Loss-aversion or loss-attention: The impact of losses on cognitive performance

Full text

Loss-aversion or loss-attention: The impact
of losses on cognitive performance
Eldad Yechiam
a,
⇑
, Guy Hochman
b
a
Max Wertheimer Minerva Center for Cognitive Studies, Faculty of Industrial Engineering and Management,
Technion – Israel Institute of Technology, Haifa 32000, Israel
b
Fuqua School of Business, Duke University, Durham, NC 27708, United States
article info
Article history:
Accepted 19 December 2012
Available online 19 January 2013
Keywords:
Negativity bias
Loss aversion
Attention
Decision making
Performance
abstract
Losses were found to improve cognitive performance, and this has
been commonly explained by increased weighting of losses com-
pared to gains (i.e., loss aversion). We examine whether effects of
losses on performance could be modulated by two alternative pro-
cesses: an attentional effect leading to increased sensitivity to task
incentives; and a contrast-related effect. Empirical data from five
studies show that losses improve performance even when the
enhanced performance runs counter to the predictions of loss aver-
sion. In Study 1–3 we show that in various settings, when an advan-
tageous option produces large gains and small losses, participants
select this alternative at a higher rate than when it does not produce
losses. Consistent with the joint influence of attention and contrast-
related processes, this effect is smaller when a disadvantageous
alternative produces the losses. In Studies 4 and 5 we find a positive
effect on performance even with no contrast effects (when a similar
loss is added to all alternatives). These findings indicate that both
attention and contrast-based processes are implicated in the effect
of losses on performance, and that a positive effect of losses on per-
formance is not tantamount to loss aversion.
Ó2013 Elsevier Inc. All rights reserved.
1. Introduction
Although they come in different shapes and flavors (e.g., taxes, deductions, payments), losses are
inherent in many of the decisions we make. In this study, we examine how losses affect the execution
0010-0285/$ - see front matter Ó2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.cogpsych.2012.12.001
⇑
Corresponding author. Fax: +972 4 829 5688.
E-mail address: yeldad@tx.technion.ac.il (E. Yechiam).
Cognitive Psychology 66 (2013) 212–231
Contents lists available at SciVerse ScienceDirect
Cognitive Psychology
journal homepage: www.elsevier.com/locate/cogpsych

of these decisions and their resulting effect on cognitive performance. In tasks ranging from simple
economic decisions to metaperceptions, previous studies have generally shown positive effects of
losses on performance (e.g., Bereby-Meyer & Erev, 1998; Costantini & Hoving, 1973; Dawson, Gilovich,
& Regan, 2002; Denes-Raj & Epstein, 1994; Haruvy & Erev, 2002; Maddox, Baldwin, & Markman, 2006;
Pope & Schweitzer, 2011; Saguy & Kteily, 2011; Yechiam & Ert, 2007). For example, in an early study
Costantini and Hoving (1973) found that the development of response inhibition among second grad-
ers was faster when tokens were removed upon making errors, than when tokens were added for suc-
cesses. Maddox et al. (2006) found similar results for the performance of adults in a complex
categorization task (see Yechiam & Hochman, in press). An independent line of research examined
performance in decision tasks (e.g., Bereby-Meyer & Erev, 1998; Denes-Raj & Epstein, 1994). Ber-
eby-Meyer and Erev (1998) coined the term the ‘‘successful loser’’ effect to denote the positive effect
of losses on decision performance. For instance, in Haruvy and Erev (2002), adult participants were
required to repeatedly select between two choice alternatives. In the Loss condition, one alternative
produced 10 tokens and the other 11 tokens (with certainty). In the Gain condition, these same to-
kens were presented as gains: one alternative produced 10 tokens and the other 11 tokens. Even in
this very simple task, participants converged much faster to the better choice in the condition where
payoffs were presented as losses.
On the other hand, some studies (e.g., Slovic, Finucane, Peters, & MacGregor, 2002; Thaler, Tversky,
Kahneman, & Schwartz, 1997) have demonstrated a reverse effect of losses. These studies also focused
on decision performance. For example, Slovic et al. (2002) gave participants a choice between a sure
outcome (of $2 or $4) and a lower expected value gamble. In the Gain condition, the gamble produced
a 7/36 chance to win $9, and $0 otherwise. In the Loss condition, the gamble produced an additional
loss of 5 cents with 29/36 chance. Paradoxically, more choices were made from the disadvantageous
gamble in the condition where it included a loss. Thus, losses appeared to have ‘‘confused’’ partici-
pants into selecting the disadvantageous gamble.
Our goal in the current study is to contrast the predictions of different process models account-
ing for the effect of losses on cognitive performance. The most well known explanation for the ef-
fect of losses on performance is that performers are driven to avoid potential losses implicated in
failure because of loss aversion (Kahneman & Tversky, 1979), the notion that losses have greater
subjective weight than equivalent gains. We examine whether positive effects of losses on perfor-
mance could be driven by processes other than differences in weighting, especially the effect of
losses on task attention (Taylor, 1991; Yechiam & Hochman, in press) and contrast-related effects
(Slovic et al., 2002). For this purpose, we examine conditions where the positive effects of losses
on performance implied by these processes are inconsistent with loss aversion. In addition, we
examine whether understanding the relative influence of these different processes can shed light
on the apparently contradictory findings concerning the effect of losses on performance noted
above.
In most of the studies that have examined this issue, the (positive or negative) effect of losses on
performance was primarily attributed to loss aversion (e.g., Baumeister, Bratslavsky, Finkenauer, &
Vohs, 2001; Bereby-Meyer & Erev, 1998; Erev & Barron, 2005; Hossain & List, 2012; Thaler et al.,
1997). For example, Haruvy and Erev’s (2002) results can be explained by the fact that a loss of
11 tokens looms larger than a gain of 11. Consequentially, participants’ tendency to avoid 11
was stronger than their tendency to approach 11 (under loss aversion the subjective difference be-
tween two losses is also larger than between two gains). Similarly, in Thaler et al.’s (1997) study of
investment portfolios, participants invested little in a fund with a mean expected return of 1% which
yielded occasional losses and instead preferred investing in a lower risk fund with a mean return of
0.25% in which returns were all positive. Thaler et al. (1997) argued that the aversion to losses won
over participants’ desire to maximize their outcomes. The asymmetric effect of losses compared to
gains is ‘‘deemed axiomatic in the most influential theories of human decision-making’’ (Hacken-
berg, 2009) and has been treated as the most likely explanation for the effect of losses on cognitive
performance (e.g., Bereby-Meyer & Erev, 1998; Hackenberg, 2009; Hossain & List, 2012; Pope &
Schweitzer, 2011).
However, we recently proposed an alternative account (Yechiam & Hochman, in press) based on
attentional processes. Our model suggests that losses increase the overall attention allocated to the
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 213

situation, and the modulation of behavior by task payoffs.
1
Findings indeed show that losses trigger
autonomic arousal, as evidenced by increased pupil diameter and heart rate following losses compared
to respective gains (Hochman & Yechiam, 2011; Löw, Lang, Smith, & Bradley, 2008; Satterthwaite et al.,
2007). These effects of losses on autonomic arousal were obtained even in the absence of loss aversion
(Hochman, Glöckner, & Yechiam, 2010; Hochman & Yechiam, 2011). The mere increase in attention is
well known to positively affect performance under restricted conditions, which include low initial (base-
line) level of attention (i.e., the Yerkes-Dodson rule; Kahneman, 1973; Watchell, 1967; Yerkes & Dodson,
1908) and a requirement to encode rather than merely retrieve information (Craik, Govoni, Naveh-Ben-
jamin, & Anderson, 1996).
The attention-based model of losses can be formally stated in a simple form using Luce’s choice
rule (Yechiam & Hochman, in press). Under Luce’s rule the probability of selecting strategies is a func-
tion of their expectancies, representing the outcomes predicted upon selecting them, and random
noise (Luce, 1959; see also Daw, O’Doherty, Dayan, Seymour, & Dolan, 2006):
P½j¼ e
hE
j
P
j
e
hE
j
;ð1Þ
Specifically, the probability (P) of selecting a strategy jis a function of the distance between its expec-
tancy (E
j
) and the expectancy of other available strategies, but it is also affected by random noise. The
Parameter hcontrols the sensitivity of the choice probabilities to the expectancies. As hincreases the
likelihood of basing one’s strategy on the expectancies increases. The attention-based account predicts
that hwould be larger for tasks involving losses than for equivalent tasks with no losses. This argu-
ment can be directly contrasted with the assumption that the asymmetry between gains and losses
is in the relative weight of gains and losses on the expectancies. The loss aversion account and the
attention-based account do not necessarily imply completely different models. Rather, the two ac-
counts differ in the component process assumed to be affected by losses. Under loss aversion this com-
ponent involves the translation of objective outcomes into subjective valences. Under the attention-
based account losses reduce random noise and increase the sensitivity of choices to the incentive
structure of the task.
Note that while most of the findings reviewed above showing a positive effect of losses on perfor-
mance (e.g., Bereby-Meyer & Erev, 1998; Haruvy & Erev, 2002; Hossain & List, 2012) have been attrib-
uted to loss aversion, they could also be explained by an attentional effect of losses. These studies used
a choice task involving a disadvantageous option producing losses, and in this case avoiding losses and
paying attention to the task are both expected to result in more successful performance. The atten-
tional model is inconsistent, however, with the findings of Thaler et al. (1997), who demonstrated that
when losses were part of the advantageous option participants still avoided losses and chose disad-
vantageously. However, Thaler et al.’s (1997) research design was recently criticized as it confounded
the availability of losses with the size the outcomes (see extensive review in Erev, Ert, & Yechiam,
2008).
On top of loss aversion and the attention-based model, a third account attributes the effect of
losses to the specific case of selecting among gambles involving both losses and gains (Slovic
et al., 2002). Specifically, according to this account, choice alternatives contrasting small losses with
large gains appear more attractive. For example, norm theory (Kahneman & Miller, 1986) postulates
that stimuli are evaluated compared to the norm of their class. Hence, without losses an alternative
is presumably in the ‘‘gain’’ class and may be viewed as a mediocre instance relative to the set of
all positive outcomes. The addition of a minor loss moves the outcome into a mixed loss–gain do-
main, and relative to members of the mixed outcomes class the same gain may seem like a more
attractive instance. A related explanation made by Slovic et al. (2002) is that losses introduce an
1
Note that the proposed attention-based model is somewhat similar to the affective mapping account. The difference between
them is that we posit that the attentional effect of losses is global, such that it enhances the contrasts between payoffs from the
entire set of outcomes and not only for the alternative producing the losses.
214 E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231

affective contrast between outcomes produced by a choice alternative, and if the loss is small en-
ough this contrast can amplify the positive part of the gamble, thus increasing its overall attractive-
ness. Supporting this claim, Slovic et al. showed that a gamble producing 7/36 to win $9 is ranked
higher when it also produces minimal losses. For conciseness, we shall refer to these accounts as
contrast-based models.
Importantly, Slovic et al.’s (2002) findings imply that contrast-related effects of losses emerge even
at the absence of loss aversion. In addition, losses in their study had a negative effect on performance,
which can only be predicted by the contrast-based model and not by the attention-based model. This
might suggest that contrast-related effects are stronger than attentional ones. Still, Slovic et al. (2002)
only meant to demonstrate the contrast effect and did not prospectively evaluate the predictions of
the contrast-based model with those implied by the attentional model or by loss aversion. Thus, while
the three proposed accounts for the effect of losses seem plausible, to our knowledge no study has
systematically compared the predictions of loss aversion with those implied by the attentional and
contrast-based models.
1.1. Comparing the different accounts
In five studies, we examine whether even in the absence of loss aversion, effects of losses on per-
formance can be produced merely by attentional and contrast-related processes. Hence, our aim was
not to test the existence of loss aversion but rather the stronger view that loss aversion is the exclusive
driver of the effect of losses on performance. Since in many domains it is often difficult to set equal
objective magnitudes for gains and losses, in the current studies we focused on simple decision mak-
ing tasks. These kinds of tasks, which control for the probability and magnitude of gains and losses,
allow rigorous comparisons concerning the weighting assigned to each component (Baumeister
et al., 2001). We specifically investigated two main lines of contrasting predictions derived from these
three accounts.
Adding minor losses to one of the choice alternatives. The first line of predictions refers to a situa-
tion where a minor loss is added to one of the alternatives. This is similar to what was done in
Slovic et al. (2002) but they did not systematically vary the expected values of the two alternatives.
By manipulating the relative expected values of the alternative to which a loss is added (see detailed
example in Study 1), we can prospectively derive contrasting predictions implied by the three pro-
cesses. Specifically, if participants are loss averse, then (all things being equal) they should avoid the
alternative producing losses. Accordingly, the loss aversion account predicts that losses increase per-
formance only when a disadvantageous alternative includes losses; since they lead to avoiding this
alternative. By contrast, the attentional model predicts that losses improve decision performance
regardless of whether they are added to the advantageous or disadvantageous alternative. According
to this model, in both cases more attention is allocated to the task, resulting in responses that are
more aligned with the incentive scheme. Finally, the contrast-based model predicts that only an
advantageous alternative that includes a minor loss with a larger gain should lead to enhanced
performance.
Adding similar losses to all alternatives. The second line of predictions refers to a situation where
similar size losses are added all choice alternatives (which otherwise produce only gains). Both loss
aversion and the contrast-based models predict that this should yield no unique effect of losses, as
all alternatives incur the same loss and all gains are contrasted with the same loss. However, under
the attentional model such an addition is expected to increase performance due to the mere increase
in task attention.
To examine these contrasting predictions we conducted five studies. The studies are organized in
accordance with the two sets of research questions. Studies 1–3 address the effect of minor losses
produced by one of the choice alternatives, while Studies 4 and 5 address the effect of similar losses
produced by all alternatives. In both lines of studies we administered two types of tasks, experience-
based tasks in which individuals actually obtain losses and description-based tasks where the
likelihood and magnitude of potential losses are presented to the participants.
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 215

2. Adding minor losses to one of the choice alternatives
2.1. Study 1: The effect of minor losses in experience-based decisions
In this study we examined decision making in four conditions that disentangle the predictions of
loss aversion, the attention-based model, and the contrast-based model. The choice problems and task
conditions are presented in Table 1. In each choice problem there is an advantageous choice alterna-
tive, which has higher expected value, and is denoted as High-EV, and a disadvantageous choice alter-
native, denoted as Low-EV. In the ‘‘Advantageous-losing’’ problem (Problem 1), the High-EV
alternative is the one that produces losses. Specifically, in the Loss condition it produces an equal
chance to obtain either a large gain (200 tokens) or a minor loss (1 token), while in the Gain condi-
tion it produces an equal chance to obtain the same large gain or a minor gain (+1 token).
A positive effect for losses in this setting is not predicted by loss aversion. Namely, under loss aver-
sion people should perform worse in the Loss condition (i.e., select High-EV less) because they would
avoid the possible loss produced by the advantageous alternative (this is also implied under the ex-
pected utility theory assumption of dominance).
In contrast, according to the attention-based model, losses increase the sensitivity to the different
task payoffs. Therefore, since the loss is quite minor, it should enhance the ability to discriminate be-
tween alternatives Low-EV and High-EV, leading to more choices from High-EV in the Loss condition
than in the Gain condition. Thus, under the attentional model, losses are assumed to have a positive
effect on performance even where selecting advantageously leads to losses. A similar prediction is
made by norm theory and affective mapping (i.e., contrast models).
Because the loss introduces a contrast between the outcomes associated with the risky alternative,
and the loss is much smaller compared to the gain, the (risky) High-EV alternative is expected to be
more attractive with losses than with no losses. Thus, the Advantageous-losing problem alone does
not disentangle the predictions of the attention-based and contrast-based accounts.
For this purpose we also added a ‘‘Disadvantageous-losing’’ problem, in which the same risky alter-
native is disadvantageous in terms of expected value (see Table 1). In this setting, under contrast-
based models the risky alternative (Low-EV) is still expected to be more attractive in the Loss
Table 1
Outline of the choice problems used in Studies 1 and 2. The top table for each choice problem (Advantageous-losing and
Disadvantageous-losing) presents the task payoffs in two conditions. Each condition (Loss versus Gain) involved choice between
two alternatives (High-EV versus Low-EV; where EV denotes expected value). This table is followed by the behavioral predictions
of loss aversion, the attentional model, and the contrast-based model. The predictions pertain to the mean choice Pof a choice
option in a specified condition. For example ‘‘P(High-EV, Loss)’’ refers to the predicted mean choice of the High-EV option in the
Loss condition.
Condition Low-EV option High-EV option
Problem 1: Advantageous-losing
Loss 35 with certainty 1 with probability 0.5, 200 otherwise (EV = 100.5)
Gain 35 with certainty 1 with probability 0.5, 200 otherwise (EV = 100.5)
Model Predictions
Loss aversion P(High-EV, Loss) < P(High-EV, Gain)
Attention P(High-EV, Loss) > P(High-EV, Gain)
Contrast P(High-EV, Loss) > P(High-EV, Gain)
Condition Low-EV option High-EV option
Problem 2: Disadvantageous-losing
Loss 1 with probability 0.5, 200 otherwise (EV = 100.5) 135 with certainty
Gain 1 with probability 0.5, 200 otherwise (EV = 100.5) 135 with certainty
Model Predictions
Loss aversion P(High-EV, Loss) > P(High-EV, Gain)
Attention P(High-EV, Loss) > P (High-EV, Gain)
Contrast P(High-EV, Loss) < P(High-EV, Gain)
216 E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231

compared to the Gain condition, since it contrasts a large gain with a small loss. Therefore, losses are
expected to impair performance. By contrast, under the attentional model, losses should result in few-
er choices from the Low-EV alternative because they increase the sensitivity to the payoff structure. If
both processes are evident then since they are assumed to work in opposite directions an interaction is
expected. Namely, the positive effect of losses on performance should be higher in the ‘‘Advantageous-
losing’’ problem than in the ‘‘Disadvantageous-losing’’ problem.
In Study 1 we examined these two problems using experience-based decision tasks in a form sim-
ilar to that used by Haruvy and Erev (2002). In this type of task the participant is not provided with full
descriptions of the outcome distributions, but rather has to learn them by making choices and receiv-
ing feedback (see review in Rakow & Newell, 2010). As repeated measures are provided for each per-
former in each choice problem this enhances statistical power for evaluations using quantitative
models, which we present following our main analysis. The participants’ outcomes were generated
by randomly sampling from the outcomes of Problems 1 and 2 (see Table 1) on each of 100 trials.
In order to reduce the transparency of the task, a noise factor ranging from 5 to 5 (rounded to the
closest integer) was randomly drawn on each trial and added to the constant outcome (35 or 135)
in all conditions.
2.1.1. Method
2.1.1.1. Participants. One-hundred and twenty-two Technion students (63 males and 59 females) took
part in the study after responding to ads asking for participation in a paid experimental study. Fifty-
seven participants performed the Advantageous-losing problem and 65 performed the Disadvanta-
geous-losing problem. All participants received a participation fee of NIS 20 as well as an additional
amount based on their performance.
2.1.1.2. Measure and apparatus. The experimental task involved making 100 repeated selections be-
tween choice options that appeared as virtual buttons. It was presented on 19-in. computer screens
(button sizes were 0.7 1.4 in.). Button clicking was performed using a standard computer mouse.
Upon pressing a button with the mouse, the image of the button changed to a ‘‘pressed’’ form. The
two buttons were labeled only as A and B. The participants received no prior information about the
payoff distributions or the number of trials. The allocation of alternatives Low-EV and High-EV to but-
tons A and B was randomized for each participant, but was kept constant throughout the 100 trials.
Each choice was followed by a realization of the selected alternative, which was randomly drawn from
the relevant distributions described above. Two types of feedback immediately followed each choice:
(1) The basic payoff for the chosen and unchosen alternative,
2
which appeared on each button for 2 s,
and (2) an accumulating payoff counter, which was displayed constantly. The dependent variable was
the proportion of High-EV selections across trials.
2.1.1.3. Procedure. Participants sat in cubicles divided by partitions (4–6 participants were tested at a
time). The allocation to the Gain/Loss conditions was random. Due to this random mechanism for the
Advantageous-losing problem, 29 students (55% male) were allocated to the Gain condition and 28
students (50% male) were allocated to the Loss condition. The participants performing the Disadvan-
tageous-losing problem were likewise randomly allocated to the Gain condition (34 participants, 53%
male) and Loss condition (31 students 48% male). There were no significant differences in age between
conditions (the average age in all conditions was 25).
Participants in all conditions received the following written instructions: ‘‘In this experiment you
will perform a decision making task. Your basic payoff is NIS 20. Additionally, you will earn NIS 1 for
every 1000 game points. In the presented window you will immediately see two buttons, A and B.
Your task is to select between buttons by pressing them. You can press a button several times repeat-
edly (as much as you wish) or switch between buttons (as you wish). The payment for your selection
will appear on the button you have chosen and under the two buttons. Also, in each trial you would be
able to see the results from the unselected button on the button you did not press. Your accumulating
2
Foregone payoffs were added in order to reduce noise due to early convergence to local optima (Denrell, 2007).
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 217

payoff will appear at the bottom of the screen. Please notice: The outcome obtained after each selec-
tion is affected only by the last selection and not by your previous choices (there is no dependency
between rounds)’’.
Three participants who performed the Advantageous-losing problem (two in the Gain and one in
the Loss condition) selected the same button (i.e., choice alternative) throughout the entire 100 trials.
Possibly, these individuals ignored the payoff structure all together, and they were thus excluded from
the analysis.
2.1.2. Results
The participants’ learning curves appear in Fig. 1. In the Advantageous-losing problem, losses led to
more selections from the advantageous alternative. As can be seen, starting from the second block of
trials, the rate of selections from the High-EV option was higher in the Loss condition than in the Gain
condition. In the Disadvantageous-losing problem, losses also led to more selections from the advan-
tageous alternative, but the effect was weaker than in the Advantageous-losing problem.
Across trials, in the Advantageous-losing problem the average rate of High-EV selections in the
Gain condition was 56.1% (SE = 3.3) while in the Loss condition it was 65.6% (SE = 2.8). In the Disad-
vantageous-losing problem the average rate of High-EV selections in the Gain condition was 62.0%
(SE = 2.8) and in the Loss condition it was 67.9% (SE = 3.8). A repeated measures ANOVA was con-
ducted with trial block (of 25 trials) as a within-subjects factor and choice problem (Advantageous-
losing versus Disadvantageous-losing) and condition (Gain versus Loss) as between-subjects factors.
The analysis revealed a main effect of choice problem (F(1, 115) = 64.4, p< .001) but not of condition
Fig. 1. Study 1 results: average proportion of selections from the advantageous High-EV option in four blocks of 25 trials, in the
Gain and Loss conditions. Top: Problem 1 (Advantageous losses). Bottom: Problem 2 (Disadvantageous losses).
218 E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231

(F(1,115) = 0.30, p= .58). Also, there was a significant interaction between the experimental condition
and trial block (F(3,345) = 2.65, p= .049), denoting the emergence of a positive effect of losses on per-
formance in later trials. Moreover, there was a significant interaction of choice problem and condition
(F(1,115) = 5.69, p= .02), suggesting that the effect of losses on performance was highly contingent on
the choice task, being more prominent in the Advantageous-losing problem.
Post-hoc tests showed that in the Advantageous-losing problem there was a significant difference
between the Gain and Loss conditions (F(1, 55) = 4.75, p= .02). In this choice problem, losses had a par-
adoxical effect of increasing the proportion of selections from the advantageous alternative producing
losses. Examination of specific trial blocks showed that the effect of condition was significant only in
blocks 3 and 4 (F(1,55) = 5.45, p= .02; F(1,55) = 5.74, p= .02, respectively), namely in the second half
of the task. By contrast, the effect of losses in the Disadvantageous-losing problem was not significant
(F(1,60) = 1.57, p= .21). The fact that the positive effect of losses on decision performance was smaller
when the disadvantageous alternative included the contrast is consistent with the joint influence of
attentional processes and contrast effects.
Our interpretation of this result is that in the Advantageous-losing problem contrast and atten-
tional effect were working in the same direction. Thus, in this problem both processes contributed
to enhancing the attractiveness of the advantageous alternative with losses, resulting in the observed
reliable positive effect of losses on performance. By contrast, in the Disadvantageous-losing problem
the contrast and attention-based effects of losses worked in opposite directions. Thus, in this problem
the positive effect induced by attention was smaller.
The results of this experiment cannot be driven only by loss aversion, as in the Advantageous-
losing problem participants behaved as if losses made the risky alternative more attractive. Still, we
could not discard the option that because participants gave greater weight to losses, this resulted in
other effects involved in learning (for instance, enhanced sensitivity to all payoffs, as implied by the
attentional model). To examine this possibility further, we modeled the participants’ trial to trial
choices.
2.1.3. Quantitative modeling
An asymmetric effect of losses compared to gains can be expressed in changes in parameters
reflecting different components of the basic reinforcement learning model. A general reinforcement
learning paradigm introduced by Busemeyer and Myung (1992), called the Expectancy Valence (EV)
model, can capture the possible effects of losses on these different components. This model includes
the essential parameters of most plausible models of experiential tasks (see e.g., Camerer & Ho, 1999;
Denrell, 2007; Erev & Roth, 1998; Worthy, Maddox, & Markman, 2008) and has been specifically val-
idated in experience-based tasks (Yechiam & Busemeyer, 2005, 2008). Like other reinforcement learn-
ing models, it is composed of three rules reflecting the effect of different component processes: First, a
utility function is used to represent the evaluation of outcomes experienced immediately after each
choice. Second, a learning rule is used to form an expectancy for each alternative, which is a summary
score for all past utilities produced by each alternative. Third, a choice rule selects the alternative
based on the comparison of the expectancies.
2.1.3.1. Utility rule. The model assumes that losses and gains could be given different weights by indi-
vidual decision makers. The utility for trial tis denoted u(t), and is calculated as follows:
If xðtÞ<0;uðtÞ¼wjxðtÞj
c
;
If xðtÞ>0;uðtÞ¼ð2wÞxðtÞ
c
:ð2Þ
The term x(t) denotes the amount of money won or lost on trial t,wis a parameter that indicates the
relative weight to losses versus gains, and
c
is a parameter that determines the curvature of the utility
function. Possible values of wwere limited between 0 and 2 (where loss neutrality implies strict aver-
aging of payoffs, namely w= 1).
3
The loss aversion model implies that wwill be larger than 1, reflecting
3
The upper-bound of 2 and the implication that loss neutrality is at w= 1 enables comparing a condition with losses and a
condition with no loss, with no bias in the form of a constant multiplying the utilities, under the assumption of loss neutrality.
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 219

greater weight to losses than to gains. Note that for the small amounts of money used in the present
experiment,
c
= 1 was found to be sufficient (estimation of
c
produced only minor improvements).
2.1.3.2. Learning rule. The term expectancy is used in reinforcement learning as a summary score of
past utilities produced by each alternative. A delta learning rule was used for updating the expectan-
cies (see Busemeyer & Myung, 1992; Sarin & Vahid, 1999). According to this learning rule, the expec-
tancy E
j
(t) for each alternative jon each trial tis updated as follows:
E
j
ðtÞ¼E
j
ðt1Þþ/½uðtÞEjðt1Þ;ð3Þ
where jis a given choice alternative. Since foregone payoffs were administered, the expectancy was
updated for both alternatives simultaneously, assuming equal weight to foregone and obtained pay-
offs (following Erev & Haruvy, in press; Otto & Love, 2010; Yechiam & Rakow, 2011). The learning rate
(or recency) parameter /describes the degree to which the expectancy reflects the influence of the
most recent outcomes or more distant past experiences (0 6/61). The delta learning rule has been
shown to have better fit at the individual level than several alternative models (see e.g., Worthy et al.,
2008; Yechiam & Busemeyer, 2008; Yechiam & Ert, 2007). In this component as well one can assume
an asymmetry resulting from losses, for instance greater recency in a condition with losses.
2.1.3.3. Choice rule. The probability of choosing an alternative is assumed to be a strength ratio of the
expectancy of that alternative relative to all others, using Luce’s rule (see Eq. (1) above), as follows:
P½j;tþ1¼ e
hE
j
ðtÞ
P
j
e
hE
j
ðtÞ
;h¼3
c
1;ð4Þ
The parameter hcontrols the consistency of the choice probabilities and the expectancies. In our anal-
ysis, the final parameter hwas set to 3
c
1, where 0 6c610. A value of cbetween 0 and 10 enables
examining the range between practically random choices (c= 0) and practically deterministic choices
(c= 10). To make the model as simple as possible, the value of hwas kept independent of the trial
number, as in Ahn, Busemeyer, Wagenmakers, and Stout (2008). The attention-based model implies
that the parameter hwould be higher in conditions with losses compared to conditions involving only
gains.
2.1.4. Implementation and results of the modeling analysis
The model was evaluated for its ability to predict ‘one step ahead’ choices on each trial in each of
the experimental conditions. Specifically, the model parameters were estimated separately for each
individual based on the fit of the prediction for trial t+ 1 to the actual choice, using log likelihood
(LL) estimation. The parameter optimization process followed a robust combination of grid-search
and simplex (Nelder & Mead, 1965) search methods (as detailed in Ahn et al., 2008). In the gain con-
dition the weight to loss parameter was redundant, therefore it was not estimated (w= 1).
The fit of the EV model was compared to a baseline model assuming that the choices are generated
by a statistical Bernoulli process (e.g., Busmeyer & Stout, 2002; Gureckis & Love, 2009). According to
this model, for each participant there is a fixed probability of selecting each of the two alternatives,
which is a free parameter of the model (since the task includes two choice options, only one free
parameter is required). The difference between the EV and baseline model fits was corrected according
to the Bayesian Information Criterion (BIC; Schwartz, 1978):
BIC ¼2½LL
EV
LL
Baseline
þklnðNÞ;ð5Þ
where kequals the difference between the EV model and the baseline model in the number of param-
eters and Nis the number of trials. Because this is a test of differences between models, positive BIC
values denote an advantage for the EV model.
The full EV model was superior to the baseline model in the Loss condition (Advantageous-losing:
BIC = 6.3; Disadvantageous-losing: BIC = 3.1) but not in the Gain condition (Advantageous-losing:
BIC = 8.0; Disadvantageous-losing: BIC = 18.9). We therefore proceeded with caution to interpret
the EV model’s estimated parameters. Table 2 presents the mean estimated parameters in the four
220 E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231

experimental conditions. The main results were as follows. First, no loss aversion was found in any of
the two choice problems. The value of the wparameter in the Loss condition was close to 1 in both
problems and it did not deviate significantly from 1.
4
Note that in the Loss condition the EV model
was found to have adequate fit compared to the baseline model; hence, we substantiated the argument
that the positive effect of losses on performance emerged simultaneously with no loss aversion.
Additionally, the choice sensitivity parameter cwas considerably higher in the Loss condition com-
pared to the Gain condition. This effect was highly significant in both choice problems (Advantageous-
losing: t(55) = 2.49, p= .02; Disadvantageous-losing: t(60) = 4.51, p< .001). In the Advantageous-
losing problem losses also had an unexpected positive effect on the participants’ learning rate,
t(55) = 2.40, p= .02. Since the fit of the EV model in the Gain condition was poor, comparisons between
parameter estimates in the Gain and Loss condition should be interpreted cautiously. Still, the pattern
of results suggests that the overall sensitivity to the payoff structure and the learning rate were af-
fected by losses.
2.2. Study 2: The effect of minor losses in description-based decisions
In the first study we focused on experienced losses. Under the attentional model of losses (Yechiam
& Hochman, in press), positive effects of losses on performance should also emerge for potential losses,
such as hypothetical losses in a gamble. By contrast, Slovic et al. (2002) showed that actually in one-
shot hypothetical gambles losses had a negative effect on performance; which they interpreted as the
result of a contrast effect. Hence, combining our results in Study 1 with Slovic et al.’s (2002) findings
one might deduce that the attentional effects of losses are more prominent in experience-based deci-
sions where participants actually obtain losses than in description-based decisions. To evaluate
whether the attentional effect of losses is indeed different in decisions from description, we replicated
Study 1 using one-shot gambles of the sort used by Kahneman and Tversky (1979) and Slovic et al.
(2002).
As in Study 1, we examined whether we find larger effects of losses on performance when losses
are produced by an advantageous alternative, consistent with joint effects of contrast and atten-
tion-based processes. We used the Advantageous-losing and Disadvantageous-losing problems with
the exact payoffs described in Table 1.
2.2.1. Method
2.2.1.1. Participants. A survey was administered to 491 students (226 males and 263 females, two
un-identified) drawn from the pool of experimental study participants at the Technion. From these
participants, 268 performed the Advantageous-losing problem, and 100 participants performed the
original Disadvantageous-losing problem (see Table 1), while 123 participants performed a slightly
modified version of the Disadvantageous-losing problem, as indicated below. The experiment was
Table 2
Estimated parameters of the expectancy valence model for Study 1. Averages and standard errors (in parentheses) of the three
model parameters: weight to losses versus gains (W), recency (/), and choice sensitivity (c).
Problem Condition Parameter
Loss weight (W) Recency ð/ÞChoice sensitivity (c)
Adv. Losing Gain – 0.21 (0.07) 3.06 (0.83)
Loss 1.08 (0.07) 0.46 (0.07)
*
5.50 (0.50)
*
Disadv. Losing Gain – 0.21 (0.05) 3.00 (0.65)
Loss 1.05 (0.06) 0.23 (0.06) 6.40 (0.35)
*
*
p< .05 (comparison of the parameters in the Loss and Gain conditions in each of the two choice problems).
4
Similar findings of loss-neutral wvalues were obtained by Ahn et al. (2008) in their study of two experiential tasks with
asymmetric expected values, as well as in other cognitive modeling analyses (e.g., Busmeyer & Stout, 2002; Wetzels,
Vandekerckhove, Tuerlinckx, & Wagenmakers, 2010; Yechiam & Busemeyer, 2005).
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 221

advertised by an email. A reward of NIS 100 (at the time, about $30) was promised to six participants,
selected by a raffle. Participation was done by following a link and entering the experiment’s website.
The participation rate was 73%.
2.2.1.2. Procedure. Participants were presented with a Qualtrics web-based questionnaire composed of
a single choice problem either in the Gain or the Loss condition (for example, the Gain condition item
in the Advantageous-losing problem appears in the appendix). The allocation of participants to the
conditions was random. Due to this random mechanism for the Advantageous-Problem, 147 students
(50% males) were allocated to the Gain condition and 121 (49% male) were allocated to the Loss con-
dition. For the Disadvantageous-losing Problem, 42 students (52% male) were randomly allocated to
the Gain condition and 58 students (52% male) to the Loss condition. There were no significant differ-
ences in age between conditions (the average age in all conditions was 25). Six participants were ran-
domly drawn at the end of the experiment and were rewarded as promised.
2.2.2. Results
In the Advantageous-losing problem, the average proportion of selections from the High-EV option
in the Gain condition was 56.4% (SE = 4.7), while in the Loss condition it increased to 68.6% (SE = 6.2).
In the Disadvantageous-losing problem the average proportion of selections from the High-EV option
in the Gain condition was 87.9% (SE = 4.0) while in the Loss condition it increased to 92.8% (SE = 4.3).
The difference between the two choice problems was statistically significant (
v
2
(1) = 27.08, p< .001),
with more choices from the High-EV option in the Disadvantageous-losing problem where the advan-
tageous alternative was also the safe alternative. Across problems participants made more selections
from the High-EV option in the Loss condition than in the Gain condition (
v
2
(1) = 4.62, p= .02). An
analysis of each separate choice problem showed, however, that the positive effect of losses was sta-
tistically significant in the Disadvantageous-losing problem (
v
2
(1) = 4.14, p= .04) but not in the
Advantageous-losing problem (
v
2
(1) = 0.66 p= .42).
Thus, in description-based decisions as well, we observed that the positive effect of losses on per-
formance was task specific. The finding that the positive effect of losses on performance remained po-
sitive but diminished in the Disadvantageous-losing problem (where losses accompany the low-value
gamble) suggests that while losses led to increased sensitivity to the task payoffs, they also induced a
contrast-based effect. However, an alternative explanation for the results of this experiment is a ceil-
ing effect, due to the participants being highly risk averse in the Disadvantageous-losing problem.
To further examine these two explanations we administered a version of the Disadvantageous-los-
ing problem where the outcome from alternative High-EV was changed from 135 to 105. We expected
that this would reduce the tendency to avoid the risky Low-EV option. This problem was administered
in the same survey-based method as the previous two problems. One-hundred and twenty-three par-
ticipants (43 females and 79 males) were randomly allocated to the Gain and Loss conditions (n= 62,
61, respectively). The results replicated the small but non-significant positive effect of losses on per-
formance. The average proportion of selections from the High-EV option in the Gain condition was
83.6% (SE = 4.1) and in the Loss condition it increased to 88.7% (SE = 4.8), but the difference was not
significant (
v
2
(1) = 0.67, p= .45). This suggests that similarly to what we observed in experience-based
tasks, the positive effects of losses on performance are more prominent in cases where they are pro-
duced by an advantageous alternative.
To summarize, when minor losses were produced by the advantageous alternative, they enhanced
performance. This is consistent with both the attentional and contrast effects of losses. When minor
losses were produced by a disadvantageous alternative, they had a much weaker positive effect on
performance, presumably because the attention-based effects were counter-acted by the contrast-
based effects.
2.3. Study 3: Replication using Slovic et al. (2002) settings
The finding of our previous study showed that when losses were part of a disadvantageous gamble,
they had a weak and non-significant positive effect on performance. This finding is inconsistent with
the results of Slovic et al. (2002) who found a negative effect of losses on performance in description-
222 E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231

based decisions. There are multiple differences between the decision problems used in Study 2 and in
Slovic et al.’s (2002) experiment. For example, our study involved 50:50 outcomes and Slovic et al.
used small probability gains; our study was conducted in the lab whereas Slovic et al. administered
the questionnaire to students on campus and in classes; and the payoffs were different. In order to
better understand the discrepancy, we conducted an experiment using Slovic et al.’s (2002) settings.
Participants performed one of two choice problems, which were administered with gains only and
with the addition of a minor loss:
Condition Low-EV option High-EV option
Problem 3: Similar-EV problem
Gain 36 with probability 7/36 (EV = 7) 8 with certainty
Loss 36 with probability 7/36, 1 otherwise (EV = 6.03) 8 with certainty
Problem 4: Different-EV problem
Gain 36 with probability 7/36 (EV = 7) 16 with certainty
Loss 36 with probability 7/36, 1 otherwise (EV = 6.03) 16 with certainty
In Slovic et al.’s (2002) study, there were two magnitudes for the minor loss (in different experi-
ments) and we used the larger of the two. The only other difference between Problem 3 and 4 payoffs
and the original payoffs used by Slovic et al. (2002) is that all outcomes were multiplied by four to
reflect the current US dollar – Israeli Shekel conversion rate. Slovic et al. (2002) originally reported
lower performance in the Loss condition for the Similar-EV problem. They also noted that the same
results were obtained for the Different-EV problem, but did not present these findings.
5
Alternatively, under the attention-based account when the expected values of the available options
considerably differ, losses are actually expected to have a positive effect on performance. Hence, we
predicted that in the Different-EV problem losses would lead to more choices from the High-EV option
and enhance performance.
2.3.1. Method
2.3.1.1. Participants. The participants were 104 Technion students (62 males and 42 females). An
experimenter approached participants on campus (as in Slovic et al., 2002) and asked them to volun-
teer to fill in a one-question survey. An equal number of participants were allocated to the four con-
ditions of the study.
2.3.1.2. Measure and apparatus. The task involved selecting between a single pair of options titled
‘‘Alternative A’’ and ‘‘Alternative B’’. The payoffs were presented in the same format as in Study 2. Pay-
offs were based either on the Similar-EV problem or the Different-EV problem (in the Gain condition
or in the Loss condition).
2.3.1.3. Procedure. Participants were randomly allocated to the experimental conditions. For the Gain/
Loss conditions the proportions of males to females were very similar (56% compares to 63% males,
respectively). In the Similar-EV problem there were slightly more males than in the Different-EV prob-
lem (67% compared to 52%;
v
2
(1) = 2.56, p= .11). We therefore added an analysis controlling for gen-
der when comparing the two choice problems.
2.3.2. Results
In the Similar-EV problem the average proportion of selections from the High-EV option was 76.9%
(SE = 8.4) in the Gain condition, and it declined to 46.1% in the Loss condition ((SE = 10.1). This is very
similar to the pattern found by Slovic et al. (2002), which is explained by the contrast-based model. In
the Different-EV problem, however, losses had a reverse effect. The proportion of selections from the
High-EV option (i.e., the safe option) in the Gain condition was 73.1% (SE = 8.8) while in the Loss
5
They merely indicated that ‘‘a replication study with $4 as the alternative to the gamble produced similar results’’ (p. 403).
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 223

condition it increased to 84.6%. Statistical analyses showed a marginally significant performance
advantage for the Different-EV problem compared to the Similar-EV problem (
v
2
(1) = 3.72,
p= .054).
6
Across choice problems, there was no effect for the Gain/Loss condition (
v
2
(1) = 1.14,
p= .28) but a separate analysis of each problem showed that in the Similar-EV problem losses had a sig-
nificant negative effect on performance (
v
2
(1) = 5.20, p= .02), while in the Different-EV problem they
had a non-significant positive effect (
v
2
(1) = 1.04, p= .31).
Hence, our results in the Similar-EV problem replicate those found by Slovic et al. (2002). However,
the results in the Different-EV problem are different: Introducing a pronounced expected value differ-
ence between choice options eliminated the negative effect of losses on performance. These findings
are consistent with those of our previous experiments. In the Similar-EV problem, where the differ-
ences in expected value were minor, the attentional effect of losses presumably did not have any im-
pact on performance. In this case, only the contrast effect impacted the participants’ decisions, leading
to impaired performance with losses. Conversely, in the Different-EV problem, where selecting the
advantageous alternative led to a substantial performance advantage, the attentional effect of losses
impacted performance in addition to the contrast effect. This led to the slight (non-significant) perfor-
mance enhancement with losses.
3. Similar losses produced by all alternatives
3.1. Study 4: Similar losses produced by all alternatives in experience-based decisions
As noted in the introduction, another means of examining the different processes implicated in the
effect of losses on performance, is to have the same or a similar loss incurred by all choice alternatives.
Differently from the design of Studies 1–3 in which the contrast and attention based models have con-
trasting predictions, in this research design, the contrast model has no clear directional prediction,
since all options are contrasted with the same loss. Similarly, loss aversion also does not predict an
effect of losses on performance in this case. Hence, in this setting, only the attention-based model pre-
dicts a positive effect of losses on performance.
The task we designed had two types of choice trials:
Trial Low-EV option High-EV option
Problem 5: Similar losses produced by all alternatives
a 1 with probability 0.5, 200 otherwise
(EV = 100.5)
1 with probability 0.3, 200 otherwise
(EV = 140.3)
bxwith certainty xwith certainty
In each choice trial a random generator determined whether a trial would be of type a or b, and the
participant chose between the High-EV and Low-EV options. In trials of type b the value of xwas set to
5 in the Gain condition and 5 in the Loss condition. Hence, in trials of type b both alternatives pro-
duced the same outcome. Under the attention-based model, losses should enhance performance in
this setting as they increase the sensitivity to payoff. On the other hand, under the contrast-based
model, since the same contrast (between 200 and 5) is induced by losses in both choice options, they
should have no effect. Similarly, under loss aversion since the same loss is sustained from both choice
options, there should be no effect of losses on performance.
3.1.1. Method
3.1.1.1. Participants. Forty-Eight Technion students (24 males and 24 females) took part in the study
after responding to ads asking for participation in a paid experimental study. The participants received
a fixed fee of NIS 10 in addition to their performance-based stipend. Participants were randomly allo-
cated to the Gain and Loss conditions (Gain condition: n= 24, Loss condition: n= 24).
6
Controlling for the slight gender differences between conditions replicated this result (F(1,101) = 3.99, p= .05). The effect of
the Gain/Loss condition in each individual choice problem was also replicated when controlling for gender.
224 E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231

3.1.1.2. Measure and apparatus. The same lab settings were used as in Study 1. The experimental task
involved making 200 repeated selections between choice options that appeared as virtual buttons. The
layout of the experiment and the instructions were as in Study 1. The payoffs were as in Problem 5
above. The actual rate of type b trials was similar in the two conditions: 0.5008 in the Loss condition
and 0.4990 in the Gain condition. The dependent variable was the proportion of selections from the
High-EV option across trials. When analyzing the data the same number of trial blocks as in Study
1 was used (4 blocks), with each block having 50 trials.
3.1.1.3. Procedure. The same procedure was used as in Study 1. The proportion of males and females
were set to 50% in both condition. Six participants made their choices from the very same choice alter-
native throughout the 200 trials. Possibly, these individuals ignored the payoff structure altogether.
Indeed, their aggregated choice pattern was not much different from random choice, with 57% selec-
tions from High-EV throughout all choice trials. We therefore conducted the analysis without these
participants. Interestingly, two of these participants were in the Loss condition and four in the Gain
condition. This anecdotal information is consistent with the argument that participants pay more
attention in the condition with losses.
3.1.2. Results
The participants’ learning curves appear in Fig. 2. The average rate of High-EV selections across all
trials showed only a small advantage for the Loss condition over the Gain condition, with 68.7% com-
pared 63.0% choices from High-EV. However, losses did seem to have a positive effect on performance
at the very first block of trials. To examine the statistical significance of this pattern, the results were
analyzed using a repeated measures ANOVA with trial block (of 50 trials) as a within-subjects factor
and condition (Gain versus Loss) as a between-subjects factor. The analysis showed that the main ef-
fect of condition was not significant (F(1, 40) = 0.60, p= .44). However, the interaction between condi-
tion and trial block was significant (F(3, 46) = 5.06, p= .03). Planned contrast tests showed that at the
first block the difference between conditions was marginally significant in t-test (t(40) = 1.81, p= .08).
In this block, in the Loss condition 63.5% of the choices were from the High-EV option compared to
50.1% in the Gain condition, with performance in the Loss but not in the Gain condition being signif-
icantly better than random choice (Loss: t(21) = 2.37, p= .03; Gain: t(19) = 0.02, p= .98). Thus, losses
seemed to have accelerated learning in the first phases of the task even though they were incurred
equally from all choice alternatives.
3.2. Study 5: Similar losses produced by all alternatives in description-based decisions
We also examined whether having a similar loss incurred from all choice alternatives might im-
prove performance in description-based decisions. This condition was implemented in a somewhat
Fig. 2. Study 4 results: Average proportion of selections from the advantageous High-EV option in four blocks of 50 trials, in the
Gain and Loss condition.
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 225

more realistic tax-base scenario, where tax is deducted from the participant’s gains. Three conditions
were compared: No-Tax, Tax (a constant fraction paid back), and Bonus (a constant fraction added to
the participant’s tally). According to the attention-based model of losses, the implementation of the
tax should facilitate performance (i.e., maximization), a pattern we labeled as the ‘‘tax-max effect’’.
By contrast, contrast-based models do not have a direct prediction in this setting since the tax is im-
posed on all alternatives. Loss aversion actually implies a negative effect of losses on maximization
because the tax imposes a larger loss on the higher expected-value alternative.
To validate the generality of the findings to different payoffs we used a battery of prospects devel-
oped by Holt and Laury (2002). The battery, presented in Table 3, was designed to incorporate a range
of differences in expected value between prospects. Holt and Laury (2002) found a large incentive ef-
fect such that choices were better for real outcomes (from a randomly determined gamble) than for
hypothetical ones. We examined whether presenting losses in the form of tax would have an effect
in the same direction, even though taxes reduce the actual obtained outcomes.
3.2.1. Method
3.2.1.1. Participants. The participants were 105 Technion students (54 males and 51 females) who re-
sponded to ads asking for participation in a paid experimental study. They received a participation fee
of NIS 10 as well as an additional amount based on their performance. Participants were randomly
allocated into the three experimental conditions (No-Tax: n= 34, Tax: n= 38, Bonus: n= 33).
3.2.1.2. Measure and apparatus. The task involved selecting between pairs of prospects. The outcomes
and probabilities are described in Table 3. The order of the prospects was randomized for each partic-
ipant. The items were presented in the exact phrasing shown in Table 3 with the addition of the word
‘‘chance’’ after the probability and the NIS symbol. For example, participants were presented with the
following pair of items: ‘‘Alternative A: 0.1 chance to get 4.00, 0.9 chance to get 3.20, Alternative B:
0.1 chance to get 7.70, 0.9 chance to get 0.20’’ (the symbol denotes NIS). The selection between
prospects was done by pressing the button labeled ‘‘Alternative A’’ or ‘‘Alternative B’’ positioned at the
top of each prospect description. This basic task conforms almost exactly to the set of prospects pre-
sented by Holt and Laury (2002) with the exception that payoffs were in Israeli currency, and that all
outcomes were multiplied by two. In the Tax condition below each pair of items the following text was
added: ‘‘Please notice that 20% of your earnings will be paid as tax to the lab’’. In the Bonus condition
this was changed to ‘‘Please notice that in addition you will get a bonus of 20% of your total payoff’’.
The dependent variable was the proportion of selections from the alternative yielding the higher ex-
pected value, which will be referred to as the High-EV option (Option A in the first 4 rows of Table 3
and Option B in the last 5 rows).
Table 3
The battery of prospects used in Study 5 (based on Holt & Laury, 2002). The two left most columns present the outcomes (in New
Israeli Shekels, NIS). The next column shows the expected value difference between Option A and B. The right most columns
present the data: mean proportion of selections from the High-EV option in the three experimental conditions (choice of A in the
first four gambles and B in the last six).
Option A Option B EV (A–B) P(High-EV)
No-Tax Bonus Tax
0.1 to get 4.00, 0.9 to get 3.20 0.1 to get 7.70, 0.9 to get 0.20 2.34 0.94 0.91 0.92
0.2 to get 4.00, 0.8 to get 3.20 0.2 to get 7.70, 0.8 to get 0.20 1.66 0.88 0.85 0.97
0.3 to get 4.00, 0.7 to get 3.20 0.3 to get 7.70, 0.7 to get 0.20 1.00 0.79 0.88 0.87
0.4 to get 4.00, 0.6 to get 3.20 0.4 to get 7.70, 0.6 to get 0.20 0.32 0.74 0.85 0.74
0.5 to get 4.00, 0.5 to get 3.20 0.5 to get 7.70, 0.5 to get 0.20 0.36 0.24 0.21 0.58
0.6 to get 4.00, 0.4 to get 3.20 0.6 to get 7.70, 0.4 to get 0.20 1.02 0.35 0.33 0.47
0.7 to get 4.00, 0.3 to get 3.20 0.7 to get 7.70, 0.3 to get 0.20 1.70 0.59 0.67 0.82
0.8 to get 4.00, 0.2 to get 3.20 0.8 to get 7.70, 0.2 to get 0.20 2.36 0.91 0.79 0.84
0.9 to get 4.00, 0.1 to get 3.20 0.9 to get 7.70, 0.1 to get 0.20 3.02 0.91 0.88 0.92
1.0 to get 4.00 [0 to get 3.20] 1.0 to get 7.70, [0 to get 0.20] 3.70 0.88 0.94 0.87
226 E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231

3.2.1.3. Procedure. The same lab settings were used as in Study 1. The proportion of males and females
was similar in each condition (50% males in the Tax and No-Tax condition and 48% in the Bonus con-
dition). The participants were given the following written instructions: ‘‘In this study you will be
asked to select between 10 pairs of gambles that you wish to play. Each gamble has two monetary
consequences that are realized with different probabilities. The amounts and probabilities will be pre-
sented on screen. At the end of the experiment, one of the gambles will be randomly selected, and it
will be played. The amount earned will be added to your overall payoff for this experiment’’. In the No-
Tax condition this ended this part of the instruction. In the Tax condition an underlined text further
indicated that ‘‘Please notice that 20% of your earnings will be paid as tax to the lab’’. In the Bonus con-
dition this last sentence was converted to ‘‘Please notice that in addition you will get a bonus of 20% of
your total payoff’’. These one-sentence messages were also presented at the bottom of the screen
describing each pair of prospects (each message in its respective condition, as noted above). The
instructions were followed by an easy example. The participants were then asked if they had any
questions. They then selected between the pairs of prospects.
3.2.2. Results
The average proportions of selections appear in Fig. 3. As indicated in the figure, the rate of selec-
tions from the High-EV option in the Tax condition was higher than in the other two conditions. A one-
way analysis of variance showed that the difference between conditions was significant
(F(2,102) = 5.14, p< .01). Scheffe contrast analyses indicated that the differences between the Tax con-
dition and each of the other two conditions were significant (Tax, No-Tax: p= .02, Tax, Bonus: p= .01,
one tailed).
Table 3 also shows the results for individual items from Holt and Laury’s (2002) battery. As can be
seen, across items there was an advantage to the Tax condition. Maximization rates in this condition
were higher than in the Bonus or Control condition in 7 out of the 10 gambles. Maximization rates in
the Bonus condition were higher than in the Control condition in only 4 out of the 10 gambles. Com-
parison of the Tax condition and the two other conditions using Student’s t-test yielded significant re-
sults for three items (rows 2, 5, and 7 in Table 2;p< .05). By contrast, for the Bonus condition there
was no significant advantage over the remaining two conditions in any of the studied gambles.
4. General discussion
The results of the current studies show that losses have unique effects on performance in decision
tasks, which are not merely a symmetric mirror image of the effect of respective gains. However, the
mechanism leading to these effects is not necessarily as simple as an increased weight of losses
Fig. 3. Comparison of the experimental conditions in Study 5: Average proportion of selections from the High-EV option in the
No-Tax condition, Bonus condition, and Tax condition. The error bars denote the standard errors.
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 227

compared to gains. We examined two sets of new predictions concerning effect of losses on cognitive
performance derived strictly by attention-based and contrast-based processes. Our findings confirm
that effects of losses on performance can be predicted based on these processes, and that these effects
may run counter to the predictions of loss aversion.
4.1. Adding minor losses to one of the choice alternatives
In Studies 1 and 2 we found that losses enhanced the selection of an advantageous choice option,
even though this choice option was the only one that produced losses. The effect of losses in this con-
dition, which was labeled as ‘‘Advantageous losing’’, could not be explained by loss aversion. Rather, it
could be explained by either the attentional or contrast-based model. Formal modeling of trial to trial
choices showed that indeed participants exhibited loss neutrality in this setting, and that losses in-
creased the sensitivity to the entire set of incentives.
To further disentangle the predictions of the attentional and contrast-based models, we examined a
‘‘Disadvantageous losing’’ condition, in which losses are produced by a risky disadvantageous option.
In this condition, under contrast models losses should promote the selection of the risky option,
thereby impairing performance; while under the attention-based model losses should enhance
performance. Our findings in this condition showed that losses still had a positive effect on
performance in experience-based tasks and description-based tasks, though it was much weaker than
in the ‘‘Advantageous losing’’ condition. This interaction suggests that both contrast and attention-
based processes modulate the effect of losses on performance. When these processes both imply a
positive effect of losses on performance, we find a larger effect than when they have counter-acting
influences.
Our results also imply that when the contrast-related and attention-based effects of losses are in
opposite direction, the attentional effect wins by a margin, as evidenced by the positive effect of losses
on performance. This finding (obtained both in Study 1 and 2) appears to contradict those of Slovic
et al. (2002). In order to examine this disparity we used Slovic et al.’s (2002) specific design. We found
that we replicated Slovic et al.’s (2002) results in the condition with almost no expected-value differ-
ence between the alternatives. Apparently, in the absence of a performance benefit to choosing either
one of the alternatives, the contrast-related effect of losses moved decisions away from expected value
maximization. However, in a second condition where the difference in expected values was substan-
tial, we obtained a weak positive effect of losses on performance. This result, implied by the attention-
based model, was consistent with those of Study 1 and 2.
The first three studies we conducted used very different settings: experience-based decisions with
performance based compensation; description-based decisions with compensation based on a raffle;
and description-based decisions with voluntary participation and involving a very different choice
problem. Yet while this might well have impacted the data, in all three studies we obtained the same
results: When alternatives differed in their expected value losses had a positive effect on performance,
and this effect became significant (in Studies 1 and 2) when it was in the same direction as the con-
trast effect.
4.2. Similar losses produced by all alternatives
We also examined whether in the absence of contrast effects, losses would have a positive effect
on decision performance, as implied by attention-based model. In Study 4 we ‘‘planted’’ random tri-
als where losses (or small gains) were emitted by both choice alternatives. As predicted by the
attention-based model, the addition of small losses improved performance, though mostly in the
first block of trials. It appears that losses helped participants to adjust quicker to the task demands.
In Study 5 we examined a more realistic situation using a tax-based scenario where losses were in-
curred as a constant fraction of the participants’ winnings. We found that taxes had a positive effect
on maximization in a decision task, even though greater losses were sustained by making the right
selection.
228 E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231

5. Concluding remarks
Our experimental results suggest that the mechanisms leading to the unique effects of losses on
cognitive performance should be re-evaluated. Indeed, as suggested by Novemsky and Kahneman
(2005) ‘‘...a realistic theory of loss aversion is unlikely to be simple’’ (p. 126). The current findings
are consistent with the theory that losses may be treated as signals of attention and not only as signals
of avoidance. Our results thus complement previous findings showing that losses induce more con-
trolled processing than comparable gains (Dunegan, 1993) and are associated with some of the phys-
iological indices of attention (as reviewed in Yechiam & Hochman, in press). However, our findings
also suggest that the contrast between losses and gains is an additional important factor that moder-
ates the association between losses and performance.
A limitation of our experimental design is that we only examined relatively small nominal losses.
Previously, Slovic et al. (2002) showed that contrast effects were lower as a function of the size of
the loss. Possibly, larger losses may lead to loss aversion (cf. Abdellaoui, Bleichrodt, & Paraschiv,
2007; Rabin & Weizsäcker, 2009). Note, however, that our goal in this paper was not to refute
the existence of loss aversion but rather to refute two stronger arguments: 1) that loss aversion
is the only explanation for the effect of losses on cognitive performance (e.g., Bereby-Meyer & Erev,
1998; Pope & Schweitzer, 2011), and 2) that the sensitivity to losses is a single primitive construct
and whenever there are attentional effects of losses there is loss aversion and vice versa (e.g., Dun-
egan, 1993; Taylor, 1991). Our findings demonstrate that the attentional effect of losses is indeed
distinct from loss aversion, and can lead to behavioral patterns that are contradictory to those im-
plied by loss aversion.
In a broader sense, the non-specificity of the effect of losses to the stimuli that have produced the
losses sheds light on a variety of social phenomena. Specifically, both the attention- and contrast-
based models of losses imply that a negative feature may turn into an advantage when it draws atten-
tion to an overall positive nature of a person or a situation, but the attention-based model suggests
that this is moderated by more global considerations. For example, experiments in Social Psychology
using simulated interviews with job candidates have examined the effect of minor negative features
on the candidates’ evaluation. The results showed that when a generally favorable candidate had some
minor negative occurrence (e.g., spilling his/her cup of coffee), this actually increased the candidate’s
positive evaluation (Beauvois & Dubois, 1988; Nisbett & Bellows, 1977). The attention-based model
suggests, however, that the positive effect of losses in this setting is limited to an overall advantageous
candidate, and should be reversed if a poor candidate presents slightly negative behavior. Under the
joint influence of contrast and attention, the effect of losses is expected to diminish for poor candi-
dates. A similar example in the field of marketing is the ‘‘blemishing effect’’, the finding that a weak
negative feature in a particular product (e.g., a partially broken chocolate bar) improves its attractive-
ness (Ein-Gar, Shiv, & Tormala, 2012). Again, differing from the contrast-based account, the attention-
based model implies that this finding is not general, and should emerge more strongly for products
which most consumers recognize as attractive upon deliberation.
The tax-max effect we observed further suggests that the positive effect of weak negative fea-
tures can be generalized to taxes. Anomalous effects of taxes on the value of products have been pro-
posed for specific products, such as Veblen goods (high status products; e.g., Amaldoss & Jain, 2005).
Our findings suggest that given a similar tax level and a tax-per-earning scheme, taxes also improve
the selection of investment options and therefore increase preference for the most highly taxed
option. We believe that this apparent paradox, which was demonstrated in simple gambles, emerges
simply because taxes, like other forms of losses, have the effect of increasing attention to the task at
hand.
Acknowledgments
This work was supported in part by the Max Wertheimer Minerva Center for Cognitive Studies. Par-
tial results from Study 1 were presented at the 2011 conference on Subjective Probability, Utility, and
Decision Making (SPUDM 23), Kingston, UK.
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 229

Appendix A. The experimental screen in Study 2 (Problem 1, gain condition)
References
Abdellaoui, M., Bleichrodt, H., & Paraschiv, C. (2007). Loss aversion under prospect theory: A parameter-free measurement.
Management Science, 53, 1659–1674.
Ahn, W. Y., Busemeyer, J. R., Wagenmakers, E. J., & Stout, J. C. (2008). Comparison of decision learning models using the
generalization criterion method. Cognitive Science, 32, 1376–1402.
Amaldoss, W., & Jain, S. (2005). Conspicuous consumption and sophisticated thinking. Management Science, 51, 1449–1466.
Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D. (2001). Bad is stronger than good. Review of General Psychology, 5,
323–370.
Beauvois, J.-L., & Dubois, N. (1988). On-line vs memory-based orientations and self-report validity. European Journal of Social
Psychology, 27, 611–616.
Bereby-Meyer, Y., & Erev, I. (1998). On learning to become a successful loser: A comparison of alternative abstractions of
learning processes in the loss domain. Journal of Mathematical Psychology, 42, 266–286.
Busemeyer, J. R., & Myung, I. J. (1992). An adaptive approach to human decision-making: Learning theory, decision theory, and
human performance. Journal of Experimental Psychology: General, 121, 177–194.
Busmeyer, J. R., & Stout, J. C. (2002). A contribution of cognitive decision models to clinical assessment: Decomposing
performance on the Bechara gambling task. Psychological Assessment, 14, 253–262.
Camerer, C., & Ho, T.-H. (1999). Experience-weighted attraction learning in normal form games. Econometrica, 67, 827–874.
Costantini, A. F., & Hoving, K. L. (1973). The effectiveness of reward and punishment contingencies on response inhibition.
Journal of Experimental Child Psychology, 16, 484–494.
Craik, F. I., Govoni, R., Naveh-Benjamin, M., & Anderson, N. D. (1996). The effects of divided attention on encoding and retrieval
processes in human memory. Journal of Experimental Psychology: General, 125, 159–180.
Daw, N. D., O’Doherty, J. P., Dayan, P., Seymour, B., & Dolan, R. J. (2006). Cortical substrates for exploratory decisions in humans.
Nature, 441, 876–879.
Dawson, E., Gilovich, T., & Regan, D. T. (2002). Motivated reasoning and performance on the Wason selection task. Personality
and Social Psychology Bulletin, 28, 1379–1387.
Denes-Raj, V., & Epstein, S. (1994). Conflict between intuitive and rational processing: When people behave against their better
judgment. Journal of Personality and Social Psychology, 66, 819–829.
Denrell, J. (2007). Adaptive learning and risk taking. Psychological Review, 114, 177–187.
Dunegan, K. J. (1993). Framing, cognitive modes, and image theory: Toward an understanding of a glass half full. Journal of
Applied Psychology, 78, 491–503.
Ein-Gar, D., Shiv, B., & Tormala, Z. L. (2012). When blemishing leads to blossoming: The positive effect of negative information.
Journal of Consumer Research, 38, 846–859.
Erev, I., & Haruvy, E. (in press). Learning and the economics of small decisions. In J.H. Kagel & A.E. Roth (Eds.), The Handbook of
Experimental Economics. Princeton, NJ: Princeton University Press.
Erev, I., & Barron, G. (2005). On adaptation, maximization and reinforcement learning among cognitive strategies. Psychological
review, 112, 912–931.
Erev, I., Ert, E., & Yechiam, E. (2008). Loss aversion, diminishing sensitivity, and the effect of experience on repeated decisions.
Journal of Behavioral Decision Making, 21, 575–597.
230 E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231

Erev, I., & Roth, A. E. (1998). Predicting how people play games: Reinforcement learning in experimental games with unique,
mixed strategy equilibria. American Economic Review, 88, 848–881.
Gureckis, T. M., & Love, B. C. (2009). Short-term gains, long-term pains: how cues about state aid learning in dynamic
environments. Cognition, 113, 293–313.
Hackenberg, T. D. (2009). Token reinforcement: A review and analysis. Journal of the Experimental Analysis of Behavior, 91,
257–286.
Haruvy, E., & Erev, I. (2002). Interpreting parameters in learning models. In R. Zwick & A. Rapoport (Eds.), Experimental Business
Research (pp. 285–300). Norwell, MA: Kluwer Academic Publishers.
Hochman, G., Glöckner, A., & Yechiam, E. (2010). Physiological measures in identifying decision strategies. In A. Glockner & C.
Witteman (Eds.), Foundations for tracing intuitions: challenges and methods (pp. 139–159). London: Psychology Press &
Routledge.
Hochman, G., & Yechiam, E. (2011). Loss aversion in the eye and in the heart: The Autonomic Nervous System’s responses to
losses. Journal of Behavioral Decision Making, 24, 140–156.
Holt, C. A., & Laury, S. K. (2002). Risk aversion and incentive effects. American Economic Review, 92, 1644–1655.
Hossain, T., & List, J. A. (2012). The behavioralist visits the factory: Increasing productivity using simple framing manipulations.
Management Science, 58, 2151–2167.
Kahneman, D. (1973). Attention and effort. Englewood Cliffs, NJ: Prentice-Hall.
Kahneman, D., & Miller, D. T. (1986). Norm theory: Comparing reality to its alternatives. Psychological Review, 93, 136–153.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47, 263–291.
Löw, A., Lang, P. J., Smith, J. C., & Bradley, M. M. (2008). Both predator and prey: Emotional arousal in threat and reward.
Psychological Science, 19, 865–873.
Luce, R. D. (1959). Individual choice behavior. NY: Wiley.
Maddox, W. T., Baldwin, G. C., & Markman, A. B. (2006). A test of the regulatory fit hypothesis in perceptual classification
learning. Memory and Cognition, 34, 1377–1397.
Nelder, J. A., & Mead, R. (1965). A simplex method for function minimization. Computer Journal, 7, 308–313.
Nisbett, R. E., & Bellows, N. (1977). Verbal reports about causal influences on social judgments: Private access versus public
theories. Journal of Personality and Social Psychology, 9, 613–624.
Novemsky, N., & Kahneman, D. (2005). The boundaries of loss aversion. Journal of Marketing Research, 42, 119–128.
Otto, A. R., & Love, B. C. (2010). You don’t want to know what you’re missing: When information about forgone rewards impedes
dynamic decision making. Judgment and Decision Making, 5, 1–10.
Pope, D. G., & Schweitzer, M. E. (2011). Is Tiger Woods loss averse? Persistent bias in the face of experience, competition, and
high stakes. American Economic Review, 101, 129–157.
Rabin, M., & Weizsäcker, G. (2009). Narrow bracketing and dominated choices. American Economic Review, 99, 1508–1543.
Rakow, T., & Newell, B. R. (2010). Degrees of uncertainty: An overview and framework for future research on experience-based
choice. Journal of Behavioral Decision Making, 23, 1–14.
Saguy, T., & Kteily, N. (2011). Inside the opponent’s head: Perceived losses in group position predict accuracy in
metaperceptions between groups. Psychological Science, 22, 951–958.
Sarin, R., & Vahid, F. (1999). Payoff assessments without probabilities: A simple dynamic model of choice. Games and Economic
Behavior, 28, 294–309.
Satterthwaite, T. D., Green, L., Myerson, J., Parker, J., Ramaratnam, M., & Buckner, R. L. (2007). Dissociable but inter-related
systems of cognitive control and reward during decision making: Evidence from pupillometry and event-related fMRI.
Neuroimage, 37, 1017–1031.
Schwartz, G. (1978). Estimating the dimension of a model. The Annals of Statistics, 5, 461–464.
Slovic, P., Finucane, M. L., Peters, E., & MacGregor, D. G. (2002). The affect heuristic. In T. Gilovich, D. Griffin, & D. Kahneman
(Eds.), Heuristics and biases: the psychology of intuitive judgment (pp. 397–420). New York: Cambridge University Press.
Taylor, S. E. (1991). The asymmetrical impact of positive and negative events: The mobilization-minimization hypothesis.
Psychological Bulletin, 110, 67–85.
Thaler, R. H., Tversky, A., Kahneman, D., & Schwartz, A. (1997). The effect of myopia and loss aversion on risk taking: An
experimental test. Quarterly Journal of Economics, 112, 647–661.
Watchell, P. L. (1967). Conception of broad and narrow attention. Psychological Bulletin, 68, 417–429.
Wetzels, R., Vandekerckhove, J., Tuerlinckx, F., & Wagenmakers, E.-J. (2010). Bayesian parameter estimation in the expectancy
valence model of the Iowa gambling task. Journal of Mathematical Psychology, 54, 14–27.
Worthy, D. A., Maddox, W. T., & Markman, A. B. (2008). Ratio and difference comparisons of expected reward in decision making
tasks. Memory and Cognition, 36, 1460–1469.
Yechiam, E., & Hochman, G. (in press). Losses as modulators of attention: Review and analysis of the unique effects of losses over
gains. Psychological Bulletin. doi: 10.1037/a0029383.
Yechiam, E., & Busemeyer, J. R. (2005). Comparison of basic assumptions embedded in learning models for experience based
decision-making. Psychonomic Bulletin and Review, 12, 387–402.
Yechiam, E., & Busemeyer, J. R. (2008). Evaluating generalizability and parameter consistency in learning models. Games and
Economic Behavior, 63, 370–394.
Yechiam, E., & Ert, E. (2007). Evaluating the reliance on past choices in adaptive learning models. Journal of Mathematical
Psychology, 52, 75–84.
Yechiam, E., & Rakow, T. (2011). The effect of foregone payoffs on choices from experience: An individual level modeling
analysis. Experimental Psychology, 13, 1–13.
Yerkes, R. M., & Dodson, J. D. (1908). The relation of strength of stimulus to rapidity of habit-formation. Journal of Comparative
Neurology and Psychology, 18, 459–482.
E. Yechiam, G. Hochman / Cognitive Psychology 66 (2013) 212–231 231