Personality, information acquisition, and choice under uncertainty: an experimental study.
Frechette, Guillaume R. ; Schotter, Andrew ; Trevino, Isabel 等
Personality, information acquisition, and choice under uncertainty: an experimental study.
I. INTRODUCTION
The economics of decision making under risk leaves little room for
personality. Differences between people are typically summarized as
differences in their risk aversion parameter, so this parameter serves
as a sufficient statistic for all personality characteristics.
While this may be adequate to explain decision making under risk,
where the decision maker (DM) knows with certainty the probability
distributions he faces, in environments where information is sparse (one
case being decisions under uncertainty) and where DMs are not informed
about the probability distributions they face, the personality of the DM
may play a role. In such environments, it is natural for DMs to seek out
information that would give them at least a glimpse into what the set of
probability distributions they face looks like, and thereby decrease the
amount of uncertainty they face. What we find in this paper is that in
uncertain environments the choices that DMs make are closely related to
the information they have at their disposal when making their choice and
that personality variables are relevant for the type of information they
gather. Since all probability distributions are known when a DM makes a
choice under risk, personality cannot play the same role. Indeed, we
find that only a DM's risk aversion coefficient is relevant for
choice under risk.
As we discuss later in the article, there are a number of theories
that might explain the type of information DMs seek when faced with
uncertainty and these may be tied to personality characteristics. For
example, due to personality differences, DMs may hold different
(pessimistic or optimistic) priors over the uncertainty they face and
seek different information depending on their degree of pessimism.
Alternatively, as a result of their personality characteristics, they
may want to be more or less confident of their choice before making it
and therefore have a preference for skewness, which will lead them to
seek out particular types of information (see Eliaz and Schotter 2010;
Masatlioglu, Orhun, and Raymond 2016). They may also employ different
choice heuristics, which require different information as inputs and the
heuristics they use may be a function of personality variables.
The main point of our article is that while we expect personality
to be relevant for choice under uncertainty, we do not expect such a
relationship when a DM faces a choice under risk. We find support for
this conjecture. (1)
This hypothesis is important since if decision making is influenced
by the information available to the DM and if information gathering
strategies are a function of people's personality characteristics,
then our results open the door for a systematic study of the impact of
personality on economic behavior and outcomes, a study which is in its
infancy (see Borghans et al. 2008; Almlund et al. 2011 for recent and
exhaustive surveys of the personality literature and its relationship to
economic decision making; and Rustichini 2009 for a discussion on the
importance of including personality traits into decision theory).
Our results allow us to go even further by indicating that the
impact of personality on choice under uncertainty is not limited to
information gathering, but extends to choice, in the sense that when the
information gathered is held constant, personality still affects choice
in environments of uncertainty. This result is in contrast to what we
find in our control treatment about choice under risk, where agents
receive full information about the probability distributions they face
and personality ceases to be relevant for choice. Hence, the importance
of personality on choice under uncertainty appears to be different from
choice under risk.
An environment where this is particularly relevant is that of
personal finance where investors are faced with a set of investments,
the properties of which are opaque. When investors have to choose
between two projects with risky returns they tend to gather more
information about these projects in order to decrease the amount of
uncertainty they face. Two possible ways in which they gather
information are by directly requesting information about the
characteristics of these projects (e.g., by studying financial reports),
or by getting advice from experts as to what project to choose (e.g.,
hiring a financial advisor). In this study, we look at the influence of
personality on choice in each of these two environments and our
treatments are meant to reflect such situations.
This article has two main parts. In the first part, we investigate
Hypothesis I, which focuses on whether personality has a differential
impact on choice in risky and uncertain environments. In the second
part, we study the role of personality on information acquisition in
uncertain environments, and how the information acquired determines
choice. (2) We show that personality determines the type of information
sought by agents. More precisely, we present evidence that a
subject's personality characteristics, as measured by the Big Five
Personality Scale (Costa and McCrae 1992) and the Sensation Seeking
Scale (SSS; Zuckerman 1994), are correlated to what information he
decides to gather and if he decides to follow advice. (3)
The size of some of these effects we find is not trivial. In our
experiment, subjects are able to search for information about the
properties of a totally unknown probability distribution by either
asking for information about its upper tail, lower tail, or mid range.
We find that, when controlling for personality characteristics, women
are more likely to ask for information about the middle as opposed to
the lower tail. (4) In addition, the size of the marginal effects for
the statistically significant components of the Big Five (Neuroticism,
Extraversion, and Conscientiousness) on demand for the middle rather
than the lower tail are similar in size between 0.02 and 0.03 at the
average regressor of 50. The marginal impact of Sensation Seeking on
both categories is -0.014 and -0.023 for the middle and top.
respectively, at an average regressor of about 22.
Similarly, we show that when the information comes from an advisor,
personality comes into play through two channels. First, it affects the
recommendations made by the advisor. Second, it determines the
likelihood that the advice is followed. In particular, our results
indicate that the impact of risk aversion and personality on choice when
the information comes through an advisor differs significantly from
their impact in environments where subjects endogenously gather their
own information. In particular, the impact of both risk aversion and
personality of the DM is no longer correlated to choice once advice is
offered. A similar inclination to follow advice is seen in Schotter and
Sopher (2003, 2007). However, we know that some people are more likely
to take advantage of advice than others, and the question then arises as
to what types of people are more likely to follow advice when offered.
Here again we find evidence that personality variables are likely to be
a key determinant of who follows advice and also on what type of advice
is offered. (5)
There is ample evidence that personality, as measured by the Big
Five and the SSSs, correlates to important economic decisions. For
instance, Nyhus and Pons (2005) investigate the influence of the Big
Five factors on wages using household survey data from the Netherlands.
They find that the economic returns of the personality factors in wage
determination vary between educational groups and across genders. In a
similar spirit, Mueller and Plug (2006) use the Big Five Scale to
investigate how personality affected the earnings of a large group of
men and women who graduated from Wisconsin high schools in 1957 and were
reinterviewed in 1992. In a political economy context, Morton, Tyran,
and Wengstrom (2011, 2016) analyze data from a large sample of the
Danish population to study the effects that the Big Five may have on
political ideology and whether or not these traits can explain the
ideological gender gap. They find that the differences in traits between
men and women explain the tendency to be left- or right-wing oriented
through a direct effect on ideology and through the indirect effect that
these traits have on income. Muller and Schwieren (2012) study the
impact of the Big Five on behavior in the trust game and find that there
is a higher correlation to the first mover's behavior. Filiz-Ozbay
et al. (2017) study the role that cognitive ability, gender, and
personality traits have on behavior in the gift exchange game. They find
that one of the traits of the Big Five Scale, Agreeableness, plays an
important role in explaining the results. Anderson et al. (2016) analyze
a large data set for truck drivers in the United States and find that
personality traits (as measured by the Big Five) are better predictors
for credit score, job persistence, and heavy truck accidents than
economic preferences. Proto and Rustichini (2015) study the relationship
between income and life satisfaction by looking at the Big Five
personality traits and find that different traits mediate the effect
that income has on aspirations and life satisfaction. In a survey,
Borghans et al. (2008) summarize evidence from various psychology papers
about the importance of personality traits in predicting socioeconomic
outcomes including job performance, health, and academic achievement.
They show correlations for the predictive validity of IQ and the Big
Five personality factors on leadership ratings, job performance,
longevity, college grades, and years of education. Finally, Zuckerman
(2007) reviews over 2000 published articles on Sensation Seeking
self-report questionnaires to show that collectively these studies have
established that Sensation Seeking predicts risky driving, substance use
and abuse, smoking, drinking, unprotected sex, juvenile delinquency, and
adult criminal behavior.
Psychologists have studied decisions in the financial realm and how
these relate to personality. However, those studies do not really speak
to economists as they typically do not consider the DM's risk
aversion. Furthermore, in line with their experimental tradition, these
studies are not incentivized and their focus is often different, for
example, on whether considering personality adds anything to intellect
alone. Our study, although not designed to answer these questions, sheds
light on some of that debate. For instance, the fact that risk aversion
explains some of the decisions we observe, even controlling for
personality, indicates that it is a feature of the DM that is not
subsumed in the personality traits considered by those scales.
On a methodological level, the paper's contribution is to
design a data set that speaks to the question at hand by revealing
aspects of preferences and information sets that are not available in
observational data sets. First, in one treatment, the experiment allows
us to learn what features of uncertain distribution subjects want to
learn about. This choice is incentivized within the experiment and thus
the exhibited behavior reveals a preference that could be an important
component of modeling choice under uncertainty. Similarly, the design
allows us to observe not only the recommendation that advisers offer to
DMs, but also which feature of the probability distribution they decide
to focus on. These aspects of the design are novel and tie the data set
generated to the question in a unique way.
Finally, it is important to point out that this is first and
foremost an empirical paper that, we believe, is the first to establish
a connection between personality and information gathering under
uncertainty. While we do not provide a theory to explain the behavior we
observe, we do present a number of theoretical approaches that could be
used to construct one in Section V.
Despite the empirical nature of the paper, it does make a point
that we think is relevant for theorists. While the literature on
decision making under uncertainty has tended to treat the degree of
uncertainty that DMs face as fixed or exogenous, in reality the degree
of uncertainty is endogenous in the sense that DMs are able to modify it
via information gathering activities. This fact, we believe, makes
decision making under uncertainty a two-stage process where in the first
stage the DM needs to decide whether to gather information and, if so,
how. In the second stage, given the information gathered and the updated
priors about the distributions faced, the DM needs to make a choice.
What is needed then is a theory of both information gathering and
decision making under uncertainty. In this study, we document the
importance of the first stage.
The article proceeds as follows. In Section II, we describe our
experimental design and in Section III we analyze our results. The data
analysis proceeds by first testing our main hypothesis. This is followed
by an exploration of the various ways in which personality can have an
impact in the settings we study. In Section IV, we present some related
literature, while in Section V we present several possible theoretical
approaches to modeling the influence of personality on information
gathering under uncertainty. Finally, in Section VI we offer some
observations and conclusions.
II. EXPERIMENTAL DESIGN
The experiment is composed of three treatments, which we call
Control, Priority, and Advice. In each treatment, subjects have to
choose between pairs of probability distributions under different
information conditions. For all treatments, each of the sessions is
divided into two parts. The first part of the experiment involves
measuring various personality and risk aversion characteristics of the
subjects by administering three tasks: the SSS (Zuckerman 1994), the Big
Five Personality Scale (Costa and McCrae 1992), (6) and the Holt-Laury
risk aversion task (Holt and Laury 2002). The second part of the
experiment varies by treatment but always involves six choices over
lottery pairs. (7)
The probability distributions defining the lotteries are
represented by the four distributions in Figure 1. The specific
probabilities of each of these distributions are in Table A2 in the
Appendix.
In the rest of the article, we will refer to the distributions with
the following shorthand: L (Low variance) for the top left distribution,
SR (skewed right) for the top right distribution, G/L (Gains and Losses)
for the bottom left, and U for the bottom right. In most cases the
lowest possible outcome is 0 and the highest possible outcome is 20,
except for the G/L distribution which also puts positive probability
on--5 and 25. These distributions were chosen because they are all very
different from each other in important ways, such as the variance, but
they all share the same mean of 10. The subjects are informed that the
means are identical, and of the lower and upper bounds of the support.
Thus, in a standard Expected Utility model, if subjects have complete
information about the distributions, their choice should be completely
determined by the risk preference of the DM and the properties of the
lotteries.
Given the four lotteries, we can define six lottery pairs covering
all possible pair-wise combinations of these distributions. In the
Control treatment, subjects have to choose one of the lotteries from
each of the pairs of distributions that are presented to them
sequentially on their computer screen (referred to as Left or Right
distributions). This treatment serves as our control since subjects have
complete information about the probability distributions that
characterize these lotteries, thus representing an environment solely of
risk.
In the Priority treatment, subjects face the same choices as in the
Control, but they do not observe the distributions (the instructions
only inform them that the distributions all have mean 10 and all range
between -5 and 25). Instead, they are given the opportunity to learn
some salient features of each pair: the sum of the probabilities for
outcomes 4 or less, the sum of the probabilities for outcomes 16 and
above, or the sum of the probabilities for outcomes between 8 and 12.
Henceforth, we will refer to these pieces of information as the Bottom
(B), Top (T), or Middle (M) sections of the distributions. (8) Before
choosing among lotteries, subjects are asked to state their priority
over these three pieces of information by choosing which one they would
like to receive the most, second most, and third most. Then, for each
choice problem, a computer randomly determines if they will be shown
one, two, or three pieces of information (each is equally likely), and
based on the priority they state and the random number generated by the
computer, they are given the relevant information and then they make
their choice. (9) Subjects only state their preference over these three
pieces of information once, and that preference is relevant for each
pair-wise choice, but a different random number is generated for each of
the six choice problems they face, so for different choices they receive
different amounts of information.
Finally, in the Advice treatment subjects are matched in fixed
pairs. Half of the subjects are given the role of Advisors and the other
half of DMs, and subjects remain with that role for the rest of the
session. The Advisors' screens display the distributions relevant
for each of the six choice problems, but the DMs see only blank screens.
The Advisors, after observing the distributions, have to make a
recommendation to the DM they have been matched with as to which lottery
to choose (Left or Right), and justify their advice using one of the
three types of information presented in the Priority treatment: Bottom,
Top, or Middle. For example, an advisor can give one of the following
pieces of advice: "Choose Left instead of Right because the
probability of receiving 4 or less is 0.498 with Right but 0.159 with
Left," or "Choose Right instead of Left because the
probability of receiving 16 or more is 0.498 with Right but 0.185 with
Left," or "Choose Left instead of Right because the
probability of receiving an outcome between 8 and 12 is 0.293 with Left
but 0.0000061 with Right." (10) DMs do not observe the
distributions; they only observe advice for either the left or right
distribution and the reason given to them. Once they receive their
advice, they have to choose one of the lotteries. Note that the
information available to the DMs is the same in this treatment as in the
Priority treatment (when they receive only one piece of information),
but in this treatment it comes in the form of exogenous advice rather
than solicited information.
At the end of the experiment, one of the choice problems is
selected at random and the choice of the DM is played out. Advisors are
paid $3.33 for each of their recommended decisions that are followed.
Hence, advisors have incentives to at least offer advice that they think
is persuasive. (11) DMs are paid the outcome of the lottery chosen. All
subjects are also paid a $13 show-up fee.
For each treatment, two sessions were conducted, for a total of six
sessions. In total, there were 123 subjects (41 in the Control, 42 in
the Priority, and 40 in the Advice treatment). The software was z-tree
(Fischbacher 2007) for the first part and multistage (CASSELL [UCLA] and
SSEL [Caltech]) for the second part. All subjects were undergraduate
students at New York University (from all majors).
As mentioned above, our experimental design is constructed to
investigate our main hypothesis. We have a treatment where there is pure
risk and two where there is uncertainty, which can be mitigated by
either information gathering (the Priority treatment) or advice (the
Advice treatment). Seen through this lens, our design is easily
motivated and a natural starting point. Intuitively, if personality is
to be related to the choices of subjects, one would expect it to be in
an environment with uncertainty (like the Priority and Advice
treatments), rather than in a risky environment with complete
information (Control treatment). This is why we choose to present
subjects with two different types of lottery choice problems, one where
there should be little scope for personality (other than risk aversion),
and one where, due to uncertainty, there might be room for personality
characteristics to influence behavior. The key to uncovering the impact
of personality is to ask subjects to decide on what information they
want to observe because, intuitively, different people might want to
know different characteristics of the decision they face. Standard
economic theory is silent about what parts of a probability distribution
a person should seek information about, but intuition suggests that
people's personality may influence this decision. In the presence
of an adviser, it appears like the adviser would attempt to use
information about the part of the distribution that would be most
convincing to most people, but, depending on the personality of the
advisee, this information may or may not be convincing. Hence,
intuitively one would expect personality to play a role in both
treatments under uncertainty, compared to a control treatment with
complete information about the lotteries. This is the theme around which
we have designed the experiment.
III. RESULTS
This section is divided into two parts: one testing our main
hypothesis and the other examining the connection between personality,
information acquisition, and advice taking and advice giving. To analyze
our results on the information acquisition, we present a set of
observations that we then substantiate using our data.
Before we proceed, however, let us pause and briefly describe the
results of our personality and risk preference elicitation exercises to
give an insight into what the sample of subjects looks like and to
verify that our sample does not vary dramatically from the norm
associated with these personality scales. We also summarize the choices
of our subjects over lotteries in the three treatments. This is followed
by the test of Hypothesis 1. Following that, we investigate the various
channels through which personality can play a role in these
environments. While this final part of the section is exploratory, we
hope that it yields interesting insights that can serve as the basis of
future research.
A. Personality Attributes
Table 1 contains summary statistics about gender. risk preferences,
and personality traits of the subjects that participated in the
experiment.
Female is an indicator variable that takes the value of 1 for
female subjects and 0 otherwise. The Relative Risk Aversion (RRA)
coefficient takes the value of the mid-point of the interval of a RRA
specification of utility implied by the Holt-Laury choices of each
subject. (12) Neuroticism, Extraversion, Openness to Experience,
Agreeableness, and Conscientiousness are the Big Five personality traits
and are explained in more detail in Table 2. Note that the Big Five
questionnaires are designed to give a mean of 50 with a standard
deviation of 10 for each trait. The score on the SSS is presented as an
aggregate score, and also separated into its components: Thrill and
Adventure Seeking. Experience Seeking, Disinhibition, and Boredom
Susceptibility (see Zuckerman 1994). The SSS is calibrated to result in
a mean of 23.77770 and a standard deviation of 5.6 for males, and a mean
of 19.0 and standard deviation of 5.7 for females in the United States
(Zuckerman 1994). As we can see, our sample appears to conform to these
norms.
It is important to note that these personality scales were not
created to predict economic decision making. These measures are a
natural starting point because they are well established in the
psychology literature and have been found to correlate well with life
outcomes, but not necessarily with the type of controlled decisions we
study. Hence, the economic interpretation of these dimensions of
personality might be difficult (see the discussion in Section VI).
Table 3 shows the pair-wise correlations among the different
personality measures, the female indicator, and the RRA coefficient. It
is interesting to note that risk aversion is not significantly
correlated to any of the Big Five personality traits and its
correlations with the components of the SSS are not high. (13)
The choices of our subjects over lottery pairs in the three
treatments are summarized in Table 4. We consider the Control treatment
as the baseline because this is the only treatment where subjects have
full information about the distributions they face. As a result, we
might consider the choices made there as reflecting the subjects'
true preferences over these distributions. Note that in each pair, the
distribution on the right is the riskier one.
One result that is clear is that the choices made for the same
lottery pairs change as we move across treatments. For example, while
the SR distribution is greatly preferred to the L in the Control
treatment, the opposite is true when we move to the Priority treatment.
These results should give readers a first indication that
information gathering can have a dramatic impact on choice because the
only thing that varies across these treatments is the information
available to subjects and the manner in which that information is
acquired. If one considers the choices made in the Control treatment as
the welfare maximizing choices for the subjects, since they have full
information there, our results from Table 4 demonstrate the impact on
welfare of different informational conditions in the presence of
uncertainty. As we will see, a large part of this variation can be
explained by the different, personality-influenced, information
acquisition strategies that subjects use in these different treatments.
B. Impact of Personality on Choice in Risky and Uncertain
Environments: Test of Hypothesis 1
As stated above, in our experiment we expect that if personality is
to have an impact on choice it is likely that it will only be felt in
environments characterized by uncertainty and not risk. This expectation
is summarized by our main hypothesis, Hypothesis 1.
HYPOTHESIS 1: Personality characteristics, either the Big Five or
Sensation Seeking, only affect choices in treatments with uncertainty,
not in the Control treatment involving solely choices under risk.
Table 5 reports the main regressions that test this hypothesis and
find support for it. They are probit regressions where the dependent
variable is the choice of the riskier distribution (the marginal effects
are reported in Table 6). In the case of the priority treatment, this is
for the subset of cases where subjects received only one piece of
information (as these are the cases with the most uncertainty and the
closest to the condition in the Advice treatment). The regression for
that treatment also includes a set of indicator variables capturing the
subjects preferences for information: which of T, M, or B. is ranked
first, second, and third. In addition, the regression for the Advice
treatment includes the advice given, a dummy variable indicating if the
suggestion was to choose the risky or safe distribution, and two
indicator variables distinguishing if the evidence provided (the reason)
was about B, M, or T. These additional variables are not reported to
keep the table easier to read. The key results are reported as "p
value: test of H1," which indicates that the joint test that
personality variables are jointly significant cannot be rejected for the
control, but can be rejected at the 10% and 5% levels for the Priority
and Advice treatments, respectively. One may worry about the fact that
with multiple tests, the false discovery rate is not the same as the
confidence level of the test. We note that even with the crudest of
corrections for multiple hypothesis tests, the Bonferroni correction,
the joint test is still rejected at the 5% level in the case of the
Advice treatment. (14) With the correction, it misses the significance
threshold at the 10% level in the Priority treatment however, since the
corrected threshold is 0.333. (15)
C. Personality, Information Acquisition, and Choice
Having established the evidence with respect to our main question,
what follows is a more exploratory analysis investigating the respective
roles of the specific factors we control for on the various steps
leading to a final choice. As we unpack the various treatments, multiple
hypotheses tests will be performed with no correction "a la"
Bonferroni. As will become clear, the various dependent variables
explored are highly correlated and as such the proper corrections for
multiple hypothesis tests are not straightforward. Hence, one should
interpret these results with caution, but we do point out that some of
the results, taken together, appear to form a logical and plausible
chain.
Before moving to the unpacking of the Priority and Advice
treatment, we mention the other results from Table 5. First, there is
the finding that risk aversion matters for choices in the Control and
the Advice treatments. As expected, more risk averse subjects exhibit a
higher likelihood of selecting the distribution with lower variance.
Surprisingly, gender only has a significant impact in the Advice
treatment where it leads women to make riskier choices than men.
In the Priority treatment, the dummy variables indicating the order
preference for information are jointly significant (p<0.01). In the
Advice treatment, the results indicate that the advice given affects the
choices (p<0.01) as well as the reason used (joint hypothesis that
the dummy variables are equal to the excluded category: p < 0.1).
D. Information Acquisition under Uncertainty: The Priority
Treatment
For the Priority treatment, the preferences for learning different
features of the distribution are represented by the popularity of each
possible permutation of information demand and of the most popular first
choice in Table 7. As we can see in the last column on the right, half
of the subjects want to learn about the bottom part of the distribution
first, with the other two options almost equal among the rest of the
subjects. The most popular order (for one third of the subjects) is to
learn first about the bottom, followed by the top and finally about the
middle.
OBSERVATION 1. Some personality traits, risk preferences, and
gender affect the demand for information under uncertainty.
Table 8 shows the results of multinomial probits with the same set
of regressors as for the probits studying choices, but with the
information ranked first as the dependent variable. First, note that
personality measures are not jointly significant. However, some traits
are significant, in particular when considering the impact of focusing
on the middle rather than the bottom of the distribution. Higher scores
on the Neuroticism, Extraversion, and Conscientiousness scales increase
the likelihood of requesting information about the middle first, rather
than the bottom. A higher score on the SSS results in a higher
probability of demanding to know about the bottom rather than the top
first. Risk aversion also appears to have an impact, that is to say,
more risk averse subjects are more likely to want information about
outcomes in the bottom of the distribution, rather than the middle.
Women are more likely to rank the middle instead of the bottom first as
compared to men. The size of some of these effects is not trivial. The
difference between men and women in the likelihood of asking about the
middle rather than the bottom is 0.41. The size of the marginal effects
for the statistically significant components of the Big Five on demand
for the middle rather than the bottom are similar in size, between 0.02
and 0.03 at the average regressor of 50. Similarly, risk preference has
an estimated marginal effect of -0.36 with an average regressor of 0.49.
The marginal impact of Sensation Seeking on both categories is -0.014
and -0.023 for the middle and top, respectively, at an average regressor
of about 22. For all other regressors, the marginal impact is much
smaller in the case of the top category, in most cases smaller by at
least a factor of 10.
In short, when subjects face an informationally sparse environment,
some aspects of personality appear to have a significant impact on what
information they acquire.
OBSERVATION 2. The information received by DMs affects the
incidence of riskier choices in environments of uncertainty where DMs
demand information according to their priority ranking.
As we discussed before, how people choose when they are only
partially informed about the probability distributions they face is, to
a large degree, a function of the information they have chosen to gather
prior to making their choice. Given that all distributions in the
experiment have the same mean, we look at the impact of information
acquisition on the riskiness of the choice made, that is, whether or not
they choose the higher variance distribution given the information they
have gathered. We have already established that personality plays some
role in determining what information the DM seeks, next we establish the
presence of a link between the information received and choice.
Table 9 shows how the information about the distributions actually
observed affects choices in the Priority treatment (viewing all three
features is the default). Clearly, when only one piece of information is
observed, which one it is affects the decision. To get a sense of the
size of these impacts, Table 10 shows the frequency of riskier choices
in the Control treatment and in the Priority treatment, depending on
which piece of information is received for the cases where subjects
observe either one or three pieces of information. Note that subjects
who only receive information about the Top of the distribution choose
the riskier option 81% of the time, while subjects that observe
information about the Bottom and Middle choose the riskier option 39.47%
and 18.18%, respectively. This suggests that demanding and receiving
information about the Top may lead to riskier choices. When subjects
observe all three pieces of information the frequency of riskier choices
is 31%, not too different from what is observed in the control. (16)
In an informationally sparse world, that is, where uncertainty is
present, DMs may resort to many devices to help them make choices. Here
personality can come in via the likelihood of following the advice given
by the advisor.
OBSERVATION 3. Under uncertainty, personality traits and gender
affect the probability with which a DM follows advice.
As observed in prior research (see Schotter and Sopher 2003, 2007)
subjects appear eager to follow advice. In fact, in our experiment
subjects follow the advice given 85% of the time. This does not mean,
however, that personality is not important for advice following. Table
11 shows the results of probit estimates where the dependent variable
takes the value of 1 if the subject follows the advice given, and zero
otherwise. The independent variables are risk aversion, gender,
personality measures, and dummies for the advice as well as the reasons
given as advice.
The main determinants of whether advice is followed or not are
gender and personality. The personality measures are jointly significant
(p < 0.01). For example, people with high scores on Extraversion and
Agreeableness appear to follow advice more often (marginal effect of
0.008 and 0.01, respectively, at the average regressor of about 50), and
people with high scores on Openness to Experience and Conscientiousness
follow advice less often (marginal effect of -0.009 and -0.008,
respectively, at the average regressor of about 50). Also, in all of the
specifications women appear to follow advice more often than men. Risk
aversion does not explain the decision to follow advice.
While we have established a link between personality and advice
following there may also be a personality component in advice giving.
This is important because if the type of advice given is determined by
the personality of the advisor, and the likelihood of it being followed
depends on the personality of the advisee, then the match between
advisors and advisees may be important in determining the effectiveness
of advice.
OBSERVATION 4. The advice given (the suggested choice) is
correlated to gender and personality for advisors.
To support this observation, we present Table 12 which uses the
data from our subject advisors and shows the results of probit
estimations where the dependent variable takes the value of 1 if the
subjects advised the choice of the riskier distribution, and zero
otherwise. The independent variables are risk aversion, gender, and
personality measures.
In the case of advice giving, the personality measures are jointly
significant (p<0.1). As we can see, females appear to give the
riskier option as advice more often than men (marginal effect of 0.125
at the average regressor of about 0.5), and subjects who are more open
to experience appear to give the riskier advice less often (marginal
effect of -0.007 at the average regressor of about 50). The fact that
females suggest riskier options more often is interesting since women
typically are risk averse and sometimes more so than males when making
choices for themselves in situations of risk (see Croson and Gneezy
2009; Niederle 2016), suggesting a kind of split attitude for females
when it comes to choosing for themselves when facing risk and advising
others when facing uncertainty. Finally those who score higher on
average on the SSS and on Conscientiousness tend to suggest the more
risky choice (marginal effect of 0.007 and 0.006, respectively, at the
average regressor of about 50).
OBSERVATION 5. Gender and elements of personality have a
significant impact on the type of information offered as justifications
by advisors.
As we have mentioned before, in our experiment advice giving has
two parts: a recommendation and a piece of information used as a
justification for the advice. Observation 7 suggests that personality is
relevant for the recommendation but there might be an additional
personality component involved in the type of justification used to
support it. Table 13 shows what factors determine the reason given as
advice, that is, bottom, middle, or top of the distribution, using a
multinomial probit where the base outcome is to give bottom as advice.
Again the various measures of personality are jointly significant (p
< 0.01), with sensation seeking decreasing the chance of a M
suggestion compared to B, while Openness reduces the use of T compared
to B. The table also reveals that females are less likely to justify a
recommendation by pointing to the bottom of the distribution.
E. Summary of Results
Since we have presented a fair number of results, it might make
sense to pause and take stock of what we have learned before proceeding
to a discussion and our conclusions. The key results are the following.
Consistent with Hypothesis 1. the key determinant of choice under risk
is risk aversion. However, personality, risk attitudes, and gender
affect multiple aspects of behavior under uncertainty. In particular,
there is strong evidence that personality directly affects choices in
the Advice treatment and some evidence that it does in the Priority
treatment. In addition, personality also appears to have an indirect
effect on choice via the information demanded or the likelihood of
following advice.
In discussing our results further, it is useful to make a
distinction between direct and indirect relationships. The relation
between information demand and personality in the Priority treatment,
between following advice and personality in the Advice treatment, and
between personality and advice giving in the Advice treatment are all
direct relations. However, for instance, the relationship between
personality and choice in the Priority treatment is indirect since it is
mediated by the intermediate step of information demand. To summarize
these direct and indirect relationships, Table 14 presents the main
statistically significant relationships we have uncovered.
As we can seen from Table 14, when it comes to information demand,
Neuroticism, Extraversion, and Conscientiousness all increase the
probability that a subject ranks M first rather than B. With respect to
indirect relationships, in our study personality affects the information
demanded, which in turn affects choices. Table 14 contains the results
for the Priority treatment when we pool the data over the cases where
subjects received information about B, M, and T, and hence all have the
same information. In this case, personality is correlated to the
riskiness of choices, even after controlling for information
preferences. In particular. Extraversion has a positive impact on the
riskiness of choices while Agreeableness has a negative impact.
With regards to the Advice treatment. Extraversion and
Agreeableness increase the likelihood of following advice, while
Openness and Conscientiousness decrease it. Openness decreases the
probability that an advisor gives the riskier advice while
Conscientiousness and SSS increase it. Personality does not determine
riskier choices in the Advice treatment, even after controlling for the
type of advice given.
Risk aversion also plays an interesting and subtle role in all of
the relationships discussed above. As mentioned earlier, it is the only
statistically significant predictor of choice in the Control treatment.
With respect to information demand. Risk aversion decreases the demand
for M in the priority treatment, so that more risk averse agents are
less likely to rank M first as opposed to B. With respect to advice
giving, Risk aversion increases the likelihood of giving B as a
justification for choice in the Advice treatment and it decreases the
likelihood of riskier choices, but it does not have an impact on who
follows advice.
When we turn our attention to gender, we find that while it does
not have a conclusive impact on choices in the Control treatment, it
does increase the demand for M by females in the Priority treatment. In
the advice treatment, females are more likely to follow advice and more
likely to give riskier advice.
One thing that is important to point out as we look across our
regression results is that while personality traits are significant
across specifications, it is not always the same traits, nor is it
necessarily in the same direction. This is not surprising, however,
since each regression explains a different phenomenon. In particular, in
each regression subjects are presented with different types of choices
(or tasks), and there is no reason a priori why the same personality
traits should explain different tasks in the same way. For example,
while Neuroticism and Conscientiousness increase the likelihood of
asking for information about the middle of the distributions as opposed
to the bottom, it is Conscientiousness and Agreeableness that are
significant in determining whether a subject follows advice. Neuroticism
appears not to be significant here. This finding is somewhat expected
since the personality characteristics that are responsible for
information demand do not have to be the same that determine whether a
person is more likely to follow advice. Our point is that the Big Five
and SSS personality scales appear to be correlated to certain types of
behavior under uncertainty, but which constellations of traits are
important for any given type of behavior varies with the task performed
by subjects.
IV. RELATED LITERATURE
There is relatively little literature that directly relates to the
questions of personality, information acquisition, and choice discussed
here. The only study we know of that relates personality to information
demand is Gerber et al. (2011). (17) This study correlates the Big Five
to political interest, knowledge, and the consumption of different types
of political media. They use data from an internet survey of American
registered voters which attempts to be close to a representative sample
of the population. The survey they use was administered before the 2008
election and contains data on 8,664 individuals. They find that
Openness, Agreeableness, and Extraversion are all positively correlated
to the consumption (in the previous week) of at least one of the three
forms of media they study: television, internet or newspaper. When they
focus on whether the individuals watched national or local news, what
they find is that Agreeableness and Extraversion increase the likelihood
of watching national news, while Agreeableness and Conscientiousness
increase the chance of watching local news and Openness decreases it.
Clearly, their exercise is very different from ours. However, one
comparison which may be legitimate is that in their case, each of the
five personality dimensions matters for some aspect of whether
individuals consume news or not, and in what format, except for
Neuroticism. In our case, Neuroticism does affect the kind of
information demanded, but Openness and Agreeableness never come into
play. This could simply be because the realms of these two studies are
so different, or because the decision to consume some information is
different from the decision to choose what information to focus on.
We also analyze the impact that personality has on choice. The
studies that would appear the most relevant for the impact of
personality on risky choices are those related to the role of
personality in gambling. (18) McDaniel (2002) finds that the SSS is
positively correlated to interest in gambling in a sample of 555 adults
(18 and above) from the eastern United States surveyed by telephone.
(19) Lauriola and Levin (2001), using a sample of 76 Italian adults,
conclude that Openness and Neuroticism affect riskier choices (they
offer a series of choices between a safe alternative and a riskier one).
Furthermore, the impact of Neuroticism varies for the loss and gain
domains. However, their results are either not statistically significant
or barely so. Our results suggest a more complicated role for
personality, one where the impact of personality traits on choice
interacts with the way in which the information is being presented to
the subject. Nicholson et al. (2005) study a sample of students and
executives, including MBAs and executives in training programs at the
London Business School (sample size 1669) looking to validate a set of
survey questions on the propensity to take risks in various areas
(physical status, lifestyle, and livelihood, which includes career and
financial risk). (20) The answer to their question on financial risks is
significantly correlated to all five domains of the Big Five. More
specifically, they find a positive relation to Extraversion and
Openness, and a negative relation to Neuroticism, Agreeableness, and
Conscientiousness. They also report that males score higher on risk
taking, but this correlation appears to have been established without
controlling for personality (and similarly, the impact of personality is
established without controlling for any other factors). Similar to them,
we find that males take riskier choices (in our case even controlling
for personality and risk aversion).
With respect to gender, there is some evidence that women are more
risk averse than men when making decisions under risk (see Croson and
Gneezy 2009; Eckel and Grossman 2008, for two surveys). (21) Borghans et
al. (2009) study how risk and ambiguity aversion vary across men and
women and whether the differences in these parameters can be explained
by personality measures. They find that differences in ambiguity
aversion cannot be explained by personality traits. However, similar to
Croson and Gneezy (2009), they find that women are more risk averse than
men and that differences across risk aversion parameters can be
explained by personality measures, in particular by Agreeableness and
Neuroticism from the Big Five Scale, and by ambition, as measured by
Duckworth et al. (2007). Eckel and Grossman (2002) study risk attitudes
between men and women and measure personality characteristics using the
SSS. They find that women are consistently more risk averse than men and
that men appear to overestimate the risk aversion of women when
predicting choices between gambles. However, they find no significant
gender difference in the overall SSS scores and they find very low
predictive power of the SSS on gamble choices.
V. MOTIVES FOR INFORMATION GATHERING: SOME THEORETICAL APPROACHES
There are two main questions that we need to answer in order to
fully understand how personality affects decision making under
uncertainty. First, why do people desire information in the first place
and why a particular type of information? Second, what is it about a
DM's personality that leads him to desire the type of information
he does? Below we sketch a few of our thoughts on the motives that
people may have for information gathering. While a complete theory of
personality is beyond the scope of this study, we do hope that our
thoughts below can be useful to others who are interested in pursuing
these topics.
A. Pessimistic Priors
As shown by Sharpe, Martin, and Roth (2011), there is a strong
statistical correlation between dispositional optimism and four of the
Big Five personality traits (Neuroticism, Extraversion, Agreeableness,
and Conscientiousness). Probably the most straightforward answer to the
question of how personality can affect information gathering works
through a subject's level of optimism (or pessimism) about the
unknown distributions he faces and the relationship of personality
traits to this characteristic. Under this interpretation, the subject
remains an expected utility maximizer but his level of optimism simply
affects the type of priors he has over the payoff distributions he
faces. While one might think it natural for pessimists to concentrate
their attention on the left tail of the distribution, with optimists
caring more for the right tail, this may not necessarily be the case.
However, as long as pessimists and optimists seek different information,
then all that is needed is to connect a subject's level of optimism
with some constellation of personality traits in order to explain the
impact of personality on information gathering.
The type of ambiguous decision environments we place our subjects
in are relatively scary when compared to environments characterized
solely by risk. As such, they may call forth some type of ambiguity
averse behavior. A famous theory of decision making under ambiguity by
Gilboa and Schmeidler (1989) suggests that when faced with ambiguity, a
DM is likely to assume he is facing the worst possible probability
distribution in the set of feasible distributions and choose that action
which is best against this pessimistic assumption. So Gilboa and
Schmeidler's DMs are extremely pessimistic when faced with
ambiguity.
However, not all subjects are likely to be this pessimistic and.
hence, we might expect some variability across people concerning how
pessimistic they are. To this end. Ghirardato. Maccheroni. and Marinacci
(2004) have created an alternative theory where DMs choose as if they
were characterized by a combination of pessimism and optimism with a
weight, [alpha]. defining the exact convex combination of the two. If a
theory of personality and decision making under ambiguity is to be
formulated, one might investigate what factors determine a DM's a.
We expect that personality variables are likely to play a role here and
hence in determining the information that such types find desirable.
B. Probabilities Inside the Utility Function
A second possibility for why personality affects information
acquisition may stem from the idea that the prize space over which a
person's utility function is defined contains not only tangible
outcomes but also emotional states defined by probability distributions.
As Caplin and Leahy (2001) have demonstrated, the utility of a
particular outcome may depend both on the anticipated outcome itself and
on the probabilities that this outcome may occur, with the probability
entering independently into a DM's utility function. (22) This is
particularly true when the decision has an emotional component to it,
such as when medical decisions are being made and anxiety about outcomes
is paramount.
In such a situation, different personality types may be inclined to
search for different types of information because their utility at the
moment of decision making is affected by the beliefs they hold at that
moment. Neurotics may want to assure themselves that they are making a
choice that, a priori, guarantees them either the largest minimum
outcome or perhaps, as our regressions indicate, the largest middle
outcome. People who rank high on the SSS or Openness to Experience may
derive utility from thinking that they are more likely to receive a good
outcome and hence inquire about the top of the distribution, and so on.
Whatever their motive, the idea here is for DMs to choose their beliefs
optimally much like Brunnermeier and Parker (2005) suggest. They search
for information in order to find those beliefs they would like to hold
and we suspect that their preferred beliefs are a function of
personality variables.
The two sketches of a theory of personality and decision making
outlined above are certainly not exhaustive. Other theories can be
easily constructed. Still, they all would need to share some common
features. First, the role of personality may be dramatically different
as we move from risky to ambiguous environments. Second, the information
people gather will depend on their personality.
Two more theories that might appear like plausible avenues through
which personality might affect information-gathering decisions are
heuristics and preferences over higher moments of the distribution. We
review these two possibilities below. Note that while these theories are
equally applicable for decision making under risk and uncertainty it is
only in the ambiguous situations where they have an influence 011 the
information-gathering strategies of subjects.
C. Heuristics
There has been a considerable amount of work done by psychologists
(see Brandstatter, Gigerenzer, and Hertwig 2006; Gigerenzer 2004, to
name only two), and economists (e.g., Rubinstein 1988), indicating that
in a risky choice environment, where DMs see all relevant probability
distributions, rather than weighting, multiplying, and adding
probabilities and payoffs as is expected of them under the Expected
Utility Hypothesis, they employ a heuristic where they proceed
lexicographically and compare features of lotteries, that is, their
minimum payoffs or the probability of a minimum payoff. Rubinstein
(1988), for example, demonstrates that when comparing two lotteries DMs
compare the similarities of the probabilities and payoffs in a
lexicographical manner. Brandstatter, Gigerenzer, and Hertwig (2006)
proceed in a similar manner but assume a fixed order for comparisons
using what they call a "priority heuristic" which compares the
minimum gain of two gambles, then the probability of the minimum gain,
and then finally the maximum gain. This priority order is justified
empirically rather than theoretically and, as is true for Rubinstein
(1988), is assumed to be the same for all individuals.
There are some modifications that need be made on the Brandstatter,
Gigerenzer, and Hertwig (2006) and Rubinstein (1988) theories before
they can be employed here. First, those theories were constructed for
complete information settings and not for the settings we examine under
uncertainty. However, it is obvious that our subjects could use such
heuristics simply by asking questions in the order most closely
associated with either heuristic and by modifying it where necessary.
Furthermore, Brandstatter. Gigerenzer. and Hertwig (2006) and Rubinstein
(1988) assume that all people search identically using their heuristics.
Clearly, we assume heterogeneity across DMs and assume that this
heterogeneity can be explained by personality. What is missing is a
theory that connects personality and heuristic choices (and hence
information gathering).
D. Preferences over Higher Moments
In recent years, a number of empirical and theoretical papers have
been written indicating that individuals have a preference for
(positive) skewness in the distribution of payoffs they face and that
risk averse individuals are prepared to accept a lower expected payoff
or a higher level of overall riskiness if the distribution of payoffs is
more skewed to the right. (23)
These results may have direct relevance for the type of information
inquiries we might see in experiments like ours since such inquiries may
be aimed at finding out information about these higher moments. Eliaz
and Schotter (2010) demonstrate that if a DM has a preference for
confidence in his decision and. as a result, has the probability of
making the correct decision as an argument in his utility function, he
will have a preference for negative skewness. As a result, he might also
wish to gather information about these higher moments and hence ask
questions that would be informative about them.
If this is the motive for information acquisition, then if we were
to build a theory of personality and choice we would need a model that
connects a subject's personality to his preferences over moments of
a distribution.
VI. DISCUSSION
This study was motivated by a hypothesis that if personality were
to have an influence on choice, it would be in uncertain rather than
risky environments, which is substantiated by the data. We have
demonstrated that personality may have a significant impact on economic
decision making through its effects on information gathering in
environments of uncertainty. The path of this influence is in part
indirect since we establish that differences in personality
characteristics, as measured by the Big Five personality scale and the
SSS, lead DMs to seek out different types of information which then,
conditional on the information observed, alters the decisions they make.
We also show that when information is transmitted by an advisor,
personality influences both the advice given and the likelihood that the
advice is followed.
However, when decisions are made solely under risk, that is, in
environments where the DM knows with certainty the exact probability
distribution he or she faces, personality fails to be a significant
determinant of choice. In such circumstances, what mattes for choice is
the DM's level of risk aversion.
Research in economics has largely focused on understanding how
people make decisions in a world characterized by uncertainty. In this
paper, we are interested in the fact that some of this uncertainty can
be alleviated by seeking information, and we find that in this search to
diminish uncertainty personality plays a role. This finding may have
important implications in various economic settings, such as the
matching between financial advisors and advisees, or the process of
hiring people in organizations.
The decision environment appears to play a crucial role when
studying the effects of personality on choice in the presence of
uncertainty, where the probability distributions faced by the DM are
unknown. On the one hand, the impact of personality on choice appears to
be mediated through information acquisition when DMs choose the
information they wish to acquire. On the other hand, personality ceases
to be important for choice when information is received via advice,
rather than solicited directly. This implies that the decision
environment, defined by how the information is received, matters for
choice. This is plausible because some people tend to follow advice so
diligently that they might ignore the actual information offered to
justify the recommendation.
As we have suggested, if progress is to be made in connecting
personality with decision making, a theory of personality will be
needed. The existing scales to measure personality characteristics are
mostly descriptive and are not designed to predict economic outcomes.
For this reason, the meaning of these dimensions of personality might
not be straightforward to interpret when it comes to economic behavior.
For example, it might be the case that Extraversion, as measured by the
Big Five, is a reliable predictor of how likely someone is to follow
advice. However, what Extraversion captures in terms of economic
decision making and advice taking in that realm is far from obvious. The
specific questions that are used to construct these measures are
difficult to relate to observable economic choices.
One component of such a theory will certainly be the specification
of a link between the different personality characteristics, that is,
Openness to Experience, Neuroticism. and so forth, and information
search. Furthermore, a link will be needed between personality traits
and individual welfare. For example, do neurotics or conscientious types
do better because they gather more relevant information about the world
they face before making decisions, or do they do better because,
conditional on any information gathered, they make better choices?
One might envision a number of theoretical explanations for
information gathering in situations of uncertainty and the role played
by personality. For example, DMs may rely on heuristics when making
decisions under risk (see Brandstatter. Gigerenzer, and Hertwig 2006;
Gigerenzer 2004; Rubinstein 1988). Under these theories when DMs make
risky decisions, rather than weighting, multiplying, and adding
probabilities and payoffs, they proceed lexicographically and compare
features of lotteries, that is, their minimum payoffs or the probability
of a minimum payoff, and so on. Brandstatter, Gigerenzer, and Hertwig
(2006), for example, assume a "priority heuristic" where the
DM compares first the minimum gain of two gambles, then the probability
of the minimum gain, and finally the maximum gain. If such a heuristic
is used under risk, it would be interesting to understand what type of
information would be gathered under uncertainty. Personality may play a
role in this information acquisition stage.
Alternatively, the impact of personality on information gathering
may work through another related personality characteristic, for
example, the degree of pessimism of the DM. For example, Gilboa and
Schmeidler (1989) suggest that when faced with uncertainty, a DM who is
uncertainty averse is likely to assume an extremely pessimistic stance.
This would imply a demand for specific types of information in our
setting. But not all DMs are this pessimistic. Ghirardato, Maccheroni,
and Marinacci (2004) allow for a combination of optimism and pessimism
(weighted by an a parameter, 0 [less than or equal to] [alpha] [less
than or equal to] 1). This suggests that personality may affect decision
making and information acquisition under uncertainty by affecting how
pessimistic or optimistic a DM is and hence, the a they use in making
decisions. Defining a link between the Big Five personality traits, a
DM's degree of pessimism, and information gathering is part of our
future agenda.
APPENDIX
TABLE A1
Correlations between Personality Traits and Ranking
of Pieces of Information in the Priority Treatment
Position Reason Female Risk
in Ranking Aversion
(1) B -0.169 0.126
(2) T
(3) M
(1) B 0.021 0.069
(2) M
(3) T
(1) T 0.066 0.039
(2) B
(3) M
(1) T -0.107 -0.018
(2) M
(3) B
(1) M -0.119 -0.021
(2) B
(3) T
(1) M 0.319 ** -0.203
(2) T
(3) B
Position Neuroticism Extraversion Openness to
in Ranking Experience
(1) 0.17 -0.381 ** -0.122
(2)
(3)
(1) -0.269 * 0.41 *** 0.194
(2)
(3)
(1) 0.075 -0.146 -0.259 *
(2)
(3)
(1) -0.019 0.076 0.189
(2)
(3)
(1) 0.194 -0.093 0.138
(2)
(3)
(1) -0.106 0.152 -0.138
(2)
(3)
Position Agreeableness Conscientiousness Sensation
in Ranking Seeking
(1) 0.032 -0.269 * -0.056
(2)
(3)
(1) -0.04 0.123 0.26 *
(2)
(3)
(1) 0.046 0.042 -0.258 *
(2)
(3)
(1) -0.06 -0.071 0.055
(2)
(3)
(1) -0.196 -0.00 0.134
(2)
(3)
(1) 0.155 0.247 -0.15
(2)
(3)
* Significant at 10%; ** significant at 5%; *** significant at 1%.
TABLE A2
Distributions Subjects Were Shown
Probability
Choice Value L SR G/L U
0 -5 0 0 0.023504 0
1 0 0.000326 0.005157 0.026115 0.350001
2 1 0.000651 0.010314 0.029017 0.105
3 2 0.001303 0.020629 0.032241 0.0315
4 3 0.002606 0.041257 0.035824 0.00945
5 4 0.005212 0.082515 0.039804 0.002835
6 5 0.010423 0.077924 0.044227 0.000851
7 6 0.020847 0.073588 0.049141 0.000255
8 7 0.041694 0.069494 0.054601 7.65E-05
9 8 0.083388 0.065627 0.060668 2.3E-05
10 9 0.166775 0.061976 0.067409 6.89E-06
11 10 0.33355 0.058528 0.074898 2.07E-06
12 11 0.166775 0.055271 0.067409 6.89E-06
13 12 0.083388 0.052196 0.060668 2.3E-05
14 13 0.041694 0.049292 0.054601 7.65E-05
15 14 0.020847 0.04655 0.049141 0.000255
16 15 0.010423 0.04396 0.044227 0.000851
17 16 0.005212 0.041514 0.039804 0.002835
18 17 0.002606 0.039204 0.035824 0.00945
19 18 0.001303 0.037023 0.032241 0.0315
20 19 0.000651 0.034963 0.029017 0.105
21 20 0.000326 0.033018 0.026115 0.350001
22 25 0 0 0.023504 0
Variance 3.907492 26.209200 40.031050 92.224839
ABBREVIATIONS
B: Bottom
DM: Decision Maker
G/L: Gains and Losses
L: Low Variance
M: Middle
RRA: Relative Risk Aversion
SR: Skewed to the Right
SSS: Sensation Seeking Scale
T: Top
U: U-Shaped
doi: 10.1111/ecin.12438
GUILLAUME R. FRECHETTE, ANDREW SCHOTTER and ISABEL TREVINO *
* The authors would like to thank the participants of the
Experimental Economics Seminar at the University of British Columbia and
the members of the Decision Science Seminar at INSEAD for their helpful
comments. In addition, Frechette wishes to thank the NSF via grants
SES-0519045, SES0721111, and SES-0924780. Schotter wishes to thank the
NSF via grant SES-07211 11, and the support of CESS.
Frechette: Professor. Department of Economics, New York University.
New York. NY 10012. Phone 212 992 8683, Fax 212 995 3932. E-mail
[email protected]
Schotter: Professor, Department of Economics. New York University,
New York, NY 10012. Phone 212 998 8952, Fax 212 995 3932, E-mail
[email protected]
Trevino: Assistant Professor. Department of Economics, University
of California, San Diego. San Diego, CA 92093. Phone 858 534 2230, Fax
858 534 7040, E-mail
[email protected]
REFERENCES
Almlund. M., A. L. Duckworth. J. J. Hecknian, and T. Kautz.
"Personality Psychology and Economics," in Handbook of the
Economics of Education, Vol. 4, edited by E. A. Hanushek, S. Machin, and
L. Woessman. North Holland: Elsevier, 2011. 1-181.
Anderson. J., S. Burks, C. De Young, and A. Rustichini.
"Toward the Integration of Personality Theory and Decision Theory
in the Explanation of Economic Behavior." Journal of Behavioral and
Experimental Economics, 64, 2016, 122-37.
Borghans, L., A. L. Duckworth. J. J. Heckman, and B. terWeel.
"Economics and Psychology of Personality Traits." Journal of
Human Resources, 43(4), 2008, 972-1059.
Borghans. L., B. H. H. Golsteyn, J. J. Heckman. and H. Meijers.
"Gender Differences in Risk Aversion and Ambiguity Aversion."
Journal of the European Economic Association, 7(2-3). 2009, 646-58.
Brandstatter, E., G. Gigerenzer, and R. Hertwig. "The Priority
Heuristic: A Process Model of Risky Choice." Psychological Review,
113(2), 2006, 409-32.
Breslin. F. C., M. B. Sobell, H. Cappell. S. Vakili, and C. X.
Poulos. "The Effects of Alcohol, Gender, and Sensation Seeking on
the Gambling Choices of Social Drinkers." Psychology of Addictive
Behaviors, 13(3), 1999, 243-52.
Brunnermeier, M. K., and J. A. Parker. "Optimal
Expectations." American Economic Review, 95(4), 2005, 1092-118.
Caplin, A., and J. Leahy. "Psychological Expected Utility
Theory and Anticipatory Feelings." Quarterly Journal of Economics,
116(1), 2001, 55-79.
Charness, G., E. Kami, and D. Levin. "Ambiguity Attitudes: An
Experimental Investigation." Journal of Risk and Uncertainty,
46(1), 2013, 1-25.
Chiu, W. H. "Skewness Preference, Risk Aversion, and the
Precedence Relations on Stochastic Changes." Management Science,
51(12), 2005, 1816-28.
Costa, P. T., and R. R. McCrae. Revised NEO Personality Inventory
(NEO PI-R) and NEO Five-Factor Inventory (NEO-FFI) Professional Manual.
Odessa, FL: Psychological Assessment Resources, Inc., 1992.
Croson, R., and U. Gneezy. "Gender Differences in
Preferences." Journal of Economic Literature, 47(2), 2009. 448-74.
Duckworth, A. L., C. Peterson, M. D. Matthews, and D. R. Kelly.
"Grit: Perseverance and Passion for Long-Term Goals." Journal
of Personality and Social Psychology, 92, 2007. 1087-101.
Eckel, C. C., and P. J. Grossman. "Sex Differences and
Statistical Stereotyping in Attitudes Toward Financial Risk."
Evolution and Human Behavior, 23(4), 2002, 281-95.
--. "Men. Women and Risk Aversion: Experimental
Evidence," in Handbook of Experimental Economics Results, Vol. 1.
Chapter 113, edited by C. Plott and V. Smith. New York: Elsevier, 2008,
1061-73.
Eliaz, K., and A. Schotter. "Paying for Confidence: An
Experimental Study of the Demand for Non-Instrumental Information."
Games and Economic Behavior, 70(2), 2010, 304-24.
Filiz-Ozbay, E., J. C. Ham, J. H. Kagel, and E. Y. Ozbay.
Forthcoming. "The Role of Cognitive Ability. Personality Traits and
Gender in Gift Exchange Outcomes." Experimental Economics, 2017.
Fischbacher, U. "z-Tree: Zurich Toolbox for Ready-Made
Economic Experiments." Experimental Economics, 10(2), 2007, 171-78.
Gerber, A. S., G. A. Huber. D. Doherty. and C. M. Dowling.
"Personality Traits and the Consumption of Political
Information." American Politics Research, 39(1), 2011. 32-84.
Ghirardato, P., F. Maccheroni, and M. Marinacci.
"Differentiating Ambiguity and Ambiguity Attitude." Journal of
Economic Theory, 1 18(2). 2004. 133-73.
Gigerenzer, G. "Fast and Frugal Heuristics: The Tools of
Bounded Rationality," in Blackwell Handbook of Judgment and
Decision Making, edited by D. J. Koehler and N. Harvey. Oxford:
Blackwell, 2004. 62-88.
Gilboa, I., and D. Schmeidler. "Maxmin Expected Utility with a
Non-Unique Prior." Journal of Mathematical Economics, 18(2), 1989.
141-53.
Holt, C., and S. Laury. "Risk Aversion and Incentive
Effects." American Economic Review, 92(5). 2002, 1644-55.
Jani, D., J.-H. Jang, and Y.-H. Hwang. "Big Five Factors of
Personality and Tourists' Internet Search Behavior." Asia
Pacific Journal of Tourism Research, 19(5), 2014, 600-15.
Lauriola, M., and I. P. Levin. "Personality Traits and Risky
Decision-Making in a Controlled Experimental Task: An Exploratory
Study." Personalin' and Individual Differences, 31, 2001,
215-26.
Lo, A. R., D. V. Repin, and B. N. Steenbarger. "Fear and Greed
in Financial Markets: A Clinical Study of DayTraders." American
Economic Review: Papers and Proceedings. 95(2), 2005, 352-59.
Masatlioglu, Y., Y. Orhun, and C. Raymond. "Intrinsic
Preferences for Information and Skewness." Mimeo, University of
Maryland. 2016.
McDaniel, S. R. "Investigating the Roles of Gambling Interest
and Impulsive Sensation Seeking on Consumer Enjoyment of Promotional
Games." Social Behavior and Personality, 30(1), 2002, 53-64.
Menezes, C., C. Geiss, and J. Tressler. "Increasing Downside
Risk." American Economic Review. 70(5), 1980, 921-32.
Modnak, J. J., and K. D. Halperin. "A Framework for the Study
of Personality and Political Behaviour." British Journal of
Political Science, 38(2), 2008. 335-62.
Morton, R., J.-R. Tyran, and E. Wengstrom. "Income and
Ideology: How Personality Traits, Cognitive Abilities, and Education
Shape Political Attitudes." Mimeo, University of Copenhagen, 2011.
--. "Personality Traits and the Gender Gap in Ideology,"
in The Political Economy of Social Choices, edited by M. Gallego and N.
Schofield. Cham, Switzerland: Springer International Publishing, 2016.
153-85.
Mueller. G., and E. Plug. "Estimating the Effect of
Personality on Male and Female Earnings." Industrial and Labor
Relations Review. 60(1), 2006, 3-22.
Muller, J., and C. Schwieren. "What Can the Big Five
Personality Factors Contribute to Explain Small-Scale Economic
Behavior?" TI Discussion Paper No. 12-028/1. 2012.
Nicholson, N., E. Soane, M. Fenton-O'Creevy, and P. Willman.
"Personality and Domain-Specific Risk Taking." Journal of Risk
Research, 8(2), 2005, 157-76.
Niederle, M. "Gender," in Handbook of Experimental
Economics, Vol. 2, edited by J. Kagel and A. E. Roth. Princeton, NJ:
Princeton University Press, 2016, 481-553.
Niederle. M., and L. Vesterlund. "Gender and
Competition." Annual Review in Economics, 3, 2011, 601-30.
Nyhus, E. K., and E. Pons. "The Effects of Personality on
Earnings." Journal of Economic Psychology, 26(3), 2005, 363-84.
Paunonen, S. V., and M. C. Ashton. "Big Five Factors and
Facets and the Prediction of Behavior." Journal of Personality and
Social Psychology, 81(3), 2001, 524-39.
Proto, E., and A. Rustichini. "Life Satisfaction, Income, and
Personality." Journal of Economic Psychology. 48, 2015. 17-32.
Rubinstein. A. "Similarity and Decision-Making under Risk (Is
There a Utility Theory Resolution to the Allais Paradox?)." Journal
of Economic Theory. 46(1). 1988. 145-53.
Rustichini. A. "Neuroeconomics: What Have We Found, and What
Should We Search For." Current Opinion in Neurobiology, 19. 2009,
672-77.
Schaninger, C. M., and D. Sciglimpaglia. "The Influence of
Cognitive Personality Traits and Demographics on Consumer Information
Acquisition." Journal of Consumer Research, 8(2). 1981. 208-16.
Schotter, A., and B. Sopher. "Social Learning and Coordination
Conventions in Intergenerational Games: An Experimental Study."
Journal of Political Economy, 111(3), 2003.498-529.
--. "Advice and Behavior in Intergenerational Ultimatum Games:
An Experimental Approach." Games and Economic Behavior. 58(2).
2007. 365-93.
Scott. R. C., and P. A. Horvath. "On the Direction of
Preference for Moments of Higher Order than the Variance." Journal
of Finance, 35(4). 1980. 915-19.
Sharpe, J. P., N. R. Martin, and K. A. Roth. "Optimism and the
Big Five Factors of Personality: Beyond Neuroticism and
Extraversion." Personality and Individual Differences, 51(8),
2011.946-51.
Zuckerman. M. Behavioral Expressions and Biosocial Bases of
Sensation Seeking. New York: Cambridge University Press, 1994.
--. Sensation Seeking and Risky Behavior. Washington, DC: American
Psychological Association, 2007.
(1.) Our experiments are not meant to show that personality only
matters in situations of uncertainty. Rather, we deem it plausible to
expect personality to play a greater role in this type of informational
environments, which appears to be a natural and relevant environment in
which to study personality.
(2.) It is important to point out that in our experiments we
contrast two extreme informational situations; one where a DM faces two
completely known probability distributions, and hence has no opportunity
or need to gather information (what we call risk), and one where the DM
has only very minimal information about the distributions he must choose
between and therefore has an incentive to gather information (what we
call uncertainty). We study these two extremes because they present the
starkest contrast between situations where information gathering is
possible and desirable and situations where it is not. This does not
mean, however, that there is not a middle ground where only risk is
present but information gathering is still possible. However, we have
purposefully avoided these situations in order to examine the more
obvious cases where the ex-ante information that subjects hold is very
sparse, that is, where they do not know the full probability
distributions, and where information gathering is clearly important.
(3.) See Schaninger and Sciglimpaglia (1981) for an early
psychology study on the influence of personality traits on consumer
information acquisition, or Jani, Jang, and Hwang (2014) for a recent
study linking the Big Five Scale with tourists' internet search
behavior.
(4.) The dummy for gender has an estimated marginal effect of 0.41
at the average regressor of 0.47.
(5.) Charness, Karni, and Levin (2013) study the effect that
incentivized persuasion (similar to our advice treatment) has on
ambiguity attitudes. They find that ambiguity-seeking and
ambiguity-incoherent subjects are very likely to follow the
recommendations of ambiguity-neutral subjects.
(6.) We implemented the questionnaires using form V of the
Sensation Seeking Scale (SSS-V) as described by Zuckerman (1994), and
the short (120 items) version of the IPIPNEO Big Five questionnaire
available at http://www.personal .psu.edu/j5j/IPIP/ipipneo 120.htm.
(7.) Instructions for all parts and treatments can be found online
at https://files.nyu.edu/gf35/public/print/Frechette_2011c_inst.pdf.
(8.) This dictation of information preferences was the simplest
method we could think of to obtain the preferences of the subjects over
parts of the distributions that they want to learn about.
(9.) Notice that subjects in this treatment still face uncertainty
even if they are given the three available pieces of information, as
they cannot assess the exact probability of each individual outcome.
(10.) This advice is all prescripted.
(11.) The inclusion of recommendations is a natural way for an
adviser to convey information. It would be posible, however, to provide
information without a recommendation. but the design of incentives for
the advisers would be less clear.
(12.) 0, 9, and 10 choices of the safe options do not correspond to
a finite range of RRA coefficient and consequently subjects with such
decisions are dropped when considering the implied RRA.
(13.) This observation may further illustrate the fact that these
personality scales are not properly designed for economic decision
making. Even if the Big Five were constructed in such a way that all
personality characteristics can be associated to one of these traits, it
is not clear which combination of traits (if any) could characterize a
person's level of risk aversion.
(14.) It is well known that the Bonferroni correction is too
conservative.
(15.) The correction is [alpha] divided by the number of hypotheses
tested.
(16.) Notice that even when subjects observe all three pieces of
information they still face some uncertainty since they do not know the
probabilities associated to each specific outcome.
(17.) Modnak and Halperin (2008) also correlates the Big Five to
media consumption, but it is more general consumption rather than on a
specific topic.
(18.) Even though in gambling situations probabilities can be
objectively known, it appears reasonable to argue that individuals are
not fully cognizant of them. Thus, it is similar to an environment with
uncertainty.
(19.) One study, by Paunonen and Ashton (2001), correlates the Big
Five to a survey question about buying lottery tickets and another about
the willingness to gamble. Unfortunately, they do not provide
information in the paper about which of the five components has a
statistically significant correlation to the answers. Another study by
Breslin et al. (1999) focuses on the interaction of drinking and
Sensation Seeking and the impact this has on risky choice behavior in
the gains versus losses domain.
(20.) Lo, Repin. and Steenbarger (2005) studies an even more
specialized group, namely day-traders that were taking part in an online
training for day-traders. They did not find that any of the Big Five
dimensions correlated significantly with trading performance.
(21.) Nevertheless, as pointed out by Niederle and Vesterlund
(2011). there are some studies that do not find gender differences in
risk preferences.
(22.) See also Brunnermeier and Parker (2005) for a model where
probabilities or beliefs enter directly into a DM's utility
function.
(23.) See Scott and Horvath (1980) for an early contribution and
Chiu (2005) for a more thorough choice theoretic treatment of the issue.
Menezes, Geiss, and Tressler (1980) discuss skewness in a
choice-theoretic framework by introducing the concept of increasing
downside risk, a concept that may have relevance for our discussion
here.
Caption: FIGURE 1 Distributions
TABLE 1
Summary Statistics
Variable Mean SD Min Max Obs
Female 0.47 123
Holt-Laury choices 5.37 1.58 0 9 123
RRA (a) 0.38 0.43 -0.72 1.17 121
Neuroticism 49.55 8.58 25.78 72.47 123
Extraversion 50.05 8.64 29.66 76.79 123
Openness 51.90 10.17 21.81 75.88 123
Agreeableness 50.44 9.44 28.13 69.68 123
Conscientiousness 52.17 9.37 31.56 72.32 123
SSS 21.73 6.67 8 35 123
Thrill 6.94 2.59 0 10 123
Experience 6.21 2.03 2 10 123
Disinhibition 4.92 2.71 0 10 123
Boredom 3.66 2.16 0 10 123
Note: RRA, Relative Risk Aversion; SD, standard
deviation; SSS, Sensation Seeking Scale,
aggregate score
(a) RRA implied by Holt-Laury choices.
TABLE 2
Big Five Personality Traits and Their Facets
Trait Facet Description
Neuroticism
Identifies Anxiety Level of free floating
individual anxiety
tendency to
experience Angry Tendency to experience anger,
psychological hostility frustration, bitterness, etc.
distress
Depression Tendency to experience guilt,
sadness, despondency and
loneliness
Self Shyness or social anxiety
consciousness
Impulsiveness Tendency to act on cravings
and urges rather than
delaying gratification
Vulnerability General susceptibility
to stress
Extraversion
Quantity and Warmth Interest in and friendliness
intensity of toward others
energy directed
outward into Gregariousness Preference for the company
the social of others
world
Assertiveness Social ascendancy and
forcefulness of expression
Activity Pace of living
Excitement Need for environmental
seeking stimulation
Positive Tendency to experience
emotion positive emotions
Openness to
experience
The active Fantasy Receptivity to the inner
seeking and world of imagination
appreciation
of experiences Esthetics Openness to inner feelings
for their and emotions
own sake
Feelings Social ascendancy and
forcefulness of expression
Actions Openness to new experiences
on a practical level
Ideas Intellectual curiosity
Values Readiness to re-examine own
values and those of authority
figures
Agreeableness
The kinds of Trust Belief in the sincerity and
interactions an good intentions of others
individual
prefers, from Straight- Frankness in expression
compassion to forwardness
tough
mindedness Altruism Active concern for the
welfare of others
Compliance Response to interpersonal
conflict
Modesty Tendency to play down own
achievements and be humble
Tender Attitude of sympathy for
mindedness others
Conscientiousness
Degree of Competence Belief in own self efficacy
organization,
persistence, Order Personal organization
control, and
motivation in Dutifulness Emphasis placed on importance
goal directed of fulfilling moral
behavior obligations
Achievement Need for personal achievement
striving and sense of direction
Self discipline Capacity to begin and
complete tasks despite
boredom or distractions
Deliberation Tendency to think things
through before acting or
speaking
Source: Costa and McCrae 1992.
TABLE 3
Correlations Between Personality Traits, RRA, and Gender
Risk
Female Aversion Neuroticism
Female 1
Risk aversion 0.21 ** 1
Neuroticism -0.14 0.09 1
Extraversion -0.17 * -0.09 -0.48 ***
Openness -0.13 -0.09 -0.20 **
Agreeableness 0.03 -0.06 -0.14
Conscientiousness -0.06 -0.07 -0.46 ***
SSS -0.25 *** -0.17 * -0.16 *
SSS: thrill -0.21 ** -0.16 * -0.20 **
SSS: experience -0.05 -0.07 -0.06
SSS: disinhibition -0.16 * -0.02 -0.09
SSS: boredom -0.28 *** -0 26 *** -0.07
Extraversion Openness Agreeableness
Female
Risk aversion
Neuroticism
Extraversion 1
Openness 0.32 *** 1
Agreeableness 0.05 0.30 *** 1
Conscientiousness 0.25 *** 0.18 ** 0.22 **
SSS 0.44 *** 0.46 *** -0.15 *
SSS: thrill 0.33 *** 0.20 ** -0.04
SSS: experience 0.23 *** 0.57 *** 0.07
SSS: disinhibition 0.41 *** 0.30 *** -0.16 *
SSS: boredom 0.23 ** 0.26 *** -0 29 ***
Conscientiousness SSS SSS:
Thrill
Female
Risk aversion
Neuroticism
Extraversion
Openness
Agreeableness
Conscientiousness 1
SSS -0.12 1
SSS: thrill -0.08 0.64 *** 1
SSS: experience -0.13 075 *** 0 29 ***
SSS: disinhibition -0.06 079 *** 0 28 ***
SSS: boredom -0.10 0.61 *** 0.16 *
SSS: SSS: SSS:
Experience Disinhibition Boredom
Female
Risk aversion
Neuroticism
Extraversion
Openness
Agreeableness
Conscientiousness
SSS
SSS: thrill
SSS: experience 1
SSS: disinhibition 0.58 *** 1
SSS: boredom 0.31 *** 0.31 *** 1
* Significant at 10%; ** significant at 5%;
*** significant at 1%.
TABLE 4
Frequency Choice for the Riskier Distribution,
by Treatment
Treatment Control Priority Advice
Pair Frequency Frequency Frequency
(%) (%) (%)
L vs. SR 60.98 16.67 *** 40.00
L vs. G/L 26.83 38.10 30.00
L vs. U 12.20 35.71 ** 15.00
SR vs. G/L 60.98 45.24 45.00
SR vs. U 26.83 50.00 ** 35.00
G/L vs. U 39.02 50.00 40.00
* Significantly different from the frequency in the
Control treatment at 10%; ** significant at 5%;
*** significant at 1%.
TABLE 5
Probit Estimates of the Factors Correlated to
Riskier Choices
Treatments
Variable Control Priority (a) Advice (b)
RRA -0.346 ** 0.054 -0.603 **
(0.174) (0.537) (0.263)
Female -0.184 -0.643 .575 **
(0.169) (0.459) (0.270)
Neuroticism 0.001 0.002 0.004
(0.009) (0.034) (0.011)
Extraversion -0.014 -0.096 ** 0.020
(0.016) (0.041) (0.017)
Openness -0.005 0.073 ** -0.013
(0.010) (0.030) (0.017)
Agreeableness 0.002 0.008 -0.022 *
(0.009) (0.022) (0.013)
Conscientiousness -0.004 0.053 ** -0.003
(0.008) (0.024) (0.012)
SSS 0.008 0.006 0.020
(0.022) (0.040) (0.022)
Constant 0.711 -2.818 -0.993
(0.729) (3.841) (1.724)
p Value: test of H1 0.854 0.055 0.012
Note: Clustered (by subject) standard errors in
parentheses.
(a) This is for the cases where subjects receive one piece
of information. Not reported: The subjects' information
preference (over the order for B, M. and T) is also
controlled for as a set of dummy variables.
(b) Not reported: The advice received is also controlled for
as a dummy variable (risky or safe distribution). Not
reported: The reason used for the advice (B, M, or T).
* Significant at 10%; ** significant at 5%; *** significant
at 1%.
TABLE 6
Marginal Effects of the Factors Correlated to
Riskier Choices
Treatments
Variable Control Priority (a) Advice (b)
RRA -0.131 ** 0.021 -0.200 **
(0.066) (0.209) (0.090)
Female -0.069 -0.249 0.230 **
(0.063) (0.175) (0.103)
Neuroticism 0.000 0.001 * 0.002
(0.004) (0.013) (0.004)
Extraversion -0.005 -0.037 ** 0.010
(0.006) (0.016) (0.006)
Openness -0.002 0.028 ** -0.006
(0.004) (0.012) (0.005)
Agreeableness 0.001 0.003 -0.007 *
(0.003) (0.008) (0.004)
Conscientiousness -0.002 0.021 ** -0.000
(0.003) (0.009) (0.004)
SSS 0.003 0.002 0.006
(0.008) (0.015) (0.007)
Note: Clustered (by subject) standard errors in
parentheses.
(a) This is for the cases where subjects receive one piece of
information. Not reported: The subjects' information
preference (over the order for B. M. and T) is also controlled
for as a set of dummy variables.
(b) Not reported: The advice received is also controlled for
as a dummy variable (risky or safe distribution). Not reported:
The reason used for the advice (B, M, or T).
* Significant at 10%; ** significant at 5%; *** significant
at 1%.
TABLE 7
Preferences over Different Information
Info Rank Ranked
First
Bottom 1 1 2 2 3 3 50.00
Middle 3 2 1 3 1 2 26.19
Top 2 3 3 1 2 1 23.81
Frequency 33.33 16.67 7.14 7.14 19.05 16.67
(%)
TABLE 8
Multinomial Probit of the Factors Correlated to
Demand for the First Priority in the Priority
Treatment
Estimates Marginals
Variable M T M T
RRA -2.24 * -1.389 -0.363 -0.173
(1.232) (1.052) (0.234) (0.247)
Female 2.438 ** 0.847 0.411 *** 0.015
(1.053) (0.811) (0.156) (0.177)
Neuroticism 0.18 ** 0.08 0.032 ** 0.005
(0.076) (0.057) (0.014) (0.014)
Extraversion 0.168 * 0.091 0.028 * 0.009
(0.092) (0.064) (0.017) (0.016)
Openness 0.05 0.049 0.007 0.009
(0.049) (0.047) (0.010) (0.011)
Agreeableness -0.059 -0.049 -0.008 -0.008
(0.061) (0.049) (0.012) (0.012)
Conscientiousness 0.11 ** 0.033 0.021 ** -0.001
(0.054) (0.047) (0.011) (N/A)
SSS -0.116 -0.125 * -0.014 -0.023
(0.089) (0.074) (0.017) (0.018)
Constant -20.763 ** -7.667
(8.94) (6.966)
Notes: Bottom is the base outcome; standard errors in
parentheses; one standard error is reported as N/A because
the software cannot compute it.
* Significant at 10%; ** significant at 5%; *** significant
at 1%.
TABLE 9
Probit Estimate of the Relation between
Information Observed and Riskier Choices in
the Priority Treatment
Variable 10
Observed bottom 0.383
(0.348)
Observed top 1.148 ***
(0.437)
Observed middle -0.757 *
(0.394)
Observed B and T 0923 ***
(0.227)
Observed B and M -0.486
(0.373)
Observed M and T -0.070
(0.389)
Preference: B-T-M -0.735 **
(0.349)
Preference: B-M-T -0.300
(0.418)
Preference: T-B-M -0.545
(0.344)
Preference: T-M-B 0.024
(0.312)
Preference: M-B-T -0.219
(0.247)
Constant -0.103
(0.275)
Notes: The default is to observe three pieces of
information; clustered (by subject) standard errors
in parentheses.
* Significant at 10%; ** significant at 5%; *** significant
at 1%.
TABLE 10
Frequency of Choice of the Riskier Distribution
in the Control and Priority Treatments, for Some
of the Key Cases of Information Observed
Treatment Information Frequency (%)
Control 37.80
Priority Bottom 39.47
Middle 18.18
Top 80.95
B-M-T 30.77
TABLE 11
Probit of the Factors Correlated to Following
Advice, for Deciders in the Advice Treatment
Variable Estimates Marginals
Advice for the risky -1.101 ** -0.185 *
distribution (0.506) (0.100)
Information about B 0.65 0.062 *
(0.525) (0.037)
Information about T 1.089 * 0.106 **
(0.591) (0.046)
RRA -0.221 -0.029
(0.528) (0.067)
Female 1.501 *** 0.144 ***
(0.474) (0.029)
Neuroticism 0.007 0.001
(0.015) (0.002)
Extraversion 0.063 *** 0.008 ***
(0.02) 0.002)
Openness -0.065 ** -0.009 **
(0.02) (0.003)
Agreeableness 0.074 *** 0.01 ***
(0.021) 0.003)
Conscientiousness -0.055 *** -0.007 ***
(0.021) (0.003)
SSS 0.038 0.005
(0.05) (0.006)
Constant -1.091
(2.162)
Note: Clustered (by subject) standard errors in
parentheses.
* Significant at 10%; ** significant at 5%; *** significant
at 1%.
TABLE 12
Probit of the Factors Correlated to Giving
Advice toward the Riskier Option, for Advisers
in the Advice Treatment
Variable Estimates Marginals
RRA 0.114 0.043
(0.151) (0.057)
Female 0.34 ** 0.125 **
(0.152) (0.054)
Neuroticism 0.011 0.004
(0.010) (0.004)
Extraversion 0.003 0.001
(0.007) (0.003)
Openness -0.018 *** -0.007 ***
(0.006) (0.002)
Agreeableness -0.010 -0.004
(0.007) (0.003)
Conscientiousness 0.019 ** 0.007 **
(0.010) (0.004)
SSS 0.016 ** 0.006 **
(0.007) (0.003)
Constant -1.212
(0.870)
Note: Clustered (by subject) standard errors in
parentheses.
* Significant at 10%; ** significant at 5%; *** significant
at 1%.
TABLE 13
Multinomial Probit Estimate of the Factors
Correlated to the Reason Given as Advice, for
Advisers
Estimates Marginals
Variable M T M T
RRA -0.567 -0.645 -0.035 -0.050
(0.56) (0.497) (0.103) (0.078)
Female 1.223 ** 1.393 *** 0.095 0.113 *
(0.476) (0.35) (0.113) (0.062)
Neuroticism 0.032 0.023 0.005 -0.001
(0.039) (0.022) (0.008) (0.005)
Extraversion 0.036 0.039 0.003 0.003
(0.039) (0.025) (0.007) (0.004)
Openness -0.038 -0.068 *** 0.003 -0.010 ***
(0.032) (0.023) (0.005) (0.003)
Agreeableness -0.002 -0.016 0.003 -0.004
(0.028) (0.015) (0.006) (0.004)
Conscientiousness -0.018 0.023 -0.010 0.010*
(0.032) (0.025) (0.007) (0.005)
SSS -0.074 ** -0.043 -0.013* 0.004
(0.037) (0.027) (0.007) (0.004)
Constant 1.838 0.827
(3.472) (2.479)
Notes: Bottom was used as the base outcome; standard
errors in parentheses; one standard error is reported as N/A
because the software cannot compute it.
* Significant at 10%; ** significant at 5%; *** significant
at 1%.
TABLE 14
Key (Statistically Significant) Relations between
Personality Traits and Behavior
Treatment N E O A C SSS
Direct
Control Riskier choice
Priority Info, demand (a) + M + M + M -T
Advice Follow advice + - + -
Advice Give risky advice - + +
Indirect
Priority Riskier choice + -
(b)
Advice Riskier choice
Notes: A, Agreeableness; C, Conscientiousness; E.
Extraversion; N. Neuroticism; O, Openness to Experience.
(a) Compared to the baseline of ranking B first in
the priority.
(b) For the case when subjects observe three pieces of
information.
COPYRIGHT 2017 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2017 Gale, Cengage Learning. All rights reserved.
