The Risk-as-feelings hypothesis in a Theory-of-planned-behaviour perspective

Judgment and Decision Making, Vol. 4, No. 7, December 2009, pp. 567–586
The Risk-as-feelings hypothesis in a Theory-of-planned-behaviour
perspective
Therese Kobbeltvedt? and Katharina Wolff
Norwegian School of Economics and Business Administration and University of Bergen, Norway
Abstract
The Theory of Planned Behaviour (TpB: Ajzen, 1985; 1991) is based on a utility framework, and the Risk-as-Feelings
hypothesis (RaF: Loewenstein, Weber, Hsee, & Welch, 2001) is a feelings-based behavioural model. The TpB and RaF
are ?rst compared and contrasted. Two empirical studies investigated the predictive power of consequence-based vs.
affect-based evaluative judgements for behavioural intentions: Study 1 (n = 94) applied a regression model to examine
the predictive value of a subset of shared variables, unique TpB variables, and unique RaF variables for intentions to have
unsafe sex. Study 2 (n = 357) experimentally examined whether intentions are driven by consequences or feelings, in two
decision vignettes with opposite qualities: A positive hedonic experience with potential negative consequences (unsafe
sex) vs. a negative hedonic experience with potential positive consequences (back surgery). The results supported the
TpB by emphasising the role of outcome-expectations in the construction of intentions, and the RaF by showing the
importance of affective subcomponents in attitudes.
Keywords: risk, feelings, affect, planned behaviour, decision, consequences, ambivalence.
1 Introduction
suggested underlying mechanism for the tension between
behaviour and consequences is the fundamental differ-
Humans sometimes act in order to reach an outcome,
ence in time perspective associated with ongoing activ-
and other times out of enjoyment for the activity itself.
ities and their consequences. It has been suggested that
This difference has been referred to as extrinsic vs. intrin-
the need for self-regulation exists in most individuals due
sic motivation (Higgins & Bryant, 1982), incentives vs.
to an inconsistency between short- and long-term conse-
motivation (De Grada, Kruglanski, Mannetti, & Pierro,
quences of an option (Soman et al., 2005). Intertemporal
1999), and utilitarian vs. hedonically driven behaviour
discounting exempli?es one such failure to balance the
(Dhar, 2000), and represents a general dichotomy be-
subjective value of immediate to more distant goals, and a
tween action as goal and action as mean towards a goal.
systematic devaluation of future as opposed to immediate
There is often a con?ict between the hedonic quality
outcomes (Camerer, Loewenstein, & Weber, 1989). The
of the behaviour and its consequences. Although hedo-
present paper focuses on dilemmas where the value(s) as-
nic qualities could be seen as a particular kind of be-
sociated with the performance of the behaviour and the
havioural consequence, the experiential part of such a
value(s) of the consequences are in con?ict, even when
(dis)pleasure is, by de?nition, in the present, whereas
holding time-perspective constant.
other behavioural outputs lie in the future. Thus, one
Several theoretical accounts may shed light on the dif-
ference between actions as goals or as means. Meta-
?This research was ?nanced by a grant from the Department of
theories of human motivation, such as self-determination
Psychosocial Science, University of Bergen (Grunnforskningsmidler).
theory (Baard, Deci, & Ryan, 2004; Deci, Koestner, &
We thank Ann Elin Jønsson Orland and Anne Hoyer Sørensen for
research assistance. Thanks are due to Britt-Marie Drottz-Sjöberg,
Ryan, 1999; Ryan & Deci, 2000; Ryan, Huta, & Deci,
George Loewenstein, and Icek Ajzen for comments that were of help
2008) centre on motivational mechanisms, claiming that a
and inspiration to this research. Thanks to Marius Barstad for in-
set of basic motivational needs drive behaviour. The need
put to the hypothetical scenarios. Finally, thanks to our reviewers for
for competence, autonomy, and belongingness are pro-
their insightful comments to a previous version of the manuscript. Ad-
dresss: Therese Kobbeltvedt, The Norwegian School of Economics
posed as general behavioural driving forces, fairly inde-
and Business Administration (NHH), Department of Strategy and Man-
pendently of time-perspectives. Thus, in addition to mo-
agement and Research Centre for Health Promotion (HEMIL), Fac-
tivation towards mastery and independence, social norms
ulty of Psychology, University of Bergen, Norway; Katharina Wolff,
direct our behaviour. The most socially accepted action
Department of Psychosocial Science, Faculty of Psychology, Uni-
versity of Bergen, Norway. Email: Therese.Kobbeltvedt@nhh.no or
is the default option, and complying with it may not be
Therese.Kobbeltvedt@psych.uib.no.
experienced as decision making.
567

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Risk-as-feelings and Theory-of-planned-behavior
568
Other theories focus on preventing factors, such as
such as e.g. exercise behaviour (Lowe, Eves, & Carroll,
cognitive limitations. To this end, several dual-process
2002), condom use (Richard, de Vries, & van der Pligt,
theories (Chaiken & Trope, 1999) demonstrate our ability
1998), and smoking cessation (Droomers, Schrijbers, &
and unique human predisposition to pay attention to po-
Mackenbach, 2004).
tentially contradictory sources of information, e.g., affec-
A core assumption of the TpB is the idea that planned
tive/experiential vs. reason based/consequential. These
behaviour is driven by behavioural intentions (Ajzen,
authors (e.g., Chen & Chaiken, 1999; Epstein & Pacini,
2001). The TpB speci?cally suggests which subcompo-
1999) circumvent the issue of time perspective indirectly,
nents determine intentions; i.e., the individual’s attitudes
by showing how the intuitive, basic, and effortless expe-
towards the behaviour, subjective norms, and perceived
riential/heuristic system for processing is faster than the
behavioural control. Attitudes are de?ned as the indi-
intentional, reason-based, and effortful rational/analytical
vidual’s evaluation of whether performing the behaviour
system. The rapid processing of affective and immediate
is either positive or negative, thus a valence-based sub-
information may result in behaviour which is more in-
jective evaluation. Subjective norms represent perceived
clined to follow the affective rather than the analytical
social pressure to perform the behaviour, indicating the
system.
appreciation of the given behaviour within a social con-
There is ample evidence that intentions may be di-
text. Hence, this variable explicitly acknowledges that
rected at activities just for the pleasure of it, with little
our need to relate to others is shaping behaviour (Baard
consideration of the potential consequences, e.g., smok-
et al., 2004; Deci et al., 1999; Ryan & Deci, 2000; Ryan
ing, heavy drinking, unsafe sex, hazardous driving, and
et al., 2008), and that preferences are socially context-
unhealthy eating. On the basis of this tension between
dependent. Perceived behavioural control refers to the
consequences and feelings, the present study set out
individual’s perception of control over performing the be-
to compare and contrast two theoretical psychological
haviour. This re?ects the idea that an ability to regulate
frameworks, which identify key structures and processes
or control one self is at the core of motivational processes
of planned- and feelings-based behaviour; the Theory of
in general, and is in line with several theoretical perspec-
planned behaviour (TpB: Ajzen, 1985; 1991; 2001) and
tives on self-regulation (Baumeister & Vohs, 2004).
the Risk-as-feelings hypothesis (RaF: Loewenstein et al.,
Furthermore, the TpB speci?es the antecedents of atti-
2001). Furthermore, we aimed at an empirical investiga-
tudes, norms, and perceived control. In the case of atti-
tion of core predictors for behavioural intentions in de-
tudes, these are a small set of speci?c salient behavioural
cision making under ambivalence. Decision ambivalence
beliefs. The behavioural beliefs are formed by the respec-
is here de?ned as perceived tension between the hedonic
tive likelihood of anticipated outcomes of the behaviour,
quality of the behaviour and its potential consequences in
weighted by an evaluation of each of these outcome, thus
behavioural choices.
a utility based construct. The expected outcomes may in-
The present paper is organized as the following: The
clude the anticipation of positive and negative emotions,
two theoretical frameworks are presented ?rst, followed
thus, anticipated emotions (prospects about feelings) are
by a theoretical analysis which separates their basic com-
accounted for within this consequentialist model (Ajzen,
ponents into shared and unique variables. Then, we
1991; Ajzen & Driver, 1991).
present the outline and results of two empirical studies,
investigating the predictive power of the models. Next,
we discuss the ?ndings in light of the theoretical frame-
1.2 Risk as feelings
works and the reviewed literature. Finally, we present an
Few behavioural models explicitly outline the be-
example of a research model, meant to capture variables
havioural output resulting from ambivalence due to con-
from several of the dominant research ?elds relevant to
?icting information from the two systems for information
decision making under ambivalence.
acquisition, but The RaF hypothesis (Loewenstein et al.,
2001) is an exception. The RaF perspective has been used
1.1 The Theory of planned behaviour
to feed into models that predict action selection in psy-
chological risk-return models (Weber & Johnson, 2008;
The TpB is among the most widely used models for pre-
Weber & Milliman, 1997). Apart from showing the high
dicting speci?c human behaviour (Ajzen, 1985; 1991).
potential for disagreement between feelings and cogni-
It is based on a utility framework, and de?ned as conse-
tive evaluations, the RaF model suggests that, when such
quentialist, according to the terminology of Loewenstein
a tension arises, behaviour tends to be driven by anticipa-
and colleagues (2001). The TpB has been applied for pre-
tory feelings, e.g., feelings experienced at the moment of
dicting speci?c behaviours, which may involve con?icts
decision making. By integrating outcome-related factors,
between short- and long term goals, affect and cognition,
such as anticipated outcomes, including anticipated emo-
or hedonism and consequences, within various domains
tions, the model incorporates several of the variables typ-

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Risk-as-feelings and Theory-of-planned-behavior
569
ically accounted for by the intentional/analytical system,
The TpB treats lack of correspondence between in-
and incorporates empirical evidence showing that the af-
tentions and behaviour as either a measurement prob-
fective/intuitive system may overrule the cognitive evalu-
lem, a question about the stability of an intention, or
ations when these are in con?ict.
a behavioural-control issue (Ajzen, 1991). The mea-
The RaF (Loewenstein et al., 2001) framework readily
surement problem is best handled by eliciting behaviour-
explains a range of behaviours which demonstrate diver-
speci?c intentions, whereas the stability-of-intention
gence between cognitive evaluations and feelings, e.g.,
problem will be lower the closer the time gap in-between
failure to act in accordance with one’s values, to com-
the reported intention and the actual behavioural out-
ply with one’s intentions and goals, or seemingly irra-
put (Ajzen, 1991; Isen, 2005). Behavioural dilemmas
tional behaviour, such as speci?c phobias and various
where there is cross-pressure between hedonic feelings
forms of affect-driven activities ranging from interper-
and behavioural consequences exemplify one type of
sonal relationships to appetitive/aversive motives in gen-
behavioural-control issue, which may lead to low cor-
eral. Loewenstein and colleagues (2001) argue that the
respondence between intentions and behaviour. Cross-
RaF perspective is feelings-based, in opposition to virtu-
pressure between social norms and attitudes is another
ally all other models aimed at describing and predicting
type of dilemma challenging behavioural control within
human behaviour, as these are consequentialist in nature
the TpB perspective. The RaF perspective depicts a
(see Figures 1–3 in Loewenstein et al., 2001).1
slightly different tension, the one lying inherently in peo-
ple’s (implicit) judgements of perceived riskiness. The
behavioural effects may nevertheless be similar, with low
1.3 Consequentialist vs. feelings-based be-
behavioural predictability as a result. We will devote our
havioural models
attention to such behavioural tensions in the empirical
section of the present paper.
From the theoretical perspective, the tension between act-
ing in order to reach a certain outcome, in contrast to
just acting out of enjoyment of the activity, has been
1.4 TpB and RaF: Shared variables
approached as a fundamental distinction between inten-
Despite the initial presentation of these models as contra-
tional behaviour and incidental- or feelings-based be-
dictory, we argue that they have certain predictors and in-
haviour (Dhar, 2000). This distinction is in accordance
terrelations between predictors in common. Based on the
with the distinction applied between consequence-based
explicit verbal and graphical presentations of subcompo-
vs. feelings-based behaviour throughout the present pa-
nents put forward in the TpB (Ajzen, 1985; 1991) and the
per. To reach a better understanding of such phenomena,
RaF (Loewenstein et al., 2001) respectively, the variables
by scrutinizing the feelings-based and the consequential-
have been divided into three categories (see Figure 1):
ist perspective on human behaviour, we now turn to a
The orange boxes indicate variables which are common
more direct comparison of the TpB (Ajzen, 1985; 1991)
for both models. The green boxes re?ect the variables
and the RaF (Loewenstein et al., 2001).
unique for the TpB, and the blue boxes are the unique
The TpB (Ajzen, 1985; 1991) and the RaF (Loewen-
RaF variables. We shall ?rst focus on the shared vari-
stein et al., 2001) are both aimed at identifying the struc-
ables (Figure 1, orange boxes).
tural antecedents of human behaviour, which may be used
The TpB and RaF share the general assumption that
for predicting future behaviour. According to the TpB,
behaviour will be guided by an evaluation of the be-
the direct antecedents of behaviour are intentions, which
haviour. The TpB has been validated extensively across
in turn are determined by attitudes, social norms, and per-
a range of settings, and is speci?c about the formative
ceived behavioural control. More recently, Loewenstein
indices of these global evaluations: The determinants of
and colleagues (2001) highlighted how feelings play a
the global attitudinal evaluations are behavioural beliefs,
large role in determining behaviour. According to their
along with their pertaining subjective probabilities and
model, both consequence-related factors, such as antici-
outcome evaluations. For the more recent RaF hypoth-
pated outcomes and subjective probabilities and feelings-
esis, the factors anticipated outcomes, including antic-
based factors such as vividness of the consequences and
ipated emotions, are thought to jointly affect cognitive
background mood, give rise to cognitive evaluations and
evaluations, and feelings, together with the respective
feelings. The RaF hypothesis focuses on how cogni-
subjective probabilities.
tive evaluations and emotional feelings may diverge, and
Despite the different degrees of speci?city in determin-
claims that feelings frequently drive behaviour.
ing the structural components of the attitudes, the TpB
and RaF are based on the same general assumption, that
1Loewenstein and colleagues (2001) refer to the RaF approach as a
hypothesis/perspective rather than a model, but present it graphically as
evaluations of a speci?c behaviour will be guided by an-
a model.
ticipated outcomes in combination with subjective prob-

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Risk-as-feelings and Theory-of-planned-behavior
570
Figure 1: TpB and RaF variables. Colour codes: Orange = shared variables, Green = Unique TpB variables, Blue =
Unique RaF variables.
abilities. The TpB determines the weighting function of
cial pressure to perform a certain behaviour and are de-
the various components of the attitude, as the product
termined by a set of normative beliefs, i.e., the individu-
of the subjective probability and outcome evaluation per-
als beliefs about the perceived wishes of signi?cant oth-
taining to the behaviour. The RaF hypothesis highlights
ers, weighted by the individual’s willingness to comply
the dynamic interplay of anticipated outcomes and sub-
with these perceived expectations. Perceived control over
jective probabilities.
the behaviour is in turn determined by a respondent’s be-
According to the RaF model, anticipated outcomes
liefs about a set of salient control factors, i.e., beliefs
may in?uence cognitive evaluations and feelings sepa-
about the presence or absence of resources and obsta-
rately. This is also the case for subjective probability.
cles which may facilitate or impede the performance of
The dismantling of cognitive evaluations and feelings,
the behaviour, weighted by the respondent’s perceived
and their unique inputs from anticipated outcomes and
control over each of these factors. The greater the per-
subjective probabilities, allows for more nuances than a
ceived social pressure and the greater the perceived con-
mere valence-based global evaluation of the behaviour.
trol, the more likely will the individual perform the be-
Furthermore, the RaF depicts how cognitive evaluations
haviour (Ajzen, 1985; 1991). The social aspects of be-
and feelings are mutually in?uenced by each other, again
havioural control may relate to moral-based or feelings
emphasising a dynamic interplay. The TpB does not de-
based evaluations pertaining to the behaviour. However,
?ne the nature of outcome evaluations with respect to be-
low behavioural control could also be related to the mere
ing more or less affective or cognitive in origin. How-
complexity of the behaviour at hand, given that certain
ever, Ajzen (1991) holds the TpB’s behavioural beliefs
behaviours are simply more dif?cult to carry out than oth-
to include outcome expectations of both predominantly
ers. Thus, lack of behavioural control in this model, could
cognitive and affective qualities (Ajzen, 1991; Ajzen &
be emotional or non-emotional. This is in contrast to the
Driver, 1991).
more narrow focus of the RaF perspective, which centres
on the particular interplay of feelings and cognitive eval-
uations, and the resulting behavioural ambivalence.
1.4.1 TpB unique variables
There are also predictors unique to each model: Com-
1.4.2 RaF unique variables
ponents speci?c to the TpB (Figure1, green boxes) are
subjective norms and perceived behavioural control. The
The components speci?c to the RaF (Figure 1, blue
subjective norms constitute the individual’s perceived so-
boxes) are feelings, speci?cally anticipatory feelings,

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Risk-as-feelings and Theory-of-planned-behavior
571
which are experienced at the moment of decision making.
applied to predict a range of behaviours which may be
Another broad unique RaF component is labelled “other
unplanned, e.g., condom use (Richard et al., 1998) and
factors”. These include different types of contextual fac-
smoking (Droomers et al., 2004). The target behaviour,
tors which will increase any feelings-based behavioural
unsafe sex, is therefore in accordance with the traditional
impact. Some of the examples provided by Loewen-
use of TpB.
stein and colleagues (2001) are vividness, e.g., how vivid
Based on the idea that knowledge follows historical
one may imagine the potential consequences of the be-
lines, and the argument that new models and perspectives
haviour, a factor found to correlate with affect and bod-
need to be tested against existing paradigms, a stepwise
ily arousal (Adolphs & Damasio, 2001), and risk judge-
approach was chosen in Study 1. The regression model
ments (Slovic, 1987). Furthermore, the immediacy of
investigates whether the unique TpB variables increase
the behavioural consequences and background mood are
the predictability of behavioural intentions, above that ac-
speci?cally taken into account in the RaF. A ?nal unique
counted for by the shared variables, thus yielding support
element of the RaF, contrary to the TpB, is the inclusion
to the consequentialist perspective. If inclusion of the
of the possibility that cognitive evaluations and anticipa-
unique RaF variables, exceeds the predictability of the
tory feelings have independent effects on behaviour, with
behavioural intentions beyond that provided by the TpB
the resulting potential for divergence in forces stemming
and the shared variables, this would support the feelings-
from the cognitive evaluations and feelings, respectively.
based perspective. By entering the shared variables as a
This divergence is much in accordance with the litera-
separate unit into the regression, the most parsimonious
ture on self-regulation in general (Baumeister & Vohs,
version in terms of number of predictors, of the model-
2004), however, the particularly new notion of the RaF
overlap could be tested separately, and yield insight into
hypothesis, is the assumption that, when there is a con-
whether the overlapping variables are the strongest pre-
?ict between cognitive evaluations and feelings, feelings
dictors.
will drive behaviour (Loewenstein et al., 2001). An in-
depth understanding of the mechanisms underlying lack
of behavioural control within individuals, and low be-
2.1 Method
havioural predictability in general, may require further
2.1.1 Sample and procedure
sub-divisions of types of behavioural control and a more
explicit operationalisation.
Data were collected on campus during a lecture break, by
requesting the subjects to ?ll inn a questionnaire about
decision dilemmas from everyday life. In total 94 under-
2 Study 1: The predictive power of graduate students of psychology participated: 22 males
shared, unique TpB, and unique and 72 females, with mean age of 20.6 years.
RaF variables
2.1.2 Material
Following the theoretical analysis unpacking the shared
The questionnaire presented a vignette about unsafe sex:
and unique behavioural predictors from the TpB and RaF
“Imagine you are having a night out and you meet some-
model, Study 1 compares the predictive power of sub-
one you have noticed before. You are happy when this
sets of shared and unique variables of the two models
person shows an interest in you, and you connect immedi-
(Figure 1). A vignette about “unsafe sex” was the target
ately. The rest of the evening the two of you stay together.
behaviour, and formed the basis for elicitations of inten-
You realise that you like this person really well. When the
tions, the TpB, and the RaF variables. There were several
clubs and bars are closing, you do not want to split up, so
reasons for choosing unsafe sex as the target behaviour:
you go home to your place. You start to get intimate, and
First, sexual self-regulation is among the most basic ar-
you are aroused. You notice how good it feels. But, you
eas for self-regulation (Baumeister & Vohs, 2004), and
do not have a condom. You are thinking about the day
represents a situation with con?icting valence of hedo-
after, and how uncomfortable you will feel.”
nistic and consequential inputs. Moreover, this dilemma
The vignette was followed by an elicitation of self-
elicits decision processes which are affect-rich enough
report measures of the variables extracted from the two
to be relevant for the RaF model. It is also likely to be
theoretical models TpB and RaF. To prevent any order
based on normative beliefs, such as norm-pressure in a
effect on the intention to have unsafe sex, the various in-
social-contextual setting. Unsafe sex is hardly the pro-
dependent variables were presented before and after the
totype of a planned behaviour,2 but the TpB has been
elicitation of the intention, in two different versions of the
2It is possible that under certain conditions unsafe sex is planned and
questionnaire. These were randomly distributed among
intended.
the subjects.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Risk-as-feelings and Theory-of-planned-behavior
572
2.1.3 Measures
Unique TpB variables
Perceived behavioural control was measured with four
The shared and unique components from the TpB and
items: “How dif?cult is it not to have sex in this situ-
RaF models (Figure 1) were operationalised and adapted
ation?” rated from 1very easy to 7very dif?cult. “It is up to
to the behaviour at hand; unsafe sex. The TpB variables
me if I have sex or not” was rated from 1I agree completely to
were constructed according to the recommendations pro-
7I completely disagree. “If I wanted to refrain from having sex,
vided by Ajzen (2002a).
I would be able to do so” was rated from 1I agree completely to
7I completely disagree. Finally “How much control would you
Shared variables
have in this situation?” rated from 1no control to 7full control
Intention. The dependent variable, the intention to
was reversed before added to the sumscore; Cronbach’s
have unsafe sex, was measured with two items: “Do you
alpha = .67 (M = 2.49, SD = 1.10).
consider having unsafe sex with him/her?” 1certainly not to
Social norms (family and friends) were measured with
7yes, de?nitely and “How probable is it that the two of you
four items: “Do you believe that your family approves of
will have sex?” 1very improbable to 7highly probable. A sum-
you having sex in this situation?” followed by “Would
score yielded Cronbach’s Alpha = .84 (M = 2.67, SD =
you take their opinion into account?” and “Do you think
1.78).
your friends approve of you having sex in this situation?”
Attitudes were measured with 10 bipolar eval-
followed by “Would you take their opinion into account?”
uations capturing both affectively and cognitively
All items were indicated from 1yes, de?nitely to 7not at all. The
based evaluations pertaining to unsafe sex:
“Hav-
two items measuring norms were multiplied separately
ing unsafe sex” is “boring/exciting, turn-off/turn-on,
for family and friends (MFamily = 4.52, SD = 3.87; MFriends
uncomfortable/comfortable, stressing/relaxing, danger-
= 9.42, SD = 6.91).
ous/safe, unwise/wise, unnecessary/necessary, irrespon-
Unique RaF variables
sible/responsible, health damaging/health promoting,
Vividness was measured with “How vividly can you
bad/good”. All the bipolar evaluations were anchored in
imagine, after having had unsafe sex” each of the same
1 and 7.
8 potential consequences as measured in behavioural be-
The 10 items were submitted to a principal com-
liefs; “pregnancy, catch HIV, catch other sexually trans-
ponent analysis (varimax rotatation), and yielded two
mitted diseases, regretting, positive self image, feel-
factors, which explained altogether 56 % of the vari-
ing more attractive, getting a new boyfriend/girlfriend,
ance in attitudes. The ?rst factor “Cognitive attitudes”
feeling proud” indicated on scales from 1very vividly to
(31 % explained variance) included the items “irrespon-
7very diffuse. Cronbach’s Alpha reached .67 (M = 4.80, SD
sible/responsible, dangerous/safe, unwise/wise, stress-
= 1.03).
ing/relaxing, health damaging/health promoting, unnec-
Anticipatory emotions were measured with “Thinking
essary/necessary”. Cronbach’s Alpha of the sumscore
about having unsafe sex makes/gives me:” with 7 items
was .79 (M = 1.65, SD =.77). The second factor “Af-
measuring “sweat, butter?ies, cold, turned on, excited,
fective attitudes” (25 % explained variance) included the
dizzy, heartbeat” from 1high degree to 7low degree. Cronbach’s
items “turn-off/turn-on, uncomfortable/comfortable, bor-
Alpha reached .82 (M = 5.72, SD = 1.23).
ing/exciting, and bad/good”, Cronbach’s Alpha of the
Mood was measured with one item: “How would you
sumscore was .79 (M = 2.86, SD = 1.17).
describe your general mood today?” from 1very positive to
Behavioral beliefs were measured with two sets of 8
7very negative (M = 2.65, SD = 1.06).
items. The 8 pairs of items were multiplied with each
other (probability * evaluation of consequences). We
2.1.4 Design and statistics
?rst elicited the probability associated with a set of 8
A block-wise regression design was applied to partial out
potential negative and positive consequences of having
the separate contribution from (i) the shared variables, (ii)
unsafe sex; “pregnancy, catch HIV, catch other sexually
the unique TpB variables, and (iii) the unique RaF vari-
transmitted diseases, regretting, positive self image, feel-
ables to predicting the intention to have unsafe sex.
ing more attractive, getting a new boyfriend/girlfriend,
feeling proud” from 1very improbable to 7highly probable. Then
2.2 Results
we measured the evaluation of the same potential con-
sequences (how it would have been to experience each
As shown in Table 1, ?rst the (i) shared set of variables
of them) from 1bad to 7good. A sumscore based on the
was entered in the step-wise multiple regression analy-
multiplied items was entered into the regression analysis.
sis. The results suggested that the variables “cognitive
Cronbach”s alpha of the sumscore was .62 (M = 8.56, SD
attitudes, affective attitudes, and behavioural beliefs” ex-
= 4.06).
plained 57 % of the variance (p < .001) of the intention

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Risk-as-feelings and Theory-of-planned-behavior
573
Table 1: Multiple regression analysis. Prediction of intention to have unsafe sex (n = 94) (*** p<.001, ** p<.01, *
p<.05).
B
SEB
B
R2
?R2
Step 1. Shared variables
57***
Cognitive attitudes/evaluations
.29
.22
.13
Affective attitudes/evaluations
.95
.14
.63***
Behavioral beliefs
.03
.04
.08
Step 2. Unique TpB variables
.64***
.07**
Cognitive attitudes/evaluations
.11
.21
.05
Affective attitudes/evaluations
.78
.14
.51***
Behavioral beliefs
.03
.03
.06
Subjective norms (family)
?.05
.03
?.11
Subjective norms (friends)
.04
.02
.15*
Perceived control
.42
.13
.26**
Step 3. Unique RaF variables
.68***
.04*
Cognitive attitudes/evaluations
.10
.21
.04
Affective attitudes/evaluations
.71
.14
.47***
Behavioral beliefs
.06
.03
.13
Subjective norms (familty)
?.05
.03
?.10
Subjective norms (friends)
.04
.02
.16*
Perceived control
.44
.13
.27**
Vividness
.36
.12
.20**
Anticipatory emotions
.00
.10
.00
Mood
?.07
.11
?.04
to have unsafe sex. “Affective attitudes” was the only
ables before the TpB variables into the regression. This
signi?cant predictor; ? = .63, p < .001. By entering (ii)
did not yield any signi?cant changes in the results re-
the variables unique for the TpB “subjective norms (fam-
ported here in terms of signi?cant predictors or explana-
ily)”, “subjective norms (friends)”, and “perceived con-
tory power of the models.
trol” there was a 7% increase (p < .05) in explained vari-
ance, with altogether 64% (p < .001) of the variance ac-
2.3 Discussion
counted for. “Affective attitudes”; ? = .51, p < .001, “sub-
jective norms (friends)”; ? = .15, p < .05, and “perceived
The unsafe-sex vignette described a behaviour typically
control”; ? = .26, p < .05 were the signi?cant predic-
carried out because the activity is pleasant in itself, and/or
tors. By then adding (iii) the variables unique for the RaF
due to high intrinsic motivation. This is an affect-rich
model “vividness”, “anticipatory emotions”, and “mood”
type of decision situation, where feelings and emotions
into the model, the explained variance increased by 4 %
would be expected to play a large part. Despite the limi-
(p < .05), and altogether 68 % of the variance of the in-
tation of vignette studies to elicit behavioural intentions,
tention to have unsafe sex were explained (p < .001). The
and the remaining variables from the TpB and RaF, eth-
signi?cant predictors in the ?nal regression model were
ical restrictions leave out real-behaviour experiments in
“affective attitudes”; ? = .47, p < .001, “social norms
the unsafe-sex domain.
(friends)”; ? = 16, p < .05, “perceived control”; ? = .27, p
The results of Study 1 are important for two reasons.
< .01, and “vividness” of the consequences; ? = .20, p < .
First, we ?nd support for the idea that both feelings-based
01. Thus, the ?nal predictors consisted of a mixed subset
and consequence-based reasoning is underlying decision
of shared and unique variables from both TpB and RaF.
pertaining to unsafe sex. The fact that the ?nal predictors
The data were also analysed by entering the RaF vari-
consisted of a mixed subset of variables from both TpB

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Risk-as-feelings and Theory-of-planned-behavior
574
and RaF is noteworthy because it may contribute to tone
3 Study 2
down a static and dualistic view of human reasoning and
decision making as either fairly normative, rational and
3.1 Feelings or consequences: Ambivalence
focussed on consequences or as very intuitive, heuristic,
in decision making
and affect-based. Rather than focussing on these forms
of processing as two different phenomena, we argue that
Study 2 was carried out to further examine the relation-
there is a continuous dynamic interplay between them.
ship between the hedonistic qualities of behaviour and
At least in the type of decision dilemma used in Study
the consequences, by contrasting two decision vignettes
1, variables like global evaluations including affective
with opposite qualities, and where the TpB and RaF dif-
qualities, social-cognitive aspects such as norms among
fered in their behavioural predictions. As noted, some ac-
peer groups, control-expectations e.g., ability to self-
tivities are carried out because the activity is pleasurable
regulate ef?ciently in the dilemma situations, in addition
in itself, in spite of the potential negative consequences.
to how vividly one is able to imagine each of the potential
Other activities are carried out despite being painful, due
consequences, all play a signi?cant part in shaping the be-
to the prospect of positive consequences. Study 1 exam-
havioural intention. This means that even for unplanned
ined the former of these types of activities: Unsafe sex
behaviours, such as unsafe sex, variables from the conse-
may be a positive hedonic experience, but a range of neg-
quentialist models interact with the feelings-based judg-
ative consequences could follow. In Study 2, we wanted
ments.
to contrast this type of decision scenario with the oppo-
The anticipatory feelings did not contribute signi?-
site: A negative hedonic experience likely to be followed
cantly to the intention to have unsafe sex in our study. It is
by positive consequences. The decision was made to use
noteworthy that experiencing bodily emotional reactions
back surgery followed by the positive prospect of recov-
such as “butter?ies”, “turn-on”, “excited” or “heartbeat”,
ery from back pain: (4 variants of the back-surgery and
when thinking of having unsafe sex does not seem to play
unsafe-sex vignettes are presented in Appendix 1). The
a signi?cant role here. This ?nding can not be explained
back-surgery vignette (ambivalent version) read as fol-
by lack of emotional reaction, as the mean level of re-
lows:
ported anticipatory emotions was fairly high. Together
“Imagine you are in the emergency unit at the hospital
with a level of control reported as fairly low, this shows
after having hurt your back in a fall. You are in moder-
that the dilemma was successful in creating ambivalence
ate pain. The surgeon informs you that if an operation
and need for self-regulation.
may be of help it must be carried out immediately. The
The second important issue derived from our results is
operation will probably lead to severe pain for around 6
the ?nding that a set of attitudinal evaluations which in-
months, but in the long run, it may reduce the pain con-
cluded both predominantly cognitive and affective eval-
siderably. If you choose not to undergo surgery, the pain
uations could be separated into two distinct predictors,
will be stabilising at the present level, probably for the
which differed according to being either proximal or dis-
rest of your life. You feel that the pain is pretty bad al-
tal to the behaviour of concern. The factor “cognitive
ready, and fear how much worse it can get in the post-
attitudes” was characterised by more distal or even moral
surgical period. You are thinking that you would func-
evaluations such as whether having unsafe sex is wise,
tion much better in daily-life activities, if you undergo
necessary, and health promoting. The factor “affective at-
surgery.”
titudes” referred to more proximal or immediate aspects
The two vignettes basically described the unsafe-sex
of the behaviour, such as whether having unsafe sex is
situation from Study 1 and the back-surgery situation
pleasant, exiting, and arousing. The type of dichotomy
described above. There were four versions of each vi-
derived from several dual-process models of social psy-
gnette (see Appendix). The neutral version simply de-
chology (see Chaiken & Trope, 1999), and various deci-
scribed the basic facts needed to understand the situa-
sion theoretical models, ?t these two types of attitudes.
tion. The feelings version highlighted the subjects’ antic-
The cognitive attitudes are analogous to the analytical
ipatory and anticipated feelings in the decision dilemma,
and consequentialist view, whereas the affective attitudes
such as feeling sexual arousal or pain. The consequence
have more in common with the intuitive- and affect-based
version highlighted the considerations around potential
view. The affective attitudes were the signi?cant predic-
consequences of the decisions being made, e.g., suffering
tors for the intention to have unsafe sex. Despite the lim-
from regret or increasing one’s physical abilities. Finally,
itation of having only one vignette, with behaviour which
the ambivalent version outlined both the feelings and the
may be unplanned, we interpret this as partial support to
consequences. This manipulation was done to examine
the RaF model. However, the substantial contribution of
whether intentions are driven by potential consequences
factors derived from the TpB indicates a general support
of the behaviour (as described in the TpB) or rather by
to the TpB and the consequentialist view.
the feelings associated with it (as noted by the RaF).

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Risk-as-feelings and Theory-of-planned-behavior
575
TpB
RaF
7
7
Consesquences
Feeling
6
Ambivalence
6
Neutral
5
5
4
4
Intention
Intention
3
3
2
2
1
1
Unsafe sex
Back surgery
Unsafe sex
Back surgery
Figure 2: Hypotheses of TpB and RaF.
The four versions of the two vignettes yielded situa-
mainly produces negative consequences (unsafe sex) or
tions where the TpB and RaF differed in their assump-
positive consequences (back surgery). Based on the RaF,
tions. According to the TpB (Ajzen, 1991), intentions
in contrast, a general feelings-dominance was predicted:
are driven by behavioural beliefs, normative beliefs, and
High behavioural intentions in the unsafe-sex situation
control beliefs. The behavioural consequences and val-
with positive hedonic values, and low behavioural inten-
ues (including social values) combined with the person’s
tions in the back-surgery situation with negative hedonic
(dis)ability to carry out the behaviour are supposed to
values, a pattern showing opposite directionality than the
guide behavioural intentions. Thus, experiencing a de-
one predicted by the TpB.
cision scenario as more or less affect rich, should not af-
To further examine the dynamic interplay of cognitive
fect intentions. The RaF (Loewenstein et al., 2001) ar-
evaluations and feelings, we explored whether it was pos-
gues that there may be con?icts between cognitive evalu-
sible to reverse the predicted feelings-based dominance
ations and feelings, and that when such a con?ict arises,
by presenting the vignettes either more neutrally, by high-
behaviour tends to be based on feelings. Hence, a de-
lighting feelings or consequences, or by highlighting both
cision situation which highlights positive feelings, may
feelings and consequences (to create different combina-
lead to a feelings-based inclination to perform the be-
tions of ambivalence). According to the RaF there will
haviour, whereas one which involves negative feelings,
be effects of the presentation format, with high inten-
may lead to a feelings-based reluctance to engage in the
tions among those reading unsafe-sex vignettes with feel-
activity. However, a decision-situation that highlights
ings highlighted, and low intentions among those reading
the potential positive consequences of certain behaviours
back-surgery vignettes with feelings highlighted. Such
may foster the behaviour, and one highlighting the poten-
effects could not be accounted for by the TpB.
tial negative consequences of the behaviour, may impede
An additional goal was to replicate the ?ndings from
the performance of the behaviour. In these latter situa-
Study 1, that a mixed set of predictors from the conse-
tions, the predictions of whether a person will take part
quentialist and feelings-based perspective are the most
in a certain activity have the same directionality for both
powerful behavioural predictors.
TpB and RaF.
A set of distinct predictions were derived from the TpB
3.2 Method
and the RaF (Figure 2): Based on the consequentialist
TpB, we did not predict differential effects on intentions
3.2.1 Sample
based on the presentation format (neutral, feelings, con-
sequence, ambivalent). The TpB-based prediction were:
A total of 357 subjects were recruited from introduc-
Low behavioural intentions in the unsafe-sex scenario and
tory courses of psychology, economics, and mathemat-
high behavioural intentions in the back-surgery scenario,
ics. There were 230 females, 120 males, and 7 who did
indicating a general adherence to whether the behaviour
not report gender. Mean age was 22 years.

---

Judgment and Decision Making, Vol. 4, No. 7, December 2009
Risk-as-feelings and Theory-of-planned-behavior
576
3.2.2 Procedure
Shared TpB and Raf variables
Intention. The dependent variable, the behavioural
The experiments were carried out in the auditoriums dur-
intention to undergo back surgery, was measured with
ing an extended lecture break. The subjects were ran-
two items: “Do you consider undergoing surgery?”
domly distributed to four experimental groups, without
1
their awareness, by receiving different versions of a ques-
certainly not to 7 yes, de?nitely, and “how probable is it that
you will undergo surgery?” 1
tionnaire.
very improbable to 7highly probable.
The sumscores yielded Cronbach’s alpha back surgery = .82
Experimental groups
and alphaunsafe sex = .86 (Mback surgery = 6.30, SD = .90;
Group 1 received the neutral descriptions of an unsafe-
Munsafe sex =3.32, SD = 1.79).
sex vignette (positive experience, negative consequences)
Attitudes were measured with 9 bipolar evaluations in-
and a back-surgery vignette (negative experience, pos-
tended to capture both affectively and cognitively based
itive consequences). Group 2 read the vignettes which
evaluations pertaining to back surgery: “To undergo
highlighted feelings. Group 3 read the ones which high-
back surgery is: painful/pain free, unwise/wise, un-
lighted consequences. Group 4 read ambivalent vignettes
necessary/necessary, irresponsible/responsible, frighten-
with both feelings and consequences outlined (see Ap-
ing/calming, uncomfortable/comfortable, health damag-
pendix).
ing/health promoting, bad/good, stressing/relaxing, dan-
gerous/safe”. All items were anchored in 1 and 7 re-
3.2.3 Material
spectively for the bipolar evaluations. The items were
submitted to a principal component analysis (varimax
Each questionnaire ?rst presented the unsafe-sex or the
rotation), and yielded two factors, which explained al-
back-surgery vignette in one of four versions (neutral,
together 59 % of the variance in attitudes to back
feelings, consequences, or ambivalent), followed by the
surgery. The ?rst factor “Cognitive attitudes” (32 %
elicitations of self reports for the variables extracted from
explained variance) included the evaluations “irrespon-
the TpB and the RaF. The presentation order of the
sible/responsible, unnecessary/necessary, bad/good, un-
two vignettes (unsafe sex, back surgery) was counterbal-
wise/wise, dangerous/safe, and health damaging/health
anced, as were the presentation order of the elicitation of
promoting”; Cronbach’s Alpha = .79 (M = 6.17, SD =
the intention and the TpB and the RaF measures. Thus,
.79). The second factor “Affective attitudes” (25 % ex-
some subjects reported their intention right after having
plained variance) included the evaluations “uncomfort-
read the vignette, and then gave self reports to the TpB
able/comfortable, frightening/calming, painful/pain free,
and RaF variables, whereas others ?rst gave self reports
stressing/relaxing, and dangerous/safe”; Cronbach’s Al-
to the TpB and RaF variables, and then reported their in-
pha = .79 (M = 3.33, SD = 1.20).
tention for each type of behaviour (unsafe-sex intention
The attitude-item results concerning unsafe sex were
and back-surgery intention).
a near perfect replication of the results from Study 1: A
principal component analysis (varimax rotation) yielded
two factors, which explained altogether 57 % of the vari-
3.2.4 Measures
ance of the intention to have unsafe sex. The factor “Cog-
The shared and unique components from the TpB and
nitive attitudes” (32 % explained variance) included the
RaF models were operationalised and adapted to the be-
items “irresponsible/responsible, unwise/wise, danger-
haviours at hand, as in Study 1. For the unsafe-sex vi-
ous/safe, unnecessary/necessary, health damaging/health
gnette, the shared, unique TpB and unique RaF variables
promoting, and stressing/relaxing”; Cronbach’s alpha =
were elicited as reported in Study 1,3 and will not be re-
.82 (M =1.79, SD =.79). The factor “Affective attitudes”
peated here. The back-surgery variables were measured
(25 % explained variance) included the items “turn-
as reported below. These were the same for all four
off/turn-on, uncomfortable /comfortable, boring/exciting,
versions of the questionnaire (neutral, feelings, conse-
and bad/good”; Cronbach’s alpha = .67 (M = 3.84, SD =
quences, and ambivalence). Means, Cronbach’s alphas
1.43).
and factor scores based on the unsafe-sex vignette are
Behavioral beliefs were measured with two sets of 8
reported along with these measures based on the back-
items. The 8 pairs of items were multiplied with each
surgery vignette.
other (probability * evaluation of consequences). We ?rst
elicited the probability associated with a set of poten-
3The only change from the Study 1 version of the unsafe-sex vi-
tial negative and positive consequences of back surgery
gnette was the removal of the following sentences from the neutral con-
(pain reduction, improved physical condition, improved
dition: “You start to get intimate, and you are aroused. You notice how
sleep, long term sick leave, lack of recovery, feelings of
good it feels. You are thinking about the day after, and how uncomfort-
able you will feel.” The removed sentences did appear in the feelings-
regret due to lack of recovery, rejoicing over recovery,
and ambivalent versions of Study 2 (see Appendix).
feelings of having tried everything), indicated on scales

---