediction of behavior: An empirical
investigation. Unpublished master’s thesis, University of Illinois, 1967.
BLACKWELL, D., AND GIRSHIK, M. A. Theory oj games and statistical decisions. New
York: Wiley, 1954.
CARLSON, A. R. The relationships between a behavioral intention, attitude toward the
behavior and normative beliefs about the behavior. Unpublished doctoral dis-
sertation, University of Illinois, 1968.
416 AJZEN AND E-ISHBEIN

COOK, S. W., AND SELLTIZ, C. A multiple-indicator approach to attitude measurement.


Psychological Bulletin, 1964, 62,3655.
DULANY, D. E. Hypotheses and habits in verbal “operant conditioning.” jountal of
Abnormal and Social Psychology, 1961,63,251-263.
DULANY, D. E. The separable effects of the information conveyed by a reinforcer.
Paper read at the Psychonomic Society meetings, 1964.
bLANY, D. E. Awareness, rules, and propositional control: A confrontation with
S-R behavior theory. In D. Horton and T. Dixon (Eds. ), VerbaE behavior and
S-R behavior theory. New Jersey: Prentice-Hall, 1967.
EDWARDS, W. The theory of decision making. Psychological Bulletin, 1954, 51, 380-
418.
FISHBEIN, M. An investigation of the relationships between beliefs about an object
and the attitude toward that object. Human Relations, 1963, 16, 233-240.
FISHBEIN, M. Sexual behavior and propositional control. Paper read at the Psycho-
nomic Society meetings, 1966.
FISHBEIN, M. Attitudes and the prediction of behavior. In M. Fishbein (Ed.), Read-
ings in attitude theory and measurement. New York: Wiley, 1967.
FISHBEIN, M., AND RAVEN, B. H. The AB scales: An operational definition of belief
and attitude. Human Relations, 1962, 15, 3544.
LAPIERE, R. T. Attitudes vs. action. Social Forces, 1934, 13, 230-237.
OSGOOD, C. E., SUCI, G. J., AND TANNENBAUM, P. H. The measurement of meaning.
Urbana: University of Illinois Press, 1957.
PEAK, H. Attitude and motivation. In M. Jones (Ed.), Nebraska sympo.sium on
motiuation. Lincoln: University of Nebraska Press, 1955.
ROSENBERG, M. J, Cognitive structure and attitudinal affect. Journal of Abnormal and
Social Psychology, 1956, 53, 367372.
ROSENBERG, M. J, Inconsistency arousal and reduction in attitude change. In I. D.
Steiner and M. Fishbein (Eds. ), Current studies in social psychology. New York:
Holt, 1965.
ROTTER, J. B. Social learning and clinical psychology. Englewood Cliffs, New Jersey:
Prentice-Hall, 1954.
TRIANDIS, H. C. Exploratory factor analysis of the behavioral component of social
attitudes. Journal of Abnormal and Social Psychology, 1964, 68, 420-430.
TRIANDIS, H. C. Toward an analysis of the components of interpersonal attitudes. In
C. W. Sherif and M. Sherif (Eds. ), Attitudes, ego-inuolvement, and change.
New York: Wiley, 1967.
ZAJONC, R. B. Structure of the cognitive field. Unpublished doctoral dissertation,
University of Michigan, Ann Arbor, 1954.

( Received June 24, 1968)

You might also like