Article

Risky Therefore Not Beneficial:
Replication and Extension of Finucane
et al.’s (2000) Affect Heuristic Experiment

Social Psychological and
Personality Science
2022, Vol. 13(7) 1173–1184
Ó The Author(s) 2021
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/19485506211056761
journals.sagepub.com/home/spp

Emir Efendić1*, Subramanya Prasad Chandrashekar2*,
Cheong Shing Lee3*, Lok Yan Yeung3*, Min Ji Kim3*,
Ching Yee Lee3*, and Gilad Feldman3

Abstract
Risks and benefits are negatively related in people’s minds. Finucane et al. causally demonstrated that increasing risks of a
hazard leads people to judge its benefits as lower. Vice versa, increasing benefits leads people to judge its risks as lower
(original: r = 2.74 [20.92, 20.30]). This finding is consistent with an affective explanation, and the negative relationship is
often presented as evidence for an affect heuristic. In two well-powered studies, using a more stringent analytic strategy, we
replicated the original finding. We observed a strong negative relationship between judgments of risks and benefits across
three technologies, although we do find that there was no change in risks when highlighting low benefits. We note that risks
seem to be more responsive to manipulation (as opposed to benefits) and find evidence that the negative relationship can
depend on incidental mood. We provided materials, data sets, and analyses on https://osf.io/sufjn/.
Keywords
affect heuristic, judgment and decision-making, heuristics, risk, replication

Introduction
People tend to view risks and benefits as negatively related:
the riskier something is, the less beneficial it is. However,
risks and benefits are distinct concepts and are sometimes
even positively correlated—some technologies or hazards
that are beneficial may be high or low in risk, but those
that are not beneficial are unlikely to be high in risk. In a
seminal article, Finucane et al. (2000) proposed that the
negative relationship occurs due to an affect heuristic (AH)
whereby people rely on affect when judging the risks/benefits of specific hazards. Furthermore, they demonstrated
evidence that is consistent with an affective explanation of
this relationship. Take nuclear energy for example. The
AH proposes that increasing the risks of nuclear energy
(e.g., by exalting the hazard uranium has for human health)
turns the affective evaluation associated with it negative,
thereby leading people to judge its benefits as lower. Vice
versa, increasing benefits leads to positive affect and to
people judging its risks as lower (see Table 1).

Affect Heuristic
Affect is a crucial component of people’s decision-making
(Kahneman, 2003, 2011; Lerner et al., 2015; Loewenstein
et al., 2001; Rachlin, 2003). It is argued that reliance on

affect is often a much quicker, easier, and more efficient
way to navigate the complexities of everyday decisionmaking (Damasio, 1994; Schwarz & Clore, 1983) and that
affect informs many judgments and decisions (Albarracı́n
& Kumkale, 2003; Peters et al., 2006; Schwarz, 2012; Slovic
et al., 2002; Wyer et al., 1999).
Early studies of risk perception have shown that feelings
of dread are major determinants of public perception and
acceptance of risk for a wide range of hazards (Slovic,
1987). Focusing on this link, Finucane et al. (2000) proposed that people use an affect heuristic (AH) when making risk judgments. According to this view, people may use
their affective response to a risk (e.g., ‘‘How do I feel about
nuclear energy?’’) to infer how large they consider the risk
to be. The argument is that: ‘‘Using an overall, readily
available affective impression can be far easier—more
1

Maastricht University, School of Business and Economics, Department of
Marketing and Supply Chain Management, the Netherlands
2
Hong Kong Metropolitan University, Hong Kong
3
The University of Hong Kong, Hong Kong
*
Contributed equally, joint first authors.
Corresponding Author:
Gilad Feldman, Department of Psychology, The University of Hong Kong,
Pok Fo Lam Road, Hong Kong 999077.
Email: gfeldman@hku.hk

1174

Social Psychological and Personality Science 13(7)

Table 1. Summary of the Predictions According to the Affect Heuristic (AH).
Manipulated attribute

Impact on affect

Impact on non-manipulated attribute

Risk is high
Risk is low
Benefit is high
Benefit is low

Negative affect
Positive affect
Positive affect
Negative affect

Benefit is low
Benefit is high
Risk is low
Risk is high

efficient—than weighing the pros and cons or retrieving
from memory many relevant examples, especially when the
required judgment or decision is complex or mental
resources are limited’’ (Finucane et al., 2000, p. 3).
Reliance on affect is a general process and, consistent with
an AH, a wide range of findings support the idea that affect
provides valuable information that people use to simplify
their decision-making. For instance, affect-laden imagery has
been shown to predict people’s preferences in investment
decisions (MacGregor et al., 2000), smoking (Benthin et al.,
1995), information integration (Anderson, 1981; Efendić
et al., 2019), simple choice gambles (Bateman et al., 2007),
and morality judgments (Slovic & Västfjäll, 2010).

Risks and Benefits
For a long time, the negative relationship between judgments
of risks and benefits puzzled researchers (Fischhoff et al.,
1978) as these judgments should be positively correlated or
independent of one another (Slovic, 1987). In a breakthrough
study, Alhakami and Slovic (1994) found that the negative
relationship was linked to how a person generally feels about
a hazard. Later, Finucane et al. (2000) showed that the
inverse relationship between risk and benefits was strengthened under time pressure designed to limit analytic thinking
(their Study 1) and that it is causally determined. Specifically,
manipulating one attribute—for example, increasing risk—
led to an affectively congruent but inverse relationship, that
is, decreased benefit and vice versa (their Study 2).
This inverse relationship has been observed elsewhere as
well. It has been found that when general negative affect is
evoked (i.e., participants were shown photographs depicting houses in flooded regions), this led to increased levels
of perceived risk (Keller et al., 2006). Similarly, incidental
negative affect (e.g., negative mood) was found to amplify
reliance on affect, which led to stronger negative correlations between risks and benefits (Västfjäll et al., 2014).
Interestingly, affective association with a particular hazard
has been shown to influence the interpretation of new
information. People evaluated nuclear power more negatively than solar power because of more negative feelings
associated with nuclear power (Siegrist & Sütterlin, 2014).
Similar negative associations between risk and benefits
have been found in consumer judgments of novel products
(King & Slovic, 2014), in the financial domain (Ganzach,
2000), and in wood smoke pollution (Bhullar et al., 2014).

Recently, Skagerlund et al. (2020) found that the negative
correlation is tied to cognitive reflection ability.

Replication Value and Present Research
In this article, with two well-powered studies, we aimed to
closely replicate and extend our understanding of the causal
demonstration of the negative relationship between risks
and benefits, using the same materials and procedure as in
the original paper (Finucane et al., 2000).
We chose to replicate Study 2 from Finucane et al.
(2000) for several reasons. First, while many correlational
studies have found the negative relationship, few demonstrated it causally. King and Slovic (2014) used a similar
method as Finucane and colleagues, but other work
mostly found correlational support (some research has
even failed to find the same relationship, Raue et al.,
2019). There is therefore value in demonstrating, with
sufficient statistical power, whether the causal effect is
robust. Second, the analysis approach used in the original
studies and in later demonstrations of the negative relationship (e.g., King & Slovic, 2014) were nonstandard,
failing to account for non-independence of data and relying on counting the number of times the manipulation
worked in the predicted direction—a strategy that leads
to large information loss. A more stringent analytic
approach with mixed-effect modeling ought to provide
information on the generalizability of the effect. Third,
the findings are relevant for risk communication.
Changing risk/benefit judgments by manipulating solely
one attribute (either risk or benefit) has vast applied
potential. Risk campaigns can focus on changing people’s
judgments about many plights of today’s society (e.g.,
smoking, obesity, and so on). Fourth, as of this writing,
we are unaware of any other attempts to directly replicate
this study. This is surprising given the relevance in understanding the relationship between risks and benefits, as
well as the popularity of the original article and how it
promoted the AH in the judgment and decision-making
literature. As of this writing, the original article has been
cited 3,363 times with a later updated review article being
cited 3,860 times (Slovic et al., 2007).
We also wish to highlight an important distinction. The
observation of the negative relationship is often presented
as evidence for an AH in risk judgments. For example,
observing the negative relationship leads authors to

Efendić et al.
conclude that the AH is a robust phenomenon (Skagerlund
et al., 2020). However, the original, as well as many other
studies, fail to demonstrate that it is affect that mediates
this relationship (although converging evidence on the
importance of affect would suggest this is the case). Our
aim here is to replicate the negative causal relationship
between risks and benefits. As such, this replication also
does not speak to the mechanism that underlies the relationship. Other more cognitive, rather than affective,
mechanisms remain a plausible explanation. Nevertheless,
we hope that investigating whether the causal relationship
replicates will (a) provide important insight into this interesting phenomenon and (b) serve other researchers who
wish to use the paradigm to further understand whether it
is affect or something else that explains it.
We thus consider this investigation to be a needed direct
replication. Replications should be sufficiently similar to
the original study to adequately gauge support for the original findings (LeBel et al., 2019). Furthermore, given the
prevalence of publication bias (Bakker et al., 2012), a close
replication adds value by providing evidence that strengthens or weakens the finding.

Overview of Studies
This replication was part of an ongoing replications project
(see Supplementary Figure S1 and the project process section in the supplementary material for more details). We
crowdsourced the replication using two teams, both teams
being supervised by experienced authors. Each team collected data independently and wrote detailed preregistrations. We thus report the results of two studies serving as
close replications of Study 2 from Finucane et al. (2000),
using the same methodology and the same materials.1 The
two studies differ only in the target sample, one obtained
on MTurk (U.S. participants) and the other on Prolific
(U.K. participants). The two studies were preregistered on
the OSF (MTurk: https://osf.io/ab5dw/files/; Prolific:
https://osf.io/p4qjx/files/).2 All materials, data sets, and
analysis scripts are available on OSF (https://osf.io/sufjn/).
We report how we determined the sample size, all data
exclusions (if any), all manipulations, and all measures.

Extensions
In addition to the direct replication of Study 2 from
Finucane et al. (2000), we also report two extensions. First,
we looked at the effect of naturally occurring incidental
mood on the negative relationship between judgments of
risks and benefits. In the MTurk sample, participants were
asked to rate their current levels of (a) pleasure—unpleasant vs. pleasant and (b) arousal—deactivated vs. activated
(using two affective sliders that ranged from 2100 to 100,
centered in the middle). We based our measure on core
affect that represents states experienced as simply feeling

1175
good or bad, energized, or enervated (Russell, 2003). We
use the term ‘‘naturally occurring incidental mood’’ to
highlight that this is a measured rather than manipulated
variable and that the affect in question is incidental (i.e.,
unrelated to the judgment at hand). Any affect that arises
due to changes in risk/benefit descriptions is integral (i.e.,
affect stemming from the judgment target at hand). Several
predictions can be made on how naturally occurring incidental mood could impact the negative relationship: (a)
incidental mood is misattributed (Schwarz, 2012) to risk/
benefit judgments impacting the strength of the negative
correlations, (b) incidental affect has a specific effect in that
negative incidental affect leads to high risk and low benefit,
while positive incidental affect leads to low risk and high
benefit, not impacting the strength of the negative correlations; or (c) it has a negation effect where, akin to mood
regulation models for example (Andrade, 2005), being in a
pleasurable naturally occurring mood may interfere with
people’s ability to effectively map a negative change in integral affect (e.g., by describing risks as high). Highlighting
the interaction between such incidental and integral states
can offer insights into the role of affect in the negative
relationship.
Second, we explored whether there was a stronger negative relationship when risks, as opposed to benefits, are
manipulated. Illuminating this boundary condition could
provide insight into which of these two attributes people
find more informative or important for their risk
judgments.

Method
Participants
In the first study, a total of 806 participants from the
United States were recruited through MTurk using the
TurkPrime platform (Litman et al., 2017). In the second,
a total of 1,008 participants from the United Kingdom
were recruited through Prolific. To determine the number
of participants needed, we conducted a power analysis
planning to detect the weakest effect size reported in the
original that was also significant (at p \ .05). Therefore,
given our resource constraints, we based our power analysis on having 95% power to detect a Cohen’s dz = 0.30.
This resulted in a suggested sample size of 147 participants per condition and a total of 588 across 4 betweensubject conditions. Finally, we aimed for a higher sample
size between 750 and 800 participants, as this would also
ensure we were able to detect a smaller effect size
(Cohen’s dz) of .20 at 80% power. A comparison of the
target article sample and the replication samples is provided in Table S1 in the supplementary material.
To obtain the final sample, we first excluded (30 from
MTurk sample and 40 from Prolific sample) participants
following our preregistered exclusion criteria.3 Because the

1176

Social Psychological and Personality Science 13(7)

studies were identical, we combined4 them for the final data
analysis, resulting in 1,552 participants (MTurk = 776;
Prolific = 776; MAge = 38.99, SDAge = 12.30; 822 females,
727 males, 3 would rather not say).

and confidentiality. We characterize the current replication
as a ‘‘very close replication’’ based on the framework for classification of the replications using the criteria by LeBel et al.
(2018; see Table S45 in the supplementary material).

Design, Procedure, and Measures

Results

Both studies had a 2 (Between-subject factor—Direction:
High vs. Low) 3 2 (Between-subject factor—Manipulated
Attribute: Risk vs. Benefit) 3 3 (Within-subject factor—
Technology Scenario: Nuclear Power vs. Natural Gas vs.
Food Preservative) mixed-subject design (see Table S3 and
Table S4 in the supplementary material for more details
and full descriptions of the measures and direction/
attribute information). Please note that the second study
(Prolific) included an additional experimental condition
that was excluded due to a methodological issue.5
Participants were first asked to answer questions regarding the perceived benefit and risk of all three technologies
(Nuclear Power, Natural Gas, Food Preservatives)—the
same ones used in the original study. The presentation of
the technologies was randomized. Participants were asked
two questions, in random order, for each technology,
namely: ‘‘In general, how risky [beneficial] do you consider
the use of nuclear power / natural gas / food preservatives?’’6, answering on a 10-point scale from 1 (not at all
risky [beneficial]) to 5 (moderate risk [benefit]) to 10 (very
risky [beneficial]).
Subsequently, dependent on the conditions, participants
were presented with textual vignettes designed to
change the affective quality (e.g., high risk = negative,
high benefit = positive, and so on) of the scenarios. We
used the same descriptions from the original study (https://
osf.io/y97tp/). For example, in the low benefit condition
for the hazard natural gas, participants were presented
with the following text (shortened):

Analysis Strategy

Natural gas is used as a source of energy in the US. Natural
gas has the property of being a gas at room temperature, which
allows it to be burned to produce heat. However, this same
gaseous property limits the energy tasks that natural gas can
be used for. Natural gas is not able to replace electricity for
such tasks as lighting, or the numerous jobs that need electric
motors, such as refrigeration or the operation of machinery.

After reading the information, participants again provided
answers to the risk and benefit questions for each technology
scenario. Please note that once participants were assigned to
one of the between-subject conditions, they were in that condition for all three scenarios, as the scenario was a withinsubject variable. This means that we had data from 4,656
trials. Finally, participants answered a funneling section and
provided demographic information. At the end of the study,
a short debriefing was given regarding the study’s purpose

We report both the original (i.e., repeating the same analytic strategy as in Finucane et al., 2000) and an improved
analytic approach. For the improved, we employed linear
mixed-effects models (LMEM) using the lme4 package in
R (Bates et al., 2015). Significance for fixed effects was
assessed via Satterthwaite’s degrees of freedom
(Kuznetsova et al., 2017). Unless stated otherwise, the
models adjusted for covariates at Level 1 (ratings of risks
and benefits before the experimental treatment) and Level
2 (i.e., Technology type and participants’ ID were treated
as random effects). We added pre-scores on the manipulated/nonmanipulated attribute to reduce noise of our
assessment and to check whether the preratings may moderate the effect of the manipulation. LMEMs reduce the
chance of Type I errors, account for nonindependence of
data points (e.g., within-subject observations), provide a
greater flexibility with specification of the covariance structure, and allow us to make more generalizable claims
across samples of participants and stimuli (hazards in our
case; Judd et al., 2012).

Original Data Analytic Approach (Finucane et al., 2000)
Descriptive statistics of the measures across the two samples are noted in Table S39 and Table S40 of the supplementary material. Following the original approach, we
conducted paired samples t tests (two-tailed). Specifically,
for each technology, we compared the mean pre- and postmanipulation ratings of the manipulated and the nonmanipulated attributes. Positive t-values indicate that
there was an increase in rating after manipulation.
Negative t-values indicate there was a decrease in rating
after manipulation. The results are in line with the original
finding (See Table S41–S44 in the supplementary material
for the detailed results). That is, for the manipulated attribute ratings, providing information on high and low benefits or risks led to higher and lower post-manipulation
ratings of benefits or risks. For the non-manipulated attribute, we see the inverse: providing information on high
and low benefits or risks led to lower and higher postmanipulation ratings of risk and benefits.
Furthermore, we tested the correlation between risk
and benefits using the t-values from the abovementioned
analysis. We found strong support for a negative
correlation: MTurk sample: r(10) = 2.87, 95% confidence

Efendić et al.

1177

Figure 1. t-Values for Manipulated Versus Non-Manipulated Attributes.
Note. t-values for four-direction/attribute information manipulations (HB = High Benefit; LB = Low Benefit, HR = High Risk, LR = Low Risk) for the three
technologies (nuclear power, natural gas, and food preservatives) across the two samples (MTurk and Prolific). The negative slope shows the predicted
negative relationship between risks and benefits—as benefits increase risks decrease and as risks increase benefits decrease.

interval (CI): [20.96, 20.59], p = .003; Prolific sample:
r(10) = 2.84, 95% CI = [20.95, 20.50], p \ .001.
Plotting the t-values in Figure 1, the negative slope shows
that when ratings on the manipulated attribute increase,
ratings on the non-manipulated attribute decrease (and
vice versa). Simply put, when benefits increase risks
decrease and when risks increase benefits decrease, indicating a negative relationship.

Mixed-Model Approach
Manipulation Checks. We conducted LMEMs with change in
the manipulated attribute as the DV (i.e., ratings on a
manipulated attribute after experimental treatment minus
ratings on manipulated attribute before experimental treatment; 0, therefore indicates no change, a positive value an
increase, and negative value indicates a decrease). Table 2
presents the fixed-effects coefficients with all the predictors
(See Table S11–S14 in the supplementary material for stepby-step regression results).
The significant effect of Direction shows that, regardless
of the manipulated attribute, if the direction was high
there was a positive change while if the direction was low
there was a negative change, indicating a successful

manipulation check (see Figure 2 and Tables S41–S44 for
detailed statistics).

Negative Relationship Between Risks and Benefits. To test
whether we observe a negative relationship between risks
and benefits, we looked at the effects of the manipulated
attribute on the nonmanipulated attribute. Specifically, we
regressed change in ratings of nonmanipulated attributes
(DV) on Direction, Manipulated Attribute, and their interaction, adjusting for covariates at Level 1 (Pre-rating
manipulated attribute; and three-way interaction between
pre-rating non-manipulated attribute, Direction, and
Manipulated Attribute) and Level 2 (i.e., Technology type
and participant’s ID). Table 3 summarizes these results (see
Table S20–S24 in the supplementary material for step-bystep regression results and model comparisons).
The main effect of direction supports the original finding of the negative relationship. In addition, we find that
the directionality of pre- and post-treatment changes in the
non-manipulated attribute was consistent with the predicted inverse relationship, except in the Low-benefit condition (see Figure 3 and Tables S41–S44 for detailed
statistics).

1178

Social Psychological and Personality Science 13(7)

Table 2. Estimated Fixed-Effects Coefficients of the Mixed-Effects Regression Model With Change in the Manipulated Attribute
as the DV.
Predictors

Intercept
Pre-rating manipulated attribute (PMA)
Direction (high vs. low)
Manipulated attribute (risk vs. benefit)
Direction 3 manipulated attribute
PMA 3direction
PMA 3 manipulated attribute
PMA 3 direction 3 manipulated attribute

DV: Change in non-manipulated attribute
B

SE

95% CI

p

20.09
21.09
2.56
20.27
0.49
20.10
0.01
0.16

0.06
0.03
0.07
0.07
0.14
0.06
0.06
0.12

[20.21, 0.04]
[21.15, 21.03]
[2.42, 2.69]
[20.40, 20.13]
[0.22, 0.75]
[20.22, 0.02]
[20.11, 0.14]
[20.08, 0.40]

.185
\.001
\.001
\.001
\.001
.109
.819
.199

Note. Variables were coded as follows—direction: 20.5 = low, + 0.5 = high; attribute: 20.5 = benefit, + 0.5 = risk. CI = confidence interval.

Exploratory Analysis: Mediation Effects. We also tested whether
the effect of the experimental manipulation on change in
the non-manipulated attributes was mediated by the
changes in the manipulated attribute as the analytic reasoning would suggest. To do this, we conducted a multilevel mediation analysis (this analysis was not part of the
pre-registration). Bayesian estimation of the multilevel
mediation model was performed using the bmlm R package
(Vuorre & Bolger, 2018). Because our experimental design
involved two directions (High vs. Low), we conducted two
independent mediation analyses that looked at the
responses within High and Low separately. Indeed, both
sets of mediation analysis show a significant indirect
effect of manipulation on non-manipulated attribute rating
through manipulated attribute rating (High only mediation: Mposterior = 20.54, SD=0.04, CI = [20.61, 20.47];
Low only mediation: Mposterior=0.55, SD=0.04, CI=
[0.48, 0.62]). For details results see Table S25–S26 in the
supplementary material.

Figure 2. Distribution of Ratings on Change in Manipulated
Attribute as DV by Experimental Conditions.
Note. Figure includes violin plots displaying the distribution of responses,
boxplots displaying the median, first, and third quartiles, while the mean value
is identified by the red circle.

Extensions
Naturally Occurring Incidental Mood and the Negative Relationship
Between Risks and Benefits. We conducted an analysis where
the change in ratings of manipulated attributes, level of
pleasure, level of arousal, and their interaction were set as
predictors of change in the ratings of the non-manipulated
attributes. Table 4 and Figure 4 summarize the results. As
a representation of the negative relationship between risks
and benefits, we looked at predicting change in nonmanipulated attribute with change in manipulated attribute. Indeed, a negative correlation between these two variables represents the negative relationship. We decided to
use this (rather than an interaction between the dummy
coded direction and manipulated attribute), as it is easier
to represent and interpret a potential two-way interaction
with pleasure or arousal.
We found some support that the negative relationship is
moderated by incidental pleasure (see Figure 4).

Specifically, the negative relationship was stronger among
participants who reported higher incidental pleasure in
comparison to participants who reported lower incidental
pleasure.

Risk/Benefit Strength. We also examined whether there was a
stronger negative relationship when risks, as opposed to
benefits were manipulated and the extent to which it may
depend on the manipulated conditions. For the analysis,
similar to above, we again used the change in ratings of
manipulated attributes, Manipulated Attribute (Risk vs.
Benefit), Direction, and their interaction as predictors of
change in the ratings of the non-manipulated attributes.
Table 5 and Figure 5 summarize the results.
The interaction between manipulated attribute and
CMA (change in manipulated attribute) indicates that the

Efendić et al.

1179

Table 3. Estimated Fixed-Effects Coefficients of the Mixed-Effects Regression Model With Change in the Non-Manipulated Attribute as the
DV.
Predictors

Intercept
Pre-rating manipulated attribute (PMA)
Pre-rating non-manipulated attribute (PNMA)
Direction (high vs. low)
Attribute (risk vs. benefit)
PNMA 3 Direction
PNMA 3 Attribute
Direction 3 Attribute
PNMA 3 Direction3 Attribute

DV: Change in non-manipulated attribute
B

SE

95% CI

p

20.26
20.21
20.95
21.15
0.55
0.14
20.16
21.34
0.13

0.10
0.03
0.03
0.06
0.06
0.05
0.06
0.12
0.11

[20.45, 20.06]
[20.27, 20.15]
[21.01, 20.89]
[21.27, 21.03]
[0.43, 0.67]
[0.04, 0.25]
[20.27, 20.05]
[21.58, 21.10]
[20.08, 0.35]

.009
\.001
\.001
\.001
\.001
.008
.004
\.001
.221

Note. Variables were coded as follows—direction: 20.5 = low, + 0.5 = high; attribute: 20.5 = benefit, + 0.5 = risk. CI = confidence interval.

Figure 3. Distribution of Rating on Change in Non-Manipulated
Attribute as DV by Experimental Conditions.
Note. Figure includes violin plots displaying the distribution of responses,
boxplots displaying the median, first, and third quartiles, while the mean value
is identified by the red circle.

strength of the negative relationship between the manipulated and non-manipulated attribute was stronger when
risks, as opposed to benefits, were manipulated.
Furthermore, the three-way interaction (Direction 3
Manipulated Attribute 3 CMA) suggests that the extent
of difference between risks and benefits varies as a function
of the direction of manipulation (High vs. Low).
Proceeding to conduct separate analyses for Low and High
conditions, results within the high condition show no support for interaction. However, results within the low condition do find support for the interaction (See Table S32 and
Table S33 in the supplementary material for detailed
results). This lack of consistency leads us to conclude
that the strength of the negative relationship between the
manipulated and the non-manipulated attribute being
stronger when risks, as opposed to benefits, were

Figure 4. The Interaction Between Change in Manipulated
Attribute and Pleasure on Change in Non-Manipulated Attribute

manipulated is mainly driven by participants’ responses
within the Low-Benefit condition (see Figure 5).
Specifically, we note large differences in change in ratings
of non-manipulated attribute across Risk, Mchange = 0.74
(SE = 0.05) and Benefit, Mchange = 20.14, (0.05),
manipulation within the low condition. However, those differences are much smaller within the high condition, Risk:
Mchange = 21.01 (0.06); Benefit: Mchange = 20.62, (0.05).

General Discussion
In two studies, using samples from the United States and
the United Kingdom, we re-did Study 2 from Finucane
et al. (2000). With high power and using a more precise
analytic approach, we successfully replicated and obtained
a similar effect as in the original study providing support

1180

Social Psychological and Personality Science 13(7)

Table 4. Estimated Fixed-Effects Coefficients From the Mixed-Effects Regression Model Adding Pleasure and Arousal Measures on Change
in Non-Manipulated Attribute as DV.
Predictors

Intercept
Pre-rating non-manipulated attribute (PNMA)
Pre-rating manipulated attribute (PMA)
Pleasure
Arousal
Change in manipulated attribute (CMA)
Direction (high vs. low)
Manipulated attribute (risk vs. benefit)
Pleasure 3 Arousal
Pleasure 3 CMA
Arousal 3 CMA
Pleasure 3 Arousal 3 CMA

DV: Change in non-manipulated attribute
B

SE

95% CI

p

20.59
20.63
21.05
0.03
20.06
20.70
0.30
0.39
20.02
20.09
0.05
20.03

0.15
0.05
0.04
0.05
0.05
0.05
0.09
0.08
0.03
0.04
0.04
0.03

[20.88, 20.29]
[20.72, 20.54]
[21.13, 20.97]
[20.07, 0.13]
[20.16, 0.04]
[20.79, 20.61]
[0.13, 0.48]
[0.23, 0.56]
[20.08, 0.04]
[20.16, 20.01]
[20.04, 0.13]
[20.09, 0.02]

\.001
\.001
\.001
.557
.266
\.001
.001
\.001
.536
.025
.293
.201

Note. CI = confidence interval.

Table 5. Estimated Fixed-Effects Coefficients From the Mixed-Effects Regression Model Looking at Moderation of the Negative Relationship
by Risks/Benefits.
Predictors

Intercept
Pre-rating manipulated attribute (PMA)
Pre-rating non-manipulated attribute (PNMA)
Direction (high vs. low)
Manipulated attribute (risk vs. benefit)
Change in manipulated attribute (CMA)
Direction 3 Manipulated Attribute
Direction 3 CMA
CMA 3 Manipulated Attribute
Direction 3 Manipulated Attribute 3 CMA

DV: Change in non-manipulated attribute
B

SE

95% CI

p

20.24
20.59
21.01
20.37
0.44
20.74
20.85
20.12
20.27
0.22

0.13
0.03
0.03
0.06
0.06
0.03
0.12
0.05
0.05
0.11

[20.49, 0.02]
[20.65, 20.53]
[21.06, 20.95]
[20.49, 20.24]
[0.31, 0.56]
[20.80, 20.68]
[21.09, 20.60]
[20.23, 20.02]
[20.37, 20.16]
[0.01, 0.43]

.066
\.001
\.001
\.001
\.001
\.001
\.001
.022
\.001
.037

Note. Variables were coded as follows—direction: 20.5 = low, + 0.5 = high; attribute: 20.5 = benefit, + 0.5 = risk. CI = confidence interval.

for the demonstration of a causal negative relationship
between risks and benefit judgments. Specifically, we
showed that increasing the risks of three technologies
(nuclear energy, food preservatives, and natural gas) led to
lower judgments on benefits while increasing the benefits
led to lower judgments on risks. Vice versa, decreasing
risks led to higher judgments of benefits. However, we did
not find any differences in the low-benefit conditions.
Specifically, decreasing the benefits did not lead to higher
judgments of risks (See Table S41–S44 in the supplementary material for detailed results).
In addition, we report two extensions. First, we found
that the negative relationship between risks and benefits
was stronger among participants who reported feeling
higher incidental pleasure. Concurrently, people who felt
pleasant may have generally relied more on heuristic
processing—in this case the AH (Bohner et al., 1995).
Previous findings, which manipulated negative mood,
showed increased risk perceptions (Västfjäll et al., 2014).

This may indicate that negative mood has a more pointed
effect on risk-benefit judgments, although our findings cannot speak on this as we did not have a lot of data on the
negative side of our measures, meaning we had few participants feeling low pleasure and low arousal (see Figure S5
in the supplementary material). This may have reduced our
chances of obtaining more precise findings on how incidental affect can modulate the negative relationship.
Furthermore, it is important to note that we measured
naturally occurring incidental mood whereas previous
research manipulated mood directly.
Second, we looked at whether manipulating risks or
manipulating benefits impacts the strength of the negative
relationship. Initially, our results showed the strength of
the negative relationship was stronger when risks, as
opposed to benefits, were manipulated. However, a more
detailed look shows that this effect is most likely a product
of the fact that there was no impact on the nonmanipulated attribute in the low-benefit condition (see

Efendić et al.

1181

Figure 5. Relationship Between Manipulated and Non-Manipulated Attributes as a Function of Risk/Benefit Manipulations

Table S41–S44; the original findings seem to show this as
well; see Table S34 and Table S35 in the supplementary
material for detailed results). It is worth pointing out that
manipulating low benefits did lead to a predicted change in
benefits—people judged them as considerably lower (i.e.,
there was a successful manipulation; see Table S41–S44).
But decreasing benefits did not lead to the predicted impact
on risks. This may hint at the fact that providing low benefit info is not enough to lead to perceptible changes in
affect as it may be that risks are simply better at evoking
an affective reaction (cf. Pachur et al., 2014). Our results
also hint at the fact that people may pay more attention to
risks—both increase and decrease in risks—while this is
not the case for benefits, where only increase in benefits led
to perceivable changes. Alternatively, the lack of impact on
the non-manipulated attribute in the low-benefit condition
may hint at sensitivity to the actual relationship of risks
and benefits in the world, namely, that they are often positively correlated. As mentioned in the introduction, technologies low in benefit are unlikely to be high in risk. It is
of course not incommensurable that this sensitivity exists
along a strong affective process that leads to negative relationships between risks and benefits.
Current findings may have important implications for
risk communication (Thaler & Sunstein, 2008; Yang et al.,
2014). For instance, communication efforts about new
technologies ought to contend that risk information may
overweigh other benefit information and is more malleable
to manipulate. While out of scope for this research, it may
be worth taking a closer look at what associations people
might have with the terms ‘‘risky’’ and ‘‘beneficial.’’
Specifically, people may already associate and interpret

these terms as ‘‘bad’’ (for risky) and ‘‘good’’ (for beneficial),
explaining the negative correlation.
We believe this replication strengthens the claim that it
is possible to causally affect risk and benefit judgments.
The negative relationship has been presented as a demonstration of the AH. However, while the effect is consistent
with an AH, we (as the original finding) do not provide
direct evidence that affect does mediate this negative relationship. Indeed, the negative relationship could also occur
due to a more cognitive explanation. While we show evidence that change in the manipulated attribute is a mediator between the manipulations and non-manipulated
attribute, this may be one of the potential mediators and
the underlying cause remains uncovered. Some recent
research has, for example, found more support for manipulations of availability by the recall, rather than affect, to
have a stronger impact on how risk judgments are constructed (Efendić, 2021). Nevertheless, with this replication,
we hope to encourage future researchers that this paradigm
is robust and could potentially be used to tease apart any
cognitive/affective explanations of risk/benefit judgments.
Finally, in our replication, we focused on the original
three technological scenarios as the risky hazards. While
one could argue that people’s attitudes toward these risks
have changed in the intervening 20 years since the original
study, impacting the strength of the negative relationship,
our results show similar effects. This could indicate that
either the attitudes did not change, or, equally likely, that
the manipulations of risk/benefit go well and beyond
beliefs and attitudes. In that sense, future work should look
at whether the negative relationship extends to other
hazards. For instance, Skagerlund et al. (2020) found that

1182

Social Psychological and Personality Science 13(7)

the inverse relationship extends to numerous other hazards,
activities, and technologies.
4.
Author Contributions
G.F. led the project, supervised each step of the project, conducted
the pre-registration, and ran data collection. E.E. and S.P.C. followed up on initial work by the other coauthors to verify and conduct additional analyses, and completed the manuscript draft.
E.E., S.P.C., and G.F. jointly finalized the manuscript for submission. C.S.L., L.Y.Y., M.J.K., and C.Y.L. conducted the replication and extension as part of university course work. They
conducted an initial analysis of the paper, designed the replication,
initiated the extensions, wrote the pre-registration, conducted initial data analysis, and wrote initial replication reports.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with
respect to the research, authorship, and/or publication of this
article.

The author(s) disclosed receipt of the following financial support
for the research, authorship, and/or publication of this article:
The research was supported by the Teaching Development Grant
of the University of Hong Kong. S.P.C. thanks the Institute of
International Business and Governance (IIBG), established with
the substantial support of a grant from the Research Grants
Council of the Hong Kong Special Administrative Region, China
(UGC/IDS 16/17), for its support.

ORCID iD
https://orcid.org/0000-0003-2812-6599

Supplemental Material
The supplemental material is available in the online version of the
article.

Notes
1.
2.

3.

6.

References

Funding

Gilad Feldman

5.

United Kingdom. Please see Table S2 in the supplement
for more detail.
We ran all the models below with study included as a fixed
effect and we did not find any evidence that the results differed between studies. Please see tables S9, S13, S18, and
S23 in the supplement.
The additional experimental condition presented participants both information on risk and benefit. This presentation made it impossible to test the negative relationship
and we saw fit to exclude it. Some 192 of the 968 participants in the prolific sample were in the excluded condition.
Responses from remaining 776 prolific participants was
included in the final analysis. Please see also note 2 in
Table S1 and Table S5 in supplement for more details.
In the original study, the question added the phrasing ‘‘. . .
to U.S. society as a whole’’ at the end. We used this exact
phrasing in the MTurk sample (which included people
from the United States) but decided to exclude this for the
Prolific sample as these participants were from the United
Kingdom.

We would like to thank the original authors for providing
the materials.
Note that the preregistrations follow a registered report format. This means that a manuscript-like document was produced reporting simulated random data results. Please see
also the Read-me document in the wiki page on the OSF
preregistrations here: https://osf.io/pg3ae/ for a detailed
guide on where to find information on preregistered materials, design, and analysis plan.
Indicating a low proficiency of English, self-report not
being serious about filling in the survey, who guessed the
hypothesis, have done the survey before, who failed to
complete the survey, and those not from the United States/

Albarracı́n, D., & Kumkale, G. T. (2003). Affect as information
in persuasion: A model of affect identification and discounting.
Journal of Personality and Social Psychology, 84(3), 453–469.
https://doi.org/10.1037/0022-3514.84.3.453
Alhakami, A. S., & Slovic, P. (1994). A psychological study of the
inverse relationship between perceived risk and perceived benefit. Risk Analysis, 14(6), 1085–1096. https://doi.org/10.1111/
j.1539-6924.1994.tb00080.x
Anderson, A. (1981). Foundations of information integration theory. Academic Press.
Andrade, E. B. (2005). Behavioral consequences of affect: Combining evaluative and regulatory mechanisms. Journal of Consumer Research, 32(3), 355–362. https://doi.org/10.1086/497546
Bakker, M., van Dijk, A., & Wicherts, J. M. (2012). The rules of
the game called psychological science. Perspectives on Psychological Science, 7(6), 543–554. https://doi.org/10.1177/
1745691612459060
Bateman, I., Dent, S., Peters, E., Slovic, P., & Starmer, C. (2007).
The affect heuristic and the attractiveness of simple gambles.
Journal of Behavioral Decision Making, 20(4), 365–380. https:
//doi.org/10.1002/bdm.558
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting
linear mixed-effects models using lme4. Journal of Statistical
Software, 67(1), 1–48. https://doi.org/10.18637/jss.v067.i01
Benthin, A., Slovic, P., Moran, P., Severson, H., Mertz, C. K., &
Gerrard, M. (1995). Adolescent health-threatening and healthenhancing behaviors: A study of word association and imagery. Journal of Adolescent Health, 17(3), 143–152. https:
//doi.org/10.1016/1054-139X(95)00111-5
Bhullar, N., Hine, D. W., Marks, A., Davies, C., Scott, J. G., &
Phillips, W. (2014). The affect heuristic and public support for
three types of wood smoke mitigation policies. Air Quality,
Atmosphere and Health, 7(3), 1–10. https://doi.org/10.1007/
s11869-014-0243-1

Efendić et al.
Bohner, G., Moskowitz, G. B., & Chaiken, S. (1995). The interplay of heuristic and systematic processing of social information. European Review of Social Psychology, 6(1), 33–68.
Damasio, A. R. (1994). Descartes’ error: Emotion, reason, and the
human brain. Avon.
Efendić, E., Drače, S., & Ric, F. (2019). The combination of multiple affective experiences and their impact on valuation judgments. Cognition and Emotion, 34(4), 684–699. https:
//doi.org/10.1080/02699931.2019.1675597
Efendic E. (2021). How do People Judge Risk? Availability may
Upstage Affect in the Construction of Risk Judgments. Risk
analysis: an official publication of the Society for Risk Analysis.
Advance online publication. https://doi.org/10.1111/risa.13729
Finucane, M. L., Alhakami, A., Slovic, P., & Johnson, S. M.
(2000). The affect heuristic in judgments of risks and benefits.
Journal of Behavioral Decision Making, 13(1), 1–17. https:
//doi.org/10.1002/(SICI)1099-0771(200001/03)13:1\1::AID-B
DM333.3.0.CO;2-S
Fischhoff, B., Slovic, P., Lichtenstein, S., Read, S., & Combs, B.
(1978). How safe is safe enough? A psychometric study of attitudes towards technological risks and benefits. Policy Sciences,
9(2), 127–152. https://doi.org/10.1007/BF00143739
Ganzach, Y. (2000). Judging risk and return of financial assets.
Organizational Behavior and Human Decision Processes, 83(2),
353–370. https://doi.org/10.1006/obhd.2000.2914
Judd, C. M., Westfall, J., & Kenny, D. A. (2012). Treating stimuli
as a random factor in social psychology: A new and comprehensive solution to a pervasive but largely ignored problem.
Journal of Personality and Social Psychology, 103(1), 54–69.
https://doi.org/10.1037/a0028347
Kahneman, D. (2003). A perspective on judgment and choice—
Mapping bounded rationality. American Psychologist, 58(9),
697–720. https://doi.org/10.1037/0003-066x.58.9.697
Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus, and
Giroux.
Keller, C., Siegrist, M., & Gutscher, H. (2006). The role of the
affect and availability heuristics in risk communication. Risk
Analysis, 26(3), 631–639. https://doi.org/10.1111/j.1539-692
4.2006.00773.x
King, J., & Slovic, P. (2014). The affect heuristic in early judgments of product innovations. Journal of Consumer Behaviour,
13(6), 411–428. https://doi.org/10.1002/cb.1491
Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B. (2017).
lmerTest package: Tests in linear mixed effects models. Journal
of Statistical Software, 82(1), 1–26. https://doi.org/
10.18637/jss.v082.i13
LeBel, E. P., McCarthy, R. J., Earp, B. D., Elson, M., & Vanpaemel, W. (2018). A unified framework to quantify the credibility
of scientific findings. Advances in Methods and Practices in
Psychological Science, 1(3), 389–402. https://doi.org/10.1177/
2515245918787489
LeBel, E. P., Vanpaemel, W., Cheung, I., & Campbell, L. (2019).
A brief guide to evaluate replications. Meta-Psychology, 3,1–9.
https://doi.org/10.15626/MP.2018.843
Lerner, J. S., Li, Y., Valdesolo, P., & Kassam, K. S. (2015). Emotion and decision making. Annual Review of Psychology, 66(1),
799–823. https://doi.org/10.1146/annurev-psych-010213-115043
Litman, L., Robinson, J., & Abberbock, T. (2017). TurkPrime
.com: A versatile crowdsourcing data acquisition platform for

1183
the behavioral sciences. Behavior Research Methods, 49(2),
433–442. https://doi.org/10.3758/s13428-016-0727-z
Loewenstein, G. F., Weber, E. U., Hsee, C. K., & Welch, N.
(2001). Risk as feelings. Psychological Bulletin, 127(2),
267–286. https://doi.org/10.1037/0033-2909.127.2.267
MacGregor, D. G., Slovic, P., Dreman, D., & Berry, M. (2000).
Imagery, affect, and financial judgment. The Journal of Psychology and Financial Markets, 1(2), 104–110. https://doi.org/
10.1207/S15327760JPFM0102_2
Pachur, T., Hertwig, R., & Wolkewitz, R. (2014). The affect gap
in risky choice: affect-rich outcomes attenuate attention to
probability information. Decision, 1(1), 64.
Peters, E., Västfjäll, D., Gärling, T., & Slovic, P. (2006). Affect
and decision making: A ‘‘hot’’ topic. Journal of Behavioral
Decision Making, 19(2), 79–85. https://doi.org/10.1002/
bdm.528
Rachlin, H. (2003). Bounded rationality: The adaptive toolbox
(Vol. 79). MIT Press.
Raue, M., D’Ambrosio, L. A., Ward, C., Lee, C., Jacquillat, C.,
& Coughlin, J. F. (2019). The influence of feelings while driving regular cars on the perception and acceptance of selfdriving cars. Risk Analysis, 39(2), 358–374. https:
//doi.org/10.1111/risa.13267
Russell, J. A. (2003). Core affect and the psychological construction of emotion. Psychological Review, 110(1), 145–172. https:
//doi.org/10.1037/0033-295X.110.1.145
Schwarz, N. (2012). Feelings-as-information theory. In P. A. M.
Van Lange, A. W. Kruglanski, & E. T. Higgins (Eds.), Handbook of theories of social psychology (pp. 289–308). SAGE.
Schwarz, N., & Clore, G. L. (1983). Mood, misattribution,
and judgments of well-being: Informative and directive functions of affective states. Journal of Personality and Social
Psychology, 45(3), 513–523. https://doi.org/10.1037/0022-3514
.45.3.513
Siegrist, M., & Sutterlin, B. (2014). Human and nature-caused
hazards: The affect heuristic causes biased decisions. Risk
Analysis, 34(8), 1482–1494. https://doi.org/10.1111/risa.12179
Skagerlund, K., Forsblad, M., Slovic, P., & Västfjäll, D. (2019).
The affect heuristic and risk perception—Stability across elicitation methods and individual cognitive abilities. PsyArXiv
[Preprint]. https://doi.org/10.31234/osf.io/mpvu8
Skagerlund, K., Forsblad, M., Slovic, P., & Västfjäll, D. (2020).
The affect heuristic and risk perception—Stability across elicitation methods and individual cognitive abilities. Frontiers in
Psychology, 11, Article 970. https://doi.org/10.3389/fpsyg
.2020.00970
Slovic, P. (1987). Perception of risk. Science, 236(4799), 280–285.
https://doi.org/10.1126/science.3563507
Slovic, P., Finucane, M., Peters, E., & MacGregor, D. G. (2002).
Rational actors or rational fools: Implications of the effects
heuristic for behavioral economics. Journal of Socio-Economics,
31(4), 329–342. https://doi.org/10.1016/S1053-5357 (02)00174-9
Slovic, P., Finucane, M. L., Peters, E., & MacGregor, D. G.
(2007). The affect heuristic. European Journal of Operational
Research, 177(3), 1333–1352. https://doi.org/10.1016/j.ejor
.2005.04.006
Slovic, P., & Västfjäll, D. (2010). Affect, moral intuition, and risk.
Psychological Inquiry, 21(4), 387–398. https://doi.org/10.1080/
1047840X.2010.521119

1184
Thaler, R., & Sunstein, C. R. (2008). Nudge: Improving decisions
about health, wealth, and happiness. Yale University Press.
Västfjäll, D., Peters, E., & Slovic, P. (2014). The affect heuristic,
mortality salience, and risk: Domain-specific effects of a natural disaster on risk-benefit perception. Scandinavian Journal of
Psychology, 55(6), 527–532. https://doi.org/10.1111/sjop.12166
Vuorre, M., & Bolger, N. (2018). Within-subject mediation analysis for experimental data in cognitive psychology and neuroscience. Behavior Research Methods, 50(5), 2125–2143. https:
//doi.org/10.3758/s13428-017-0980-9
Wyer, R. S., Clore, G. L., & Isbell, L. M. (1999). Affect and information processing. Advances in Experimental Social Psychology, 31, 1–77. https://doi.org/10.1016/S0065-2601(08)60271-3
Yang, Z. J., Aloe, A. M., & Feeley, T. H. (2014). Risk information seeking and processing model: A meta-analysis. Journal of
Communication,
64(1),
20–41.
https://doi.org/10.1111/
jcom.12071

Social Psychological and Personality Science 13(7)
Subramanya Prasad Chandrashekar recently completed
a research assistant professor position with the Lee Shau
Kee School of Business and Administration at the Hong
Kong Metropolitan University. His research focuses on
social status, lay-beliefs, and judgment and decision-making.
Cheong Shing Lee is a student at the University of Hong
Kong during the academic year 2019–2020.
Lok Yan Yeung is a student at the University of Hong
Kong during the academic year 2019–2020.
Min Ji Kim is a student at the University of Hong Kong
during the academic year 2019–2020.
Ching Yee Lee is a student at the University of Hong
Kong during the academic year 2019–2020.

Author Biographies
Emir Efendić is a postdoctoral scholar at the School of
Business and Economics in Maastricht University in the
Netherlands. His research focuses on judgment and decision-making.

Gilad Feldman is an assistant professor with the
University of Hong Kong psychology department. His
research focuses on judgment and decision-making.
Handling Editor: Lissa Libby

