Accepted Manuscript

Decision making and age: Factors influencing decision making under

uncertainty

Alec N. Sproten , Carsten Diener , Christian J. Fiebach ,

Christiane Schwieren

PII: S2214-8043(18)30315-X
DOI: 10.1016/j.socec.2018.07.002
Reference: JBEE 371

To appear in: Journal of Behavioral and Experimental Economics

Received date: 7 November 2016

Revised date: 7 July 2018
Accepted date: 9 July 2018

Please cite this article as: Alec N. Sproten , Carsten Diener , Christian J. Fiebach ,
Christiane Schwieren , Decision making and age: Factors influencing decision making under uncer-
tainty, Journal of Behavioral and Experimental Economics (2018), doi: 10.1016/j.socec.2018.07.002

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Highlights
 Two experiments were run to investigate age effects on uncertain decisions
 Uncertainty condition, feedback, and learning requirements were manipulated
 No age differences were found in risky conditions with a priori probabilities
 But well under ambiguity with feedback and under risk with statistical probabilities
 Several explanations for these effects are discussed

Decision making and age: Factors influencing decision making under uncertainty

T
Authors - affiliations:

IP
Alec N. Sproten - University of Erlangen-Nuremberg, Lehrstuhl für Volkswirtschaftslehre,
insb. Wirtschaftstheorie, Lange Gasse 20, 90403 Nürnberg, Germany. Correspondence to:

CR
alec.sproten@fau.de
Carsten Diener - Department of Applied Psychology, Division for Clinical and Biologic
Psychology, SRH University of Applied Sciences Heidelberg, Maria-Probst-Str. 3, 69123
Heidelberg, Germany.
US
Christian J. Fiebach - Department of Psychology, Goethe University Frankfurt, Mertonstr. 17,
60325 Frankfurt am Main, Germany.
AN
Christiane Schwieren - Alfred Weber Institut für Wirtschaftswissenschaften, Heidelberg
University, Bergheimer Str. 58, 69115 Heidelberg, Germany.
M
ED

Abstract:
In the present study, we investigate how decision making under uncertainty is affected by age. We ran
PT

two experiments with young and older adults, systematically manipulating (1) uncertainty conditions
(risk and ambiguity), (2) feedback on decisions and (3) requirements of the task regarding executive
CE

functions. Experiment 1 aims at investigating risk with a priori probabilities and ambiguity and the
effects of feedback in a card game (N = 200; older adults: 97). The results reveal no age differences in
choice behaviour under risk with a priori probabilities. If feedback is provided, we find that older
AC

adults are less ambiguity averse than young adults, whereas there is no significant age difference if no
feedback is provided. Moreover, the presence of feedback had a positive effect on the propensity to
gamble in uncertain conditions by influencing subjective probabilities for both age groups.
Experiment 2 uses the Balloon Analogue Risk Task (BART) to investigate decision making with
statistical probabilities (N = 100; older adults: 50). Here we report older adults being more risk
averse, an effect that can be explained with age differences in sensitivity to prior choices. We support
the results by comparing them to survey data and conclude that differences in uncertainty-processing
exist between young and older adults. Possible explanations of these differences are discussed.

1
 ACCEPTED MANUSCRIPT

Keywords: age differences, experiment, risk, ambiguity, feedback

JEL classification: J14, C91

T
IP
CR
US
AN
M
ED
PT
CE
AC

2
 ACCEPTED MANUSCRIPT

1. Introduction:
“The aging workforce” currently is under much scrutiny in both popular media and scientific
research. A large number of children born after 2000 in western countries could live to 100 years of
age and older – with increasing shares ageing in good cognitive health (Vaupel 2010). Therefore,
many economic decisions will be taken by individuals at higher age, and it becomes more and more
important to understand the decision making of cognitively healthy older adults. While older adults
generally avoid physical risks, face changes in medical risk taking (Hanoch, Rolison & Freund, 2018)
or perceive themselves as less risk seeking while using the internet (White, Gummerum, Wood &

T
Hanoch 2017) assuming the same being true for financial risk taking may be a foregone conclusion

IP
(Samanez-Larkin & Knutson, 2015). Although much research on individual decision making relies on
student populations (Henrich et al. 2010), the number of studies focusing on age differences and life-

CR
span changes in decision making has seen strong increases over the last years, with the current article
contributing to this research.

US
By now, a number of (economic) experiments investigates age effects on decision making by
comparing age groups (see e.g. Best & Charness 2015 for an overview and meta-analysis of studies on
AN
age effects in risky decision making). Older adults are commonly assumed to face changes in decision-
making abilities (Peters et al. 2000) and age differences in decision making under uncertainty are a
central finding in the literature (Mata et al. 2011). Yet, the literature is not always consistent: Whereas
M

some studies report increasing risk aversion with age, other studies find no age differences (e.g.
Carstensen & Hartel 2006; Mata et al. 2011). Many explanations have been drawn to explain these
ED

effects and we will refer to the most widespread explanations below.

The goal of the present paper is to add further evidence to the picture on age effects on decision
PT

making under uncertainty by studying two age groups (younger and older adults) deciding on two
experimental tasks with complementary aspects of uncertainty. We focus on uncertain conditions with
varying information on the probabilities of realizing a positive outcome. Further, we investigate the
CE

impact of feedback on outcomes on behaviour in subsequent decisions, as (a) there is agreement in the
literature that aging implies a decrease in the ability to learn from feedback (e.g. Mathewson et al.,
AC

2008) and (b) feedback is a central tool in information design when it comes to developing decision
aids for older adults. We investigate decision making of elderly participants in both decision making
under risk and decision making under ambiguity. Within both domains, we compare a situation
without feedback and with feedback, where feedback does not allow for learning (as the task is
designed in a way that feedback does not provide any information which can be used to increase
outcomes in subsequent decisions), hence only affecting emotions and subjective uncertainty via a
pathway where feedback may lead to regret for not having made the outcome-maximizing decision or
joy for having just done so. We then conduct a game that adds learning to uncertain conditions,

3
 ACCEPTED MANUSCRIPT

combining all three factors (uncertainty, feedback and learning), where participants have to recur to
feedback to learn about probabilities during the task.

Uncertainty describes a situation in which the outcome of a given condition is bound to some (known
or unknown) probabilities. As early as in the beginning of the 20th century, uncertain conditions were
classified in different subcategories, defined by their “probability situations” (Knight 1921): estimates,
statistical probabilities (sometimes also called decisions from experience), and a priori probabilities
(sometimes also called decisions from description). In current research, it is more common to refer to
‘ambiguity’ instead of estimates and to ‘risk’ instead of statistical and a priori probabilities. Ambiguity

T
(i.e., estimates) is defined as uncertain conditions in which probabilities cannot be computed

IP
empirically because of conflicting or absent information. Statistical probabilities are defined as the
option to learn the risk of a choice empirically by referring to prior choices with similar outcomes.

CR
This constitutes a rather “psychological” definition of risk, as, from an economic perspective, they are
at the border between ambiguity and risk with a priori probabilities. Nevertheless, the concept of

US
statistical probabilities is well defined and choice models can (and will) be estimated (cf. infra). A
priori probabilities define the level of risk by assignment of explicit (or easily computable)
probabilities. We structure the following literature review based on the aspects that interest us: risk vs.
AN
uncertainty, and feedback.

Age and risk

M

Several research groups have shown that risk taking decreases with age (Chaubey 1974; Deakin et al.
2004; Dohmen et al. 2012; Hallahan et al. 2004), whereas other studies have shown that risk
ED

preferences do not change (Ashman et al. 2003; Dror et al. 2000; Henninger et al. 2010; Zamarian et
al. 2008). A number of explanations for this, at the first glance, lack of consistency has been
developed over the years: most centrally the observed effects might reflect differences in the demand
PT

of the tasks used with respect to executive functions. A further explanation might be additional age-
related changes in value assessments (Samanez-Larkin & Knutson 2015) or an inverse-U shaped
CE

relationship between risk preferences and age (see Josef et al. 2016 for self-assessed risk preferences);
other reasons, such as group sampling, may as well play a role, yet to a lesser degree and are much
AC

more difficult to control for between studies. For example, Mata et al. (2011) and Best and Charness
(2015) show in their meta-analyses that older adults are more risk avoidant than young adults when
confronted with statistical probabilities but not when confronted with a priori probabilities, the former
drawing more heavily on executive functions than the latter: when confronted with statistical
probabilities, the ability to learn from prior experiences is crucial (Hertwig & Erev 2009; Hau et al.
2008), while for a priori probabilities, mathematical abilities (i.e. understanding of probabilities) play
an important role. One must however be aware of an additional distinction between such tasks:
conditions with statistical probabilities can be interpreted as conditions that start out with ambiguity

4
 ACCEPTED MANUSCRIPT

and, while exploring, turn more towards risk (while in conditions with a priori probabilities, ambiguity
should be absent from the beginning; Rolison et al. 2012).

Ambiguity

While the effects of age on risky decision making received considerable attention in the literature,
evidence on ambiguity is sparser. Some authors showed that older adults behave similar to young
adults under ambiguity (Tymula et al., 2013), others provide evidence that older adults are more

T
ambiguity averse (Zamarian et al. 2008) or behave less consistently with prior indications of choice

IP
likelihood ratings under ambiguity (Rolison & Parchur 2017). These three studies further stand out
from the literature by assessing decision making in risky and ambiguous conditions in a within subject

CR
design. This is important as older adults may exhibit larger heterogeneity in choice behaviour (due to
e.g. differences in preservation of executive functions or simply life-long experiences) and within-
subject comparisons are a way to avoid finding false positives (or negatives) in the corresponding

US
analyses. Zamarian and colleagues find that normal aging does not affect decision making under risk,
but does affect decision making under ambiguity: Older adults demonstrate less payoff maximizing
AN
behaviour than young adults. One might note that this study clearly reveals differences between
uncertain choices on two tasks, but, from an economic perspective, the measure used to assess
ambiguity, the Iowa Gambling Task (IGT; Bechara et al. 2005), has been under scrutiny for lacking a
M

concise definition of optimal decision making. In addition, the IGT may assess other factors than
ambiguity attitude alone: Learning in a situation of risk with statistical probabilities (as
ED

operationalized by Henninger, Madden & Huettel 2010), differences between “hot” and “cold”
decision making (Buelow & Suhr 2009), or reactions to negative and positive feedback. Tymula et al.
(2013), on the other hand, show that, in a task encompassing risk taking with a priori probabilities and
PT

ambiguity, older adults are more risk averse than young adults, but behave similarly to young adults
under ambiguity.
CE
AC

The role of feedback

Cognitive demands of behavioural tasks can in fact be a driving force behind age differences in
decision making. We therefore try to influence the cognitive demands of the task by studying the
effects of feedback on decisions.

Feedback can have two effects in a situation of risk: (1) influencing emotions and reinforcing
behaviour in general, but it also provides (2) an opportunity to learn. Thus, in contrast to feedback that

5
 ACCEPTED MANUSCRIPT

cannot be used to learn, such as in risk conditions with a priori probabilities, in situations with
statistical probabilities differences in learning abilities between old and young participants can
determine differences in behaviour.

In ambiguous situations, feedback reduces perceived uncertainty (Hogg & Mullin 1999; Kramer
1994), even if it is not informative for future decisions. It also influences emotions and decision
making in general (Baumeister et al. 2007) and has been shown to affect behaviour more strongly in
young than in older adults (Bellucci & Hoyer 1975; Tripp & Alsop 1999). To be more precise, when
investigating dual-process models of decision making, tasks providing immediate feedback on

T
decisions are assumed to assess “hot” decision making (hence with a strong emotional component;

IP
Loewenstein 2000), whereas tasks with delayed feedback are classified as “cold”, inducing little
emotions. To add to the complexity of the problem, older adults seem to use different strategies than

CR
younger individuals in decision making, using more affective strategies where young adults rely more
on deliberative strategies (Wood et al. 2005). Or as Huang, Wood, Berger and Hanoch (2015) put it:

US
declines in deliberative abilities lead to declines in description-based decision making, whereas the
age-related preservation of affective abilities is linked to no age effects in experience-based decisions.
This is particularly important in the light of the “risk as feelings” hypothesis (Loewenstein et al. 2001),
AN
meaning that (a) feedback, by reducing perceived uncertainty, influences the emotional state of
decision makers and (b) older adults´ decision making is more dependent on this emotional state.
According to Huang, Wood, Berger and Hanoch 2013, older adults will rely less than younger adults
M

on “mathematical” decision strategies in tasks carrying some affective component by providing

feedback, such as the task applied in this paper.
ED

The goal of our study is to advance the field towards a better understanding of the relationship
PT

between aging and decision making under financial uncertainty by systematically looking at the
factors (a) risk versus ambiguity, (b) “uninformative” feedback (which cannot be used to learn about
CE

the structure of the task1 and thus only has an “emotional” component) and (c) “informative” feedback
(allowing for learning and being cognitively more demanding). The current study follows closely the
AC

design by Mamerow, Frey and Mata (2016), spanning the afore-listed task differences: self-assessed
risk preferences, risk with statistical and risk with a priori probabilities. The ambiguity conditions of
the task in the current paper mirror more closely Tymula et al.’s task, as the learning requirements are
kept to a minimum. . Note that here, feedback does not carry the opportunity to learn, as this leads to
different results – older adults reacting more stongly to negative feedback (Eppinger & Kray 2011).

1
There is still an opportunity to learn: if, for example, a participant chose to ignore the probabilities of the task
or did not understand the set-up, feedback could be used to learn. For the remainder of the paper, we however
work with the hypothesis that all, or at least the majority, of participants could not learn from feedback.

6
 ACCEPTED MANUSCRIPT

We report results from two experiments in which we observe behavioural differences in uncertain
decisions between young and older adults (Table 1a summarizes the experimental conditions and
Table 1b provides an overview of the hypotheses which will be described below).

In the following, we posit hypotheses about behaviour. We describe our experiments and analyse the
results. Finally, we discuss the findings in the light of the current interest in age differences in decision
making.

2. The Experiments

T
2.1. Basic Design and Hypotheses

IP
In the first experiment, we investigate age differences in behaviour under risk (with a priori

CR
probabilities) and under ambiguity either with or without uninformative feedback on decision
outcomes.

US
As most of the literature suggests differences between young and old adults both in risky and
ambiguous decisions, we hypothesize that older adults have a different propensity to gamble than
young adults in (H1) risky and (H2) ambiguous conditions. The task used to assess these hypotheses is
AN
designed such that feedback can only influence emotions and the perception of probabilities, but does
not provide additional relevant information. Therefore, age differences in learning should not play a
role. As feedback reduces subjective uncertainty (Trope, 1979), even if not informing about objective
M

probabilities, providing feedback on choices with a priori probabilities should result in behaviour
indicative of a smaller distance between subjective and objective probabilities in both age groups
ED

(H3). Assuming that most participants are a priori uncertainty averse, we therefore expect gambling
behaviour to be more frequent, and an estimated probability weighting function to be closer to the
PT

actual probabilities when feedback is provided. Given results that uninformative feedback influences
behaviour more strongly in young than in older adults, and that older adults employ more emotional
strategies for decision making (while assuming that feedback adds an emotional component to tasks
CE

investigating behaviour under uncertainty), two possible outcomes arise: feedback will c.p. increase
older participants´ probability to gamble (H3a) or older adults will demonstrate a smaller difference in
AC

behaviour between feedback and non-feedback treatments (H3b).

Based on the literature reviewed and the methodological reflections given so far, we hypothesize that
whether probabilities are given or have to be learned influences the difference between the age groups.
In the second experiment, we therefore vary this factor: we administer a risk task with statistical
probabilities to our participants (statistical probabilities by definition include feedback on decision
outcomes). Feedback now allows for learning. Since older participants have been shown to learn
slower (Kausler, 1994), we expect changes in behaviour based on feedback rather for young than for
old participants. Learning in this task should in general lead to more risk taking over time (i.e., over

7
 ACCEPTED MANUSCRIPT

the repetition of trials (Mata et al., 2011)). Thus, we expect older participants to take less risk than
young participants (H4a). Alternatively, one could describe a task where probabilities in the beginning
are unknown and can be learned over time as a situation where ambiguity prevails at the onset of the
task, with ambiguity being gradually transformed to risk by gaining experience. Ideally, if such a task
is played for a sufficient number of trials, participants will know the probability structure and end up
in a situation of risk. For age effects this implies that we should expect an increase in the difference in
risk taking between young and older participants over the repetition of trials (H4b). Comparing both
kinds of risk tasks, we would expect that age has a larger effect on behaviour on a task where learning

T
is required. Thus, we expect a difference in risk taking between the two risk tasks within the group of
elderly but not for the young participants (H5). Finally, in order to get an impression of the

IP
representativeness of our results compared to the general population, we also include a widely used

CR
self-assessment question for risk preferences, and compare answers on this question to data from a
representative sample of the German population (Josef et al. 2016).

(a_priori)
Ambig. Risk
(stat.) US
Table 1a : Overview of the Experimental Design and Conditions
)Condition Exp. Risk Feedback
(no learning,
uninformative)
Feedback
(no learning,
informative)
Feedback
(learning)
AN
1 1 Yes Yes
2 1 Yes Yes
3 1 Yes Yes
4 1 Yes Yes
M

5 2 Yes Yes
ED

Table 1b: Overview of the Expected Effects

Expected Effect Hyp. Exp.
Age differences in uncertainty preferences H1, H2, 1, 2
PT

H4
Feedback reduces subjective uncertainty H3 1
- Older adults are less uncertainty averse than young adults H3a
- Older adults are more uncertainty averse than young adults H3b
CE

Statistical probabilities allow for learning through feedback 2

- Older adults are more risk averse than young adults H4a
- Distance between young and older adults increases over time H4b
AC

Difference in risk taking between risk with a priori probabilities and with statistical H5 1, 2
probabilities is more pronounced in older adults

3. Sample, Procedures and Results

3.1. Experiment 1

3.1.1. Participants

8
 ACCEPTED MANUSCRIPT

A total of 200 adults (103 young adults, 97 older adults; cf. Table 2) participated in the experiment.
All of the young adults were students at the Universities of Mannheim or Heidelberg and were on
average 23.30 years old (SD = 2.82, min: 19, max: 32). The older adults were healthy with an average
age of 67.77 years (SD = 5.98, minimum age: 58 years, maximum: 88 years). The majority of the
older adults held a college or university degree and were retired. We recruited them by word of mouth
advertisement at an adult education centre in Heidelberg and by means of an article appearing in a
local newspaper, explaining the need for participants with some computer literacy. Thereby we
generated a group of older adults not representative of the population, but well-matched to the student

T
sample with respect to education and cognitively active lifestyle. Participants’ health was assessed
using an extensive questionnaire on physical and mental disorders, but no cognitive test battery was

IP
applied. We are confident that, if cognitive impairments existed in our participants, they only could

CR
have been mild: all older participants were able to understand the experimental setup, were able to use
a computer, and many were recruited in classes at the adult education centre.

Number
Male/Female
US
Table 2: Participants
Young
103
44/59
Older
97
35/62
AN
Mean Age (SD) 23.30 (2.82) 67.77 (5.98)
Years of education* (SD) 12.70 (1.09) 11.99 (2.41)
*until graduation from school
M

3.1.2. Procedures
ED

In accordance with our hypotheses 1-3, we applied a three-factorial design with two between subject
factors (age group and presence/absence of feedback) and one within-subject factor (uncertainty type:
PT

risk vs. ambiguity) to investigate decision making under uncertainty.

We used a computerized decision making tasks that will be described in detail in the following:
CE

The risk and ambiguity task (RAT) is a computerized card game introduced by Hsu and colleagues
that measures risk behaviour with a priori probabilities and ambiguity behaviour (Hsu et al. 2005).
AC

Participants see card decks on a computer screen. These decks always contain some number of red and
of blue cards. For some decisions, the exact amount of red and blue cards in the deck is revealed to the
participants (see Figure 1), whereas for other decisions participants only know the total number of
cards in the deck. Further, both the uncertain and the sure payoffs are displayed. Participants always
have the choice to either place a bet that a chosen colour would be drawn, or to receive a sure amount
of money. Participants make 48 of these choices between a gamble and a sure amount of money. In
half of the trials, participants are faced with a priori risky decisions (i.e. the probabilities of winning
are given), and in half of the trials they are faced with ambiguous decisions (i.e. the probabilities of

9
 ACCEPTED MANUSCRIPT

winning the gamble are unknown and participants do not receive any information on the underlying
distribution of cards). Risky and ambiguous gambles alternate. Responses are made by selecting an
option in the card game displayed on the screen. When betting, participants receive the indicated
amount of ECU if the colour they chose was drawn (i.e. if a randomly drawn number matched the
probability of the game); otherwise the payoff was zero. When participants choose the certain amount,
they receive the amount automatically, without any influence of the drawn card colour on their gain.

Over all choices card distribution (respectively the total number of cards) and outcome vary.
Probabilities of winning range from 0.1 to 0.9, and payoffs of the gambles vary between € 2.40 and €

T
4.00, certain payoffs range from € 1.00 to € 3.00. Whereas participants receive full feedback on their

IP
performance in the feedback treatments, no feedback is provided in the other treatments. Specifically,
in the feedback treatments, the screen displayed after each choice which colour had been selected by

CR
the participant, if any, which colour was drawn and which amount the participant had won. In the no-
feedback treatments, only the information about participants’ choice, but no information about the

US
drawn colour or the outcome of the gamble was shown. To avoid risk hedging, only two of the
gambles were selected randomly at the end of the game to determine participants’ payoffs.
AN
Participants were allowed as much time as they needed to make their choices.

The experiment took place at the former Collaborative Research Centre 504 Lab (SFB-504) of the
University of Mannheim (young participants) and at the Alfred Weber Institute Lab (AWI-Lab) of
M

Heidelberg University (young and older participants). Participants first had to fill in a general
demographic questionnaire. Subsequently, the experimental tasks by which we examined behaviour
ED

under uncertainty started. Participants received a show-up fee of € 3. Depending on their choices
participants earned on average € 4.80 (S.D. 0.50) in addition to the show-up fee. Although this might
seem little as compared to the possible wealth older adults might dispose of, various authors (e.g.
PT

Camerer and Hogarth, 1999) show that within most economic games, the relative wealth of
participants does not influence choice behaviour. Also, in accordance with Hogarth (2005), we want to
CE

argue that, although in the study of decision making there is a bias towards studying “important”
decisions, the aggregate effects of many “small” decisions might be more important for the quality of
AC

our lives. We therefore are confident that a laboratory experiment with small amounts at stake is a
valid approach to study age differences in decision making.

10
 ACCEPTED MANUSCRIPT

T
IP
CR
US
AN
M

Figure 1: Upper part: Timeline of the game with full feedback. Lower part: Screens presented to
the participants. Left screen: choice between an ambiguous gamble and a sure amount of money.
Right screen: choice between a risky gamble and a sure amount of money. In the original, colors
ED

were red and blue.

3.1.3. Statistical methods and results

PT

In a first parametric analysis propensity to take risky gambles was measured by the number of times
CE

participants chose a gamble ranging from 0 (no gamble chosen at all) to 24 (always chosen the gamble
instead of the sure payoff) instead of a sure amount of money in risk trials. Propensity to take
ambiguous gambles was measured, mutatis mutandis, the same way in ambiguous trials. Descriptive
AC

statistics on the acceptance of gambles can be found in Figure 2 and Table 3a.

11
 ACCEPTED MANUSCRIPT

T
IP
CR
Figure 2: Box-plots of risky and ambiguous gambles split by age group and feedback

Feedback Decision US
Table 3a: Age Effects Split by Feedback
Mean(S.E.)
Young
Mean(S.E.)
Older
AN
Yes Risk 15.706(0.785) 15.673(0.829)
Yes Ambiguity 14.196(0.918) 15.673(1.004)
No Risk 13.019(0.760) 13.688(1.031)
No Ambiguity 8.635(0.841) 14.333(1.128)
M

Hypotheses 1, 2 and 3:
ED

A poisson regression was run to estimate the number of accepted gambles in the RAT based on the age
group, uncertainty condition, and feedback condition and their respective interactions. Ambiguous
PT

conditions and older age interacting with the presence of feedback influence the propensity to gamble
negatively (see Table 3b). Feedback and older age interacting with the presence of ambiguity
influence the propensity to gamble positively. Older age alone has only a marginal (yet positive)
CE

impact on the propensity to gamble and feedback does not interact with the uncertainty condition.

Table 3b: Regression Results for the RAT

AC

Parameter Coefficient S.E. p Par. estimate 95% C.I.

Ambiguity -.301 .090 < .01 .740 .620 - .883
Older age .140 .082 .089 1.150 .979 – 1.351
Feedback .270 .072 < .001 1.309 1.136 – 1.509
Older age * ambiguity .258 .094 < .01 1.294 1.076 – 1.557
Older age * feedback -.218 .095 < .05 .804 .667 - .969
Ambiguity * feedback .121 .097 .211 1.129 .934 – 1.364
Intercept 2.521 .058 < .001 12.444 11.105 – 13.943
N=200; Poisson regression with robust standard errors.

12
 ACCEPTED MANUSCRIPT

In confirmation of the results of the Poisson regression, a nonlinear stochastic choice model (see
Appendix 2 for a detailed description of the model) provides evidence that older age does not
significantly influence the curvature of an assumed utility function, whereas the model shows that the
utility function of participants becomes less concave in the presence of feedback (i.e. the willingness
to take risks increases). Ambiguity preferences are influenced by age and the presence of feedback. It
appears that feedback changes the weighting of probabilities, and that younger adults overestimate the
probabilities in ambiguous decisions more strongly than older adults.

To summarize, it appears that older age overall positively influences the propensity to accept gambles,

T
that ambiguous gambles are chosen less often than risky gambles and that feedback increases the

IP
propensity to gamble. In addition, older adults are more prone to gamble under ambiguity, while
younger adults react more strongly to feedback (which increases the propensity to gamble).

CR
3.2. Experiment 2

US
Experiment 2 is a follow-up study to experiment 1 and aims at testing hypothesis 4a and 4b. To do
AN
this, 100 of the participants of experiment 1 played an additional risk task based on statistical
probabilities (thus feedback now allows participants to learn about probabilities of winning which are
a priori not given). In addition, these participants answered a risk question which we could compare to
M

data from a representative sample of the German population to test external validity of our results
(Wagner, Frick, & Schupp, 2007), i.e. to test how closely the answers on this question in our sample
ED

match the answers given by a similar sample in the general population.

3.2.1. Procedures
PT

To measure decisions with statistical probabilities, participants made sequential choices on the
Balloon Analogue Risk Task (description below). In order to assess self-assessment of risk
CE

preferences, participants also responded to the following item (same question as in the SOEP
questionnaire (Wagner, Frick, & Schupp, 2007)): “On a scale from 0 to 10, how would you assess
your willingness to take risks?” A value of 0 corresponded to not willing to take risks at all, and a
AC

value of 10 to high willingness to take risks. Depending on their choices on the experimental task,
participants on average earned an additional € 4.75 (S.D. 0.46). The questionnaire question was not
incentivized.

3.2.1.1. The balloon analogue risk task (BART):

In the BART (Lejuez et al., 2002), participants see a series of 20 virtual balloons, one after another,
and can earn money by pumping up the balloon. The money is stored on a temporary account until
participants decide to collect the money (transfer it to a permanent account) and to proceed to the next

13
 ACCEPTED MANUSCRIPT

balloon, or until the balloon pops and the money is lost. Participants do not receive detailed
information on the probabilities that a balloon pops. Participants know that at some point, every
balloon pops, and that the explosion could happen at any moment. It could pop at the first pump or
when the balloon fills the entire screen (we implicitly assume that common-sense would tell
participants that the probability that a balloon pops after the first pump is quite low, and is increasing
the larger the balloon gets). When a balloon is filled beyond its individual exploding point, it pops on
the screen. Whenever a balloon pops, the money of the temporary account is lost. Then, the next
empty balloon appears on the screen. At every moment during a trial, the participant can interrupt the

T
pumping and press a “collect money” button. The pressing of this button transfers the money from the
temporary account to a permanent account, where it cannot be lost.

IP
After each money transfer or explosion, the trial ends, followed by a new balloon that appears on the

CR
screen until the participant has seen a total of 20 balloons. The chance that a balloon pops is randomly
generated, with a starting probability of P = 1/64 (0.0156). If the balloon does not pop at the first

US
pump, the probability that it pops at the second pump is P = 1/63, P = 1/62 at the third pump, and so on
until the 64th pump, where the probability will be P = 1/1. Every pump increases the money on the
temporary account that can be lost by popping, and diminishes the relative additional earnings by
AN
pumping. After the first pump, participants risk only 5 cents on the temporary account by an additional
pump but can increase their potential earnings by 100% by executing this pump. In contrast, after the
30th pump, participants risk the 1.5 ECU on the temporary account for an additional increase in
M

earnings of only 3.3%.

ED

Note that optimal behaviour that maximizes payoffs in expected terms is to pump 32 times. In general,
the earlier a participant stops, the more risk averse she or he is.

3.2.2. Statistical methods and results

PT

To investigate experience and age effects on the BART, we applied a fully identified repeated-
CE

measures general linear model, with number of pumps as dependent variable, trial number as within-
subjects factor, and age as between-subjects factor. Analysing the within-subject factor trial number, a
significant increase in participants’ readiness to take risks over the course of the experiment can be
AC

observed (F(13.239) = 5.575; p = .001), independently of age group2. Young adults are found to be more
risk-seeking (F(1) = 7.346; p = .008), but the change in risk taking over the course of the experiment is
not influenced by age group (trial*age group: F(13.239) = 0.867; p = .590), thus disconfirming H4b. To

2
For the repeated-measures general linear model, the Mauchly test for sphericity revealed a violation of the
sphericity-assumption (Mauchly’s W = .015; df = 189; approximate χ² = 385.685; p < .001). Therefore,
Greenhouse-Geisser-corrected results for the repeated-measures analysis will be reported (the corrected results,
however, do not reveal any fundamental differences compared to conventional, uncorrected results). Although
results are visualized with logarithmic fitting functions, the statistics reported are on a linear model as a
logarithmic transformation of the data does not reveal qualitative differences in the analyses.

14
 ACCEPTED MANUSCRIPT

visualize our results, we applied a logarithmic fitting function to the behavioural measures of each age
group (Figure 4).

T
IP
CR
Figure 4: Average behaviour of young and older adults on the BART. Error bars represent 95%

US
confidence intervals. Thin lines are observed behaviour, thick lines correspond to logarithmic fitting
functions (young = 14.451 + 1.858 × ln(x); older = 13.491 + 0.649 × ln(x)).
When taking a modelling approach to estimate each individuals’ choice function, in which a
AN
participants’ sensitivity to gains and her initial estimate of the probability of winning are central to
decision making (see Appendix 3 for a detailed description of the model), it appears that there are no
age differences in the sensitivity to gains, in the initial estimate of the probability of winning and in
M

the variance of this estimate (or participants’ “optimism”). Yet older participants exhibit a larger
response sensitivity. This means that the probability of pumping after each additional pump is higher
ED

in young adults and hence that young adults take more risk on average. When combining these results
with the outcomes of the RAT, we can show that the initial estimate of the probability of winning can
PT

be predicted by behaviour on the RAT, both under risk and under ambiguity (with risk and ambiguity
behaviour correlating in opposite direction) and that the measure of sensitivity to gains is predicted by
risk behaviour alone (Table 7 in Appendix 3).
CE

Unlike with the RAT, but as hypothesized based on the literature, young adults are more risk-seeking
compared to older adults (H4a). Although we confirm the underlying assumption of H4b that the
AC

BART measures both components of risk and of ambiguity, when controlling for behaviour on the
RAT, no significant age differences remain. Hypotheses H4b (increase of distance between young and
older adults over trials) and H5 (stronger effect for older adults when comparing statistical to a priori
probabilities) thus need to be rejected.

3.2.3. Representativeness of the sample of participants

To get an idea about the external validity of our results and to help us understanding if (but not how
so) the selection of a highly educated sample of participants can be a driving force behind the observed

15
 ACCEPTED MANUSCRIPT

effects, we compare the results of the (non-incentivized) measure of risk preferences observed in our
sample to a representative sample of the German population (from 2009 - we are aware that self-
reported scales of risk preferences only correlate weakly with behaviour on tasks assessing risk (Frey
et al., 2017), but the intention of this section is rather to provide a better understanding of the
participant sample). The German Socio-Economic Panel (GSOEP; Wagner, Frick, & Schupp, 2007)
contains different questions related to risk attitudes. The main measure assesses “willingness to take
risks, in general” (called henceforth general risk preference). Roughly 13,000 persons answered the
item relevant to our analysis. We use a sample that is spread over a wide age range (18-88 years of

T
age), with a mean age of 53.57 years (SD 16.05) and compare it to the part of our sample that
answered the same non-incentivized self-report measure of risk preferences. In our sample, despite

IP
significant differences in risk behaviour between young and older adults, no significant difference was

CR
found in the self-assessment of risk attitudes. Young adults on average reported a value of 4.98 (SD:
0.239) and older adults on average were at 5.49 (SD: 0.225), hence also not differing significantly (t(98)
= 1.534, p = .128).

US
Given that our sample consisted of students and highly educated older adults, we intended to provide a
rough estimate of how much of our results are driven by education. To control for the effects of
AN
education on the relationship between self-reported risk preference and age, we provide evidence from
the SOEP by bootstrapping 1000 random samples of young and older adults (each N = 200; 100 older
adults) with a minimum of 13 years of education. Young adults were younger than 30 years (mean:
M

25.60, SD: 1.365), older adults were between 58 and 88 years of age (mean: 67.89, SD: 4.957). It
appears that, as in our experimental group, young and older adults do not differ significantly on the
ED

self-assessment of their own general risk attitude (t(198) = 1.296, p = .196). On a scale from 0 to 10,
young adults reported an average of 4.45 (SD: 2.045) for the willingness to take risks, and older adults
PT

reported a mean value of 4.09 (SD: 2.088). Thus, while older and younger adults on average do not
differ between our study and the general population (all p > .1) it is noteworthy that standard
deviations are an order of magnitude higher in the sample of the GSOEP, i.e. preferences in our study
CE

were much more homogenous than in a highly educated random sample in the general population. In a
second step, we used a linear regression model with age and gender as predictors of risk preferences.
AC

We also introduced years of education as additional predictor into a second regression model.
Examination of the results reveals that both age (β = -.172, S.E. = .001, p < .001) and being female (β
= -.212, S.E. = .038, p < .001) are negatively related to the willingness to take risks in the sample from
the SOEP. After introducing educational level into the regression, the effects of age (β = -.170, S.E. =
.001, p < .001) and gender (β = -.208, S.E. = .038, p < .001) still hold. Education appears to also have
a positive effect on risk taking: the higher the level of education, the more likely participants consider
themselves to take risks (β = .059, S.E. = .005, p < .001). The non-significant age-difference in self-

16
 ACCEPTED MANUSCRIPT

assessed risk preference on the SOEP question in our experiment might be (at least partly) explained
by the high educational background of our participants.

4. Discussion

The aim of this study was to gain a better understanding of older adults’ decision making in situations
of financial uncertainty and to investigate age differences in the propensity to gamble in uncertain
situations. Based on findings in the literature, we hypothesized that older adults do have a different

T
propensity to gamble than young adults in risky (H1) and ambiguous (H2) conditions. We also

IP
expected that providing feedback on choices with a priori probabilities should result in behaviour
indicative of a smaller distance between subjective and objective probabilities in both age groups. We

CR
thus expected gambling behaviour to be more frequent and an estimated probability weighting
function to be closer to the actual probabilities when feedback is provided (H3). We also hypothesized

US
that feedback will c.p. increase older participants´ probability to gamble (H3a) under risk and under
ambiguity. Concurrently, we hypothesized that older participants c.p. do react less to feedback than
young participants, thus demonstrating a smaller difference in behaviour between feedback and non-
AN
feedback treatments (H3b). We finally hypothesized that whether probabilities are given or have to be
learned has an influence on age differences. We expected older participants to take less risk than
young participants on a task with statistical probabilities (H4a) and an increase in the difference in risk
M

taking between young and older participants over time (H4b). Comparing both kinds of risk tasks, we
expected that age has a larger effect on behaviour on a task where learning is required. Within our
ED

group of elderly we hypothesized a difference in risk taking between the two risk tasks which should
be less pronounced for the young participants (H5).
PT

We provide support for most, but not all of our hypotheses. Older adults appear to be equally willing
to take risks as young adults when considering a priori probabilities (i.e., as tested in the RAT task in
CE

experiment 1, disconfirming H1). Ambiguity behaviour differs with age (confirming H2). Feedback
increases the propensity to gamble, both in decisions under risk and in decisions under ambiguity,
however, the effect is strongest in young adults (confirming H3b). When making decisions based on
AC

statistical probabilities in the BART task, older adults gamble less than young adults (confirming
H4a). If learning is required, age does not appear to have a larger effect on behaviour than in a task
where learning is not required (rejecting H5).

Regarding H1, we did not find any significant age differences in the average number of risk-taking
responses on the RAT. When estimating a hypothesized utility function, it also appeared that age did
not influence the curvature of the utility function. This indicates that, when dealing with decisions
under risk where probabilities are explicitly known, older adults behave similarly to young adults

17
 ACCEPTED MANUSCRIPT

(when decisions are made in the gain domain; Cooper, Blanco & Maddox 2016). Results are different
when it comes to ambiguous choices (H2): our findings in decision-making behaviour under
conditions of ambiguity showed that older adults gamble more than young adults in ambiguous
conditions, without feedback, and that older adults tend to estimate the probability of gaining more
realistically than young adults. Studies that have assessed cognitive processes and strategies in the
elderly might provide explanations for the observed behaviour. The most likely factor that could have
influenced the results is experience. Older adults have had a lifetime to decide and develop strategies
for decisions under ambiguity. Hence, their schemas or memory traces are more developed than those

T
of young adults. One survey of bank managers, for example, revealed that older managers’ business
decisions were more aggressive than the decisions of younger managers (Brouthers et al. 2000), and

IP
various studies found that uncertain investments increased until a particular age (Riley Jr & Chow

CR
1992; Schooley & Worden 1999; Jianakoplos & Bernasek 1998). An alternative explanation for the
age difference in the propensity to gamble under ambiguity is given by Mata et al. (2007). In their
study, they found a difference in strategies used by young and older adults to make a decision: older

US
adults looked up less information and took more time to process it. If we apply this to the fact that
ambiguity is a condition with less information available than risk, we can hypothesize that ambiguous
AN
gambles are easier to process for older adults.

A novelty of this study was the investigation of the effects of feedback on uncertain choices (H3).
Feedback had a salient effect on the propensity to gamble, both by decreasing the level of risk aversion
M

(i.e. increasing the coefficient of curvature of the utility function) and by diminishing the weighting of
ambiguous probabilities. Young adults gamble significantly more in the presence of feedback than
ED

without it. The main reason for this effect lies in the fact that decisions under uncertainty strongly rely
on subjective probabilities. Feedback is an important factor in reducing subjective uncertainty (Hogg
PT

& Mullin 1999), which in our experiment led participants to gamble more than in the absence of
feedback. By design, in our RAT experiment, feedback could not have been used to learn about the
probabilities of winning (all trials had to be considered separately because each trial is an independent
CE

event). Two hypotheses arise therefore. First, the greater experience of older participants might let
them estimate probabilities more accurately than young adults, which would lead to a lower reactivity
AC

to feedback in terms of subjective probabilities. Second, older adults might simply not (or to a lower
extent) update their beliefs about subjective probabilities. The results of the first experiment provide
support for the first hypothesis, but not for the second.

Several factors might play a role explaining the age differences found in statistical but not in a priori
probabilities (H4). The most salient factor certainly is the different learning requirements of the tasks.
Learning requirements are absent for the RAT. In contrast, performance on the BART relies on the
participant’s ability to learn the probabilities of outcomes from experience. When learning from
experience is required to approach optimal (i.e. payoff-maximizing, risk neutral) behaviour, older

18
 ACCEPTED MANUSCRIPT

adults appear to be more risk averse than young adults in a limited set of trials. One reason for the
observed age difference could be that learning rates of young adults are generally steeper than those of
older adults (Merriam et al. 2007), and thus young adults approach optimal behaviour faster than older
adults. The results from our sample did not support this reasoning. A non-significant interaction effect
between age group and the propensity to gamble over trials indicated that the rate of learning did not
significantly differ between young and older adults. We observed that in both age groups the
propensity to gamble increases over the trials with a similar pace, and that the response sensitivity is
the factor that differentiates young from older adults: young adults are more willing to perform an

T
additional pump. At the same time, both age groups were risk averse. Risk-taking behaviour, on
average, stays below the risk neutral behaviour of 32 pumps per balloon over 20 trials. The difference

IP
between young and older adults mainly lies in the starting values. Older adults begin the game with a

CR
lower average number of pumps than their younger counterparts, and then subsequently increase the
number of pumps at a pace that is not significantly different from young adults. These findings are
consistent with evidence that older adults are more cautious than young adults (Rolison, Hanoch &

US
Wood 2012; Rolison, Wood & Hanoch 2017; Starns & Ratcliff 2010) and that older adults only
display differences in learning performance when learning demands are high, which arguably is not
AN
the case in the BART (Rolison, Hanoch & Wood 2012). It also is of importance to note that we
observe correlations in behaviour between the RAT and BART. Although recent evidence suggests
that the link between behavioural tasks is rather weak (Pedroni et al., 2017), these correlations provide
M

us with some confidence that the same concepts of risk and ambiguity were captured across both tasks.

The current study also has its limitations. We selected the older participants in a way that eased
ED

comparability with our young participants: All older participants were healthy, highly educated and
practised a cognitively active lifestyle. Overall, older adults are very heterogeneous in their cognitive
PT

abilities, and activity might preserve cognitive ability with aging. The analysis of the response to the
self-assessment question has shown that, even when selecting only highly educated participants, the
responses on the question were more homogenous in our sample than in the general population. We
CE

have shown that whereas age has a negative impact on the willingness to take risks in the population,
the level of education of participants has a positive impact. Although it is difficult to conclude that the
AC

non-representativeness of our sample has an effect on the results, we still think that there is a
possibility that we reduced some of the effects claimed in other studies to be attributable to age by
selecting highly educated participants. This does not mean that we are not dealing with large selection
effects in the current study, but we refer the reader for a more detailed analysis of the effects of
education, health, and other factors influencing the willingness to take risks, to other authors (Dohmen
et al. 2011), as this was not the aim of our study. Nevertheless, future experiments should focus on age
differences in more heterogeneous participant groups on both sides of the age range (Henrich et al.
2010).

19
 ACCEPTED MANUSCRIPT

In this experiment, we wilfully did not try to rule out cohort effects. All western populations are
rapidly growing older, so that there is an immediate need to describe older adults’ behaviour. Certainly
it also is of great importance to understand whether and how cohorts differ (see e.g. Malmendier &
Nagel, 2009), but our first aim was to describe the differences between young and older adults, paving
the way for future work where one should try to minimize the impact of cohort effects. In the same
line, how behaviour develops over the life-course is an interesting topic to extend the current work.
With evidence that risky decision making follows an inverse-U shaped function over the life course,
studies spanning more age segments can provide additional insights missed by the current paper.

T
5. Conclusion

IP
In conclusion, we observed that older adults’ decision-making behaviour differs from that of younger

CR
adults. In risky conditions, older adults behave like young adults if a priori probabilities apply, but in
risky conditions were statistical probabilities are part of the decision and in ambiguous conditions, age
differences appear. Young adults are more ambiguity averse if no feedback is provided, and older

US
adults exhibit a higher risk aversion when it comes to statistical probabilities. Despite several possible
explanations for our results, more work will be needed to fully understand the causes of our findings.
AN
On a more practical level, our work contributes to growing evidence that older adults’ decision making
differs from that of younger adults. In our societies, older adults represent a growing part of the
population, and a part of the population that will work until a higher age, thus also making financial
M

decisions at a higher age. Understanding how decision making is affected by age becomes crucial for
numerous situations of everyday life, and further research is needed to understand how older adults
ED

make their decisions, to help employers, policy makers, financial institutions, but also older adults
themselves, to cope with the effects of the demographic change.
PT

6. Acknowledgements

This research has been made possible by a FRONTIER-Fonds grant from the University of Heidelberg
CE

to the authors. The authors are much obliged to Daniela Jopp and Rui Mata for their helpful
comments.
AC

20
 ACCEPTED MANUSCRIPT

References
Ashman, O., Dror, I., Houlette, M., & Levy, B. (2003). Preserved risk-taking skills in old age. North
American Journal of Psychology, 5(3), 397-406.
Baumeister, R. F., Vohs, K. D., DeWall, N. C., & Zhang, L. (2007). How Emotion Shapes Behavior:
Feedback, Anticipation, and Reflection, Rather Than Direct Causation. Personality and Social
Psychology Review, 11(2), 167-203, doi:10.1177/1088868307301033.
Bechara, A., Damasio, H., Tranel, D., & Damasio, A. R. (2005). The Iowa Gambling Task and the
somatic marker hypothesis: some questions and answers. Trends in Cognitive Sciences, 9(4),
159-162.
Bellucci, G., & Hoyer, W. J. (1975). Feedback Effects on the Performance and Self-Reinforcing
Behavior of Elderly and Young Adult Women. Journal of Gerontology, 30(4), 456-460,
doi:10.1093/geronj/30.4.456.

T
Best, R., & Charness, N. (2015). Age differences in the effect of framing on risky choice: A meta-
analysis. Psychology and Aging, 30(3):688-698. http://doi.org/10.1037/a0039447

IP
Brouthers, K., Brouthers, L., & Werner, S. (2000). Influences on strategic decision-making in the
Dutch financial services industry. Journal of Management, 26(5), 863.

CR
Buelow, M., & Suhr, J. (2009). Construct Validity of the Iowa Gambling Task. Neuropsychology
Review, 19(1), 102-114, doi:10.1007/s11065-009-9083-4.
Camerer, C. F., & Hogarth, R. M. (1999). The effects of financial incentives in experiments: A review
and capital-labor-production framework. Journal of risk and uncertainty, 19(1-3), 7-42.

US
Carstensen, L. L., & Hartel, C. R. (2006). When I'm 64. Commitee on Aging Frontiers in Social
Psychology, Personality, and Developmental Psychology. Washington, DC: The National
Academy Press.
Chaubey, N. P. (1974). Effect of age on expectancy of success and on risk-taking behavior. Journal of
AN
personality and social psychology, 29(6), 774-778, doi:DOI: 10.1037/h0036178.
Cooper, J. A., Blanco, N. J., Maddox, W. T. (2017). Framing Matters: Effects of Framing on Older
Adults' Exploratory Decision-Making. Psychology and Aging, 32(1), 60-68.
Deakin, J., Aitken, M., Robbins, T., & Sahakian, B. (2004). Risk taking during decision-making in
M

normal volunteers changes with age. Journal of the International Neuropsychological Society,
10(4), 590-598.
Dohmen, T., Falk, A., Huffman, D., & Sunde, U. (2012). The intergenerational transmission of risk
ED

and trust attitudes. The Review of Economic Studies, 79(2), 645-677.

Dohmen, T., Falk, A., Huffman, D., Sunde, U., Schupp, J., & Wagner, G. G. (2011). Individual risk
attitudes: Measurement, determinants, and behavioral consequences. Journal of the European
Economic Association, 9(3), 522-550.
PT

Dror, I., Katona, M., & Mungur, K. (2000). Age differences in decision making: to take a risk or not?
Gerontology, 44(2), 67-71.
Eppinger, B., & Kray, J. (2011). To choose or to avoid: Age differences in learning from positive and
negative feedback. Journal of Cognitive Neuroscience, 23, 41-52.
CE

Frey, R., Pedroni, A., Mata, R., Rieskamp, J., & Hertwig, R. (2017). Risk preference shares the
psychometric structure of major psychological traits. Science Advances, 3
Hallahan, T., Faff, R., & McKenzie, M. (2004). An empirical investigation of personal financial risk
AC

tolerance. Financial Services Review, 13(1), 57-78.

Hanoch, Y., Rolison, J., & Freund, A. M. (2018). Does Medical Risk Perception and Risk Taking
Change With Age? Risk Analysis, 38(5), 917-928.
Hau, R., Pleskac, T. J., Kiefer, J., & Hertwig, R. (2008). The description–experience gap in risky
choice: the role of sample size and experienced probabilities. Journal of Behavioral Decision
Making, 21(5), 493-518, doi:10.1002/bdm.598.
Henninger, D. E., Madden, D. J., & Huettel, S. A. (2010). Processing speed and memory mediate age-
related differences in decision making. Psychology and aging, 25(2), 262-270,
doi:10.1037/a0019096.
Henrich, J., Heine, S., & Norenzayan, A. (2010). The weirdest people in the world? Behavioral and
Brain Sciences, 33(2-3), 61-83.
Hertwig, R., & Erev, I. (2009). The description–experience gap in risky choice. Trends in Cognitive
Sciences, 13(12), 517-523, doi:10.1016/j.tics.2009.09.004.

21
 ACCEPTED MANUSCRIPT

Hogarth, R.M. (2005). Deciding analytically or trusting your intuition? The advantages and
disadvantages of analytic and intuitive thought. In: Betsch, T. & Haberstroh, S. (Eds.), The
routines of decision making (pp. 67-82). New York: Psychology Press.
Hogg, M. A., & Mullin, B.-A. (1999). Joining groups to reduce uncertainty: Subjective uncertainty
reduction and group identification. In D. A. M. A. Hogg (Ed.), Social identity and social
cognition (pp. 249-279). Malden: Blackwell Publishing.
Hsu, M., Bhatt, M., Adolphs, R., Tranel, D., & Camerer, C. (2005). Neural systems responding to
degrees of uncertainty in human decision-making. Science, 310(5754), 1680-1683.
Huang, Y. H., Wood, S., Berger, D. E., & Hanoch, Y. (2015). Age differences in experiential and
deliberate processes in unambiguous and ambiguous decision making. Psychology and Aging,
30, 675-687.
Huang, Y., Wood, S., Berger, D., & Hanoch, Y. (2013). Risky choice in younger versus older adults:

T
Affective context matters. Judgment and Decision Making, 2, 179-187.
Jianakoplos, N., & Bernasek, A. (1998). Are women more risk averse? Economic Inquiry, 36(4), 620-

IP
630.
Josef, A. K., Richter, D., Samanez-Larkin, G. R., Wagner, G. G., Hertwig, R., & Mata, R. (2016).
Stability and change in risk-taking propensity across the adult life span. Journal of Personality

CR
and Social Psychology, 111(3), 430-450. http://doi.org/10.1037/pspp0000090
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk.
Econometrica: Journal of the Econometric Society, 47(2), 263-291.

US
Kausler, D.H. (1994). Learning and Memory in Normal Aging. San Diego, CA, US: Academic Press.
Knight, F. H. (1921). Risk, uncertainty and profit. New York: Hart, Schaffner and Marx.
Kramer, M. W. (1994). Uncertainty Reduction During Job Transitions. Management Communication
Quarterly, 7(4), 384-412, doi:10.1177/0893318994007004002.
AN
Lejuez, C., Read, J. P., Kahler, C. W., Richards, J. B., Ramsey, S. E., Stuart, G. L., et al. (2002).
Evaluation of a behavioral measure of risk taking: The Balloon Analogue Risk Task (BART).
Journal of Experimental Psychology: Applied, 8(2), 75-84.
Loewenstein, G. (2000). Emotions in economic theory and economic behavior. American
M

economic review, 90(2), 426-432.

Loewenstein, G. F., Weber, E. U., Hsee, C. K., & Welch, N. (2001). Risk as feelings. Psychological
bulletin, 127(2), 267.
ED

Malmendier, U., & Nagel, S. (2009). Depression Babies: Do Macroeconomic Experiences Affect
Risk-Taking? NBER Working Paper No. 14813.
Mamerow, L., Frey, R., & Mata, R. (2016). Risk taking across the life span: A comparison of self-
report and behavioral measures of risk taking. Psychology and Aging, 31(7), 711-723.
PT

Mata, R., Josef, A. K., Samanez‐Larkin, G. R., & Hertwig, R. (2011). Age differences in risky choice:
a meta‐analysis. Annals of the New York Academy of Sciences, 1235(1), 18-29.
Mata, R., Schooler, L. J., & Rieskamp, J. (2007). The aging decision maker: cognitive aging and the
adaptive selection of decision strategies. Psychology and aging, 22(4), 796.
CE

Mathewson, K. J., Dywan, J., Snyder, P. J., Tays, W. J. and Segalowitz, S. J. (2008), Aging and
electrocortical response to error feedback during a spatial learning task. Psychophysiology,
45(6), 936–948. doi: 10.1111/j.1469-8986.2008.00699.x
AC

Merriam, S. B., Caffarella, R. S., & Baumgartner, L. M. (2007). Learning in adulthood: A

comprehensive guide. San Francisco: John Wiley & Sons.
Pedroni, A., Frey, R., Bruhin, A., Dutilh, G., Hertwig, R., & Rieskamp, J. (2017). The risk elicitation
puzzle. Nature Human Behaviour.
Peters, E., Finucane, M. L., MacGregor, D. G., & Slovic, P. (2000). The bearable lightness of aging:
Judgment and decision processes in older adults. In N. R. Council (Ed.), The aging mind:
Opportunities in cognitive research (pp. 144-165). Washington, D.C.: The National Academy
Press.
Riley Jr, W., & Chow, K. (1992). Asset allocation and individual risk aversion. Financial Analysts
Journal, 48(6), 32-37.
Rolison, J. J., Hanoch, Y., & Wood, S. (2012). Risky decision making in younger and older adults:
The role of learning. Psychology and Aging, 27, 129-140.

22
 ACCEPTED MANUSCRIPT

Rolison, J. J., & Pachur, T. (2017). How well do we know our inner daredevil? Probing the
relationship between self-report and behavioral measures of risk taking. Journal of Behavioral
Decision Making, 30(2), 647-657.
Rolison, J., Wood, S., & Hanoch, Y. (2017). Age and adaptation: Stronger decision updating about
real world risks in older age. Risk Analysis, 37(9), 1632-1643.
Samanez-Larkin, G. R., & Knutson, B. (2015). Decision making in the ageing brain: changes in
affective and motivational circuits. Nature Reviews Neuroscience, 16(5), 278-289.
http://doi.org/10.1038/nrn3917
Schooley, D., & Worden, D. (1999). Investors' asset allocations versus life-cycle funds. Financial
Analysts Journal, 55(5), 37-43.
Starns, J. J., & Ratcliff, R. (2010). The effects of aging on the speed-accuracy compromise: Boundary
optimality in the diffusion model. [Research Support, N.I.H., Extramural]. Psychology and

T
aging, 25(2), 377-390, doi:10.1037/a0018022.
Tymula, A., Belmaker, L. A. R., Ruderman, L., Glimcher, P. W., & Levy, I. (2013). Like cognitive

IP
function, decision making across the life span shows profound age-related changes.
Proceedings of the National Academy of Sciences, 110(42), 17143-17148.
Tripp, G., & Alsop, B. (1999). Age-related changes in sensitivity to relative reward frequency. New

CR
Zealand Journal of Psychology, 28, 30-36.
Trope, Y. (1979). Uncertainty-reducing properties of achievement tasks. Journal of Personality and
Social Psychology, 37(9), 1505.

US
Vaupel, J. W. (2010). Biodemography of human ageing. Nature, 464(7288), 536-542.
Wagner, G. G., Frick, J. R., & Schupp, J. (2007). The German Socio-Economic Panel Study (SOEP)–
Evolution, Scope and Enhancements. Journal of Applied Social Science (Schmollers
Jahrbuch), 127(1), 139-169.
AN
Wallsten, T. S., Pleskac, T. J., & Lejuez, C. W. (2005). Modeling behavior in a clinically diagnostic
sequential risk-taking task. Psychological review, 112(4), 862.
White, C., Gummerum, M., Wood, S., & Hanoch, Y. (2017). Internet safety and the Silver Surfer: The
relationship between gist reasoning and adults' risky online behaviour. Journal of Behavioral
M

Decision Making, 30(4), 819-827.

Wood, S., Busemeyer, J., Koling, A., Cox, C. R., & Davis, H. (2005). Older adults as adaptive
decision makers: evidence from the Iowa Gambling Task. Psychology and aging, 20(2), 220.
Zamarian, L., Sinz, H., Bonatti, E., Gamboz, N., & Delazer, M. (2008). Normal aging affects decisions
ED

under ambiguity, but not decisions under risk. Neuropsychology, 22(5), 645-657.
PT
CE
AC

23
 ACCEPTED MANUSCRIPT

Appendix 1: Instructions of the RAT and list of choices

Appendix 1:

Instuctions:

In the current game, you are requested to resolve decision problems. You are getting paid for two of

the decisions chosen at random. Your possible gain is presented to you on the screen in experimental

T
currency units (ECU). 10 ECU correspond to 1€ - at the end of the experiment, the ECU will be

IP
converted into euro. If you have won for example 70 ECU, your payment would be 7€.

CR
In the current game, card decks will be presented to you. These card decks always contain some

US
amount of red and of blue cards. In some situations, the exact amount of red and blue cards in the deck

is revealed to you, in other situations, you only will know the total amount of cards in the deck.
AN
In the game, you have always the choice either to bet that a certain colour will be drawn out of the
M

deck, or to receive a sure amount of money. If you bet, you will receive the indicated amount of ECU

if the colour you chose is drawn; otherwise you don't receive anything. If you choose the certain
ED

amount, you will get the amount automatically, without any influence of the drawn card colour on

your gain.
PT

Prior to the game, you will have the possibility to get familiar with the task on two examples.
CE

The amount which you will be paid within each trial will be presented to you on the computer screen
AC

during the experiment.

If you have any questions, please raise your hand. One of the experimenters will come to your table

and answer your question privately.

Choices on the RAT

24
 ACCEPTED MANUSCRIPT

Condition Nr. Red Blue Total ECU bet ECU certain

Risk 1 2 18 20 12 8
2 18 12 30 16 12
3 9 1 10 19 12
4 3 9 12 16 13
5 3 12 15 20 10
6 21 9 30 20 15
7 8 32 40 16 10
8 5 10 15 12 8
9 6 14 20 18 12
10 16 24 40 16 8
11 4 36 40 20 14

T
12 18 9 27 15 10
13 26 13 39 12 8

IP
14 10 20 30 12 10
15 3 2 5 18 12

CR
16 27 3 30 14 11
17 12 28 40 16 14
18 4 1 5 14 8
19 8 12 20 18 9
20
21
22
23
7
24
8
7
21
8
2
3
28
32
10
10
US 12
19
16
13
8
13
10
10
AN
24 15 5 20 12 6
Ambiguity 1 20 20 10
2 10 16 7
3 40 20 12
M

4 30 13 7
5 40 17 7
6 30 18 6
ED

7 20 20 11
8 30 12 7
9 15 20 12
10 30 20 12
PT

11 12 19 6
12 15 16 7
13 40 14 8
CE

14 28 18 8
15 40 20 11
16 10 16 9
17 32 19 9
AC

18 39 16 6
19 10 20 11
20 27 20 8
21 5 17 8
22 20 12 5
23 5 18 11
24 20 16 6
Red = number of red cards in the deck, Blue = number of blue cards in the deck,
Total = total amount of cards in the deck, ECU bet = amount to win when
betting, ECU certain = amount to earn with the sure option

25
 ACCEPTED MANUSCRIPT

Appendix 2 – Instructions of the BART (text translated from the instruction n screens)

BART: text on the instructions screen

Now you're going to see 20 balloons, one after another, on the screen. For each balloon, you will use
the mouse to click on the button that will pump up the balloon. Each click on the mouse pumps the
balloon up a little more.

T
BUT remember, balloons pop if you pump them up too much. It is up to you to decide how much to

IP
pump up each balloon. Some of these balloons might pop after just one pump. Others might not pop
until they fill the whole screen.

CR
You get MONEY for every pump. Each pump earns 0.05 ECU. But if a balloon pops, you lose the
money you earned on that balloon. To keep the money from a balloon, stop pumping before it pops
and click on the button labeled "Collect €€€".
US
After each time you collect €€€ or pop a balloon, a new balloon will appear.
AN
At the end of the experiment, you will be paid the amount earned on the game.

Click the left mouse button to see the summary.

M
ED

BART: text on the summary screen

You make 0.05 ECU for each pump.

PT

You save the money from a balloon when you click "Collect €€€".

You lose money from a balloon when it pops.

CE

There are 20 balloons.

AC

You will be paid the exact amount you earned on the game.

Now, do you have any questions?

Click the left mouse button to begin.

26
 ACCEPTED MANUSCRIPT

Appendix 2: Modelling Approach for the RAT

In accordance with Hsu et al. (2005), a parametric analysis to estimate coefficients of risk and
ambiguity preferences on the RAT was conducted via a nonlinear stochastic choice model, combining
data from all treatments. Participants’ utility functions for money are assumed to follow a power
function u(x,α) = xα which is conveniently characterized by one parameter and widely used in
empirical estimations of this sort. Participants are assumed to weight probabilities according to
pγ
Prospect-theory’s probability weighting function w(p,γ)= 1 (Kahneman and Tversky 1979).
γ
[pγ +(1-p) ]γ

T
The α parameter of the utility function is interpreted as the risk aversion coefficient, i.e., the curvature

IP
of the utility function. The γ parameter is interpreted as the ambiguity aversion coefficient, i.e. how
much do people underestimate ambiguous probabilities (Hsu et al., 2005). We assume that people

CR
combine these weighted probabilities and utilities linearly, so that their weighted subjective expected
utility is U(p,x,α,γ) = w(p,γ)u(x,α). The tasks are binary choices in which participants either choose a
gamble to win x (with probability p) or 0, or a certain payoff c. For the risky decks, the probabilities

US
are the ratios of the cards. For the ambiguous decks, we assume p = 0.5. Based on Hsu et al. (2005),
we constrain γ = 1 in all risky conditions and estimate γ from behavioural data in the ambiguity
AN
conditions. A small γ (< 1) indicates that small probabilities are overestimated whereas large
probabilities are underestimated, and vice-versa for a large (> 1) value of γ. Including both factors α
and γ in the model for ambiguity reflects the fact that choices between ambiguous gambles and sure
M

outcomes are always influenced both by attitude towards ambiguity and by risk attitude. The
probability that the participant chooses the gamble rather than the sure amount of money c is given by
ED

1
𝑈(𝑝,𝑥,𝛼,𝛾)𝜇
the formula 𝑃(𝑝, 𝑥, 𝑐, 𝛼, 𝛾) = 1 1 , where µ is the amount of stochastic errors, or the
𝑈(𝑝,𝑥,𝛼,𝛾)𝜇 +𝑢(𝑐,𝛼)𝜇

amount of “randomness”, in participants’ choices. Besides this baseline model, we introduce in a

PT

second model age group and the presence or absence of feedback as predictor variables for α and γ,
and age group as predictor for the occurrence of stochastic errors.
CE

In the model without controlling for age and feedback variables, it appears that our participants tend to
exhibit a concave utility function with α = .738 (S.E.: .044) and to underestimate ambiguous
AC

probabilities with γ = .734 (S.E.: .040). In the second model, when controlling for age groups and the
presence or absence of feedback (cf. Table 4), it appears that feedback has a significant effect on α,
increasing the willingness to take risks (i.e. rendering u(x,α) less concave). Older age, on the other
hand, does not significantly influence the curvature of the utility function. Ambiguity preferences are
influenced by age and the presence of feedback. It appears that feedback changes the weighting of
probabilities, and that younger adults overestimate the probabilities in ambiguous decisions more

27
 ACCEPTED MANUSCRIPT

strongly than older adults. It also appears that older adults make more stochastic errors than young
adults3.

Table 4: Utility coefficients

Parameter Predictor Coeff. S.E. p
α
Constant 1.217 0.100 < .001
Old -0.194 0.238 .414
no feedback -0.586 0.121 < .001

T
γ
Constant 0.446 0.022 < .001

IP
Old 0.267 0.146 .068
no feedback 1.601 0.198 < .001
μ

CR
Constant 0.054 0.010 <.001
Old 0.179 0.073 .014
Nbr. of observations: 9600; S.E. adjusted for 200 participants

US
AN
M
ED
PT
CE
AC

3
When excluding the 57 participants (26 young adults and 31 older adults) that chose inconsistently at least
once, the results remain qualitatively similar, yet the significant age effect on µ disappears (p = .351).

28
 ACCEPTED MANUSCRIPT

Appendix 3: Modelling Approach for the BART

Wallsten, Pleskac and Lejuez (2005) have evaluated different decision making models applied to the
BART. According to these authors, in the best-fitting model, prior to beginning with each balloon,
participants engage in a prospect-theory-like evaluation of the expected outcomes to determine an
optimal number of pumps. Their probability of pumping then decreases within each balloon such that
it equals 0.5 at the optimal stopping point. Participants also incorrectly treat the conditional probability
of explosion as constant rather than increasing over pumps. This leads to the assumption that
participants treat an event as binomial distribution with ph being the subjective probability that the

T
balloon will not explode at pump number h. The prior probability 𝑝
̂0 can then be represented as a beta

IP
distribution (with parameters a0 – estimated size of p prior to observing any data – and m0 – certainty
𝑎0
about the value of p – with m0 > a0 > 0) where E(𝑝
̂)
0 = , the initial estimate of the probability may be

CR
𝑚0

̂)
interpreted as a measure of optimism that the balloon will not explode and where var(𝑝0 refers to the

variance in the participants’ initial estimate. High variance subjects will be more sensitive to new data

US
from the balloon, while low variance participants will be less sensitive. It is assumed that a decision
maker, instead of sequentially evaluating each pump, may determine an optimal number of pumps
AN
prior to each balloon (optimal in the sense that the number of pumps optimizes the perceived prospect-
theoretic gains). The reference point is set to the holdings of the temporary account (0) and therefore
all evaluations are made in terms of gains. The expected gain for i pumps on balloon h is E h(i) =
M

πh,i(ix)γ, with πh,i being the probability of pumping balloon h i times successively without exploding (x
is the amount gained by increasing the balloon; in our case 5₵). The number of pumps that maximize
this equation depends on how πh,i is expressed. Assuming that participants use Bayes’ rule to update
ED

beliefs about probabilities with experience, we can write the participants’ estimate of p following
𝑎0 + ∑ℎ−1
ℎ′ =1
(𝑚ℎ′ −𝑑ℎ′ )
experience with h balloons as 𝑝ℎ = (with dh’ taking the value of 1 if the balloon
PT

𝑚0 +∑ℎ−1 𝑚
ℎ′ =1 ℎ′

exploded, and 0 else). Therefore, we can assume, given that the probability of the balloons not
exploding is assumed to be stationary over pumps with an estimated value of ph for all i, that πh,i = 𝑝ℎ𝑖 .
CE

When denoting the optimizing number of pumps as gh, this results in an optimization of Eh(i) when
−𝛾
AC

𝑔ℎ = ln(𝑝ℎ )
. The response model uses a logistic function constructed in a way that the probability of a

participant taking the ith pump on the hth balloon, rh,i strictly decreases with each pump (it equals 0.5
1
when i = gh): 𝑟ℎ,𝑖 = 𝛽𝛿ℎ,𝑖 with 𝛿ℎ,𝑖 = 𝑖 − 𝑔ℎ . β measures the sensitivity of a participant to her prior
1+𝑒

evaluation and remains constant with experience. The functions (more specifically the γ, β, E(𝑝
̂)
0 and

𝑣𝑎𝑟(𝑝
̂)
0 parameters) are estimated individually for each participant using the maximum likelihood

method and compared on the group level to make inferences about age differences. A more
comprehensive description of the model can be found in Wallsten et al. (2005). Table 6 displays the
results:

29
 ACCEPTED MANUSCRIPT

Table 6: Model estimates of the BART

Mean (S.E.) young Mean (S.E.) older t(df) p
β .355 (.035) .521 (.067) -2.180(71.462) .033
γ .806 (.053) .677 (.062) 1.599(98) .113
E(𝑝̂)
0 .967 (.004) .964 (.005) .469(98) .640
3 .411 (.125)
𝑣𝑎𝑟(𝑝 ̂)
0 ∗ 10 .400 (.122) .063 (98) .950
t-tests for independent samples, N=100. Degrees of freedom (df) adjusted
for unequal variances in β.

It appears that, although there are no age differences in the γ parameter or in the initial estimate of the

T
probability and in the variance of this estimate (or participants’ “optimism”), older participants exhibit
a higher value of the β parameter, thus a larger response sensitivity. This means that the probability of

IP
pumping after each additional pump is higher in young adults and hence that young adults take more

CR
risk on average. A multivariate regression analysis further reveals that β can be predicted by behaviour
on the RAT, both under risk and under ambiguity (with risk and ambiguity behaviour correlating in
opposite direction) and that γ, the measure of diminishing sensitivity to gains, is predicted by risk
̂)
behaviour alone (Table 7). E(𝑝
the RAT.
0 and 𝑣𝑎𝑟(𝑝
̂)
US
0 on the other hand do not correlate with behaviour on
AN
Table 7: Regression results
β γ
Model 1 Model 2 Model 1 Model 2
M

Risk -.013 (.007)* -.020(.009)* .019(.007)** .023(.010)*

Ambiguity .013 (.006)* .018(.008)* -.007(.006) -.006(.009)
Age group .103 (.080) .184(.186) -.101(.086) -.246(.202)
Age * risk -.014(.013) .009(.014)
ED

Age * ambiguity .009(.012) .003(.013)

Constant .513 (.106)*** .535(.125)*** .517(.114)*** .461(.136)***
R² .102 .114 .096 .102
PT

Multivariate linear regression. N = 100. Standard errors in parentheses. Regression

̂)
coefficients for E(𝑝0 and 𝑣𝑎𝑟(𝑝 ̂)
0 are not reported to increase legibility.
* p < .05, ** p < .01, *** p < .001
CE
AC

30