10978_The Domain-Specific Risk-Taking Scale lacks convergence with alternative risk-taking propensity measures

Graduate Theses and Dissertations
Iowa State University Capstones, Theses and
Dissertations
2018
The Domain-Specific Risk-Taking Scale lacks
convergence with alternative risk-taking propensity
measures
Michael Tynan
Iowa State University
Follow this and additional works at: https://lib.dr.iastate.edu/etd
Part of the Social Psychology Commons
This Thesis is brought to you for free and open access by the Iowa State University Capstones, Theses and Dissertations at Iowa State University Digital
Repository. It has been accepted for inclusion in Graduate Theses and Dissertations by an authorized administrator of Iowa State University Digital
Repository. For more information, please contact digirep@iastate.edu.
Recommended Citation
Tynan, Michael, “The Domain-Specific Risk-Taking Scale lacks convergence with alternative risk-taking propensity measures” (2018).
Graduate Theses and Dissertations. 16477.
https://lib.dr.iastate.edu/etd/16477

The Domain-Specific Risk-Taking Scale lacks convergence with alternative risk-taking
propensity measures

by

Michael C. Tynan

A thesis submitted to the graduate faculty
in partial fulfillment of the requirements for the degree of
MASTER OF SCIENCE

Major: Psychology

Program of Study Committee:
Marcus Credé, Major Professor
Patrick Armstrong
L. Alison Phillips

The student author, whose presentation of the scholarship herein was approved by the
program of study committee, is solely responsible for the content of this thesis. The
Graduate College will ensure this thesis is globally accessible and will not permit
alterations after a degree is conferred.

Iowa State University
Ames, Iowa
2018

ii
TABLE OF CONTENTS
CHAPTER 1. INTRODUCTION

1

Theoretical Conceptualizations of Risk-Taking Propensity

2

Approaches to the Measurement of Risk-Taking Propensity

6

CHAPTER 2. METHOD

11

Subjects

11

Measures

11

Procedure

15

CHAPTER 3. RESULTS

17

DOSPERT Structure

17

Correlations Between DOSPERT and Non-Domain Specific Measures

20

Correlations Between DOSPERT and Self-Reported Risk-Taking Behavior
22

Correlations Between DOSPERT and Decision-making Tasks

22

DOSPERT Incremental Validity

23

CHAPTER 4. DISCUSSION

25

Limitations

28

Conclusion

29

REFERENCES

30

iii
TABLES

35
APPENDIX. IRB APPROVAL

40

iv
ABSTRACT
The domain-specific evaluative approach to risk-taking propensity allows people to differentiate
situations in which they will approach risk-related decisions from situations in which they will
avoid them. The Domain-Specific Risk-Taking Scale (DOSPERT) is the most widely used
measure of such evaluations. The current study of the DOSPERT tests alternatives to the
assumed five-domain structure, explores associations between the DOSPERT and alternative
risk-taking measures, and tests the incremental validity of the DOSPERT in predicting both self-
reported risky behavior and risky behavior in the lab. Analyses show that the DOSPERT would
benefit from a six-factor structure rather than five factors, the DOSPERT domains are weakly
correlated with the majority of alternative risk-taking propensity measures, and the DOSPERT
can predict variance in certain self-reported risky behaviors, but not risky behaviors in the lab,
after accounting for alternative measures.
Key words: Risk-taking, validity, measurement, personality
1

CHAPTER 1
INTRODUCTION

Decisions to take or avoid risks influence behavior in a wide variety of domains, ranging
from the relatively mundane (e.g., buying a lottery ticket, unhealthy food choices) to impactful
(e.g., changing careers, relationship decisions, buying a home). Because individuals are
frequently required to make such decisions, researchers have examined variables that might
reflect the general tendency to favor an action with an extremely profitable, but unlikely, or an
extremely aversive outcome over an alternative action with a less extreme but more likely
outcome (Derntl, Pintzinger, Kryspin-Exner, & Schöpf, 2014) (hereafter referred to as “risk-
taking propensity”). In economics, risk-taking propensity can partially explain behaviors like
gambling preferences, stock management, and debt accrual (Alm, McClelland, & Schultze, 1992;
Dew & Xiao, 2011; Dislich, Zinkernagel, Ortner, & Schmitt, 2010; Dohmen et al., 2011;
Kahneman & Tversky, 1979). Similarly, psychologists explore risk-taking propensity’s relations
with behaviors like reckless driving, unhealthy habits, and criminality (Arnett, 1991; Gullone,
Moore, Moss, & Boyd, 2000; Szrek, Chao, Ramlagan, Peltzer, 2012). More general social
nonconformity, unethical behaviors, aggression, and self-harm are also associated with risk-
taking propensity (Sadeh & Baskin-Sommers, 2016; Weber, Blais, & Betz, 2002).
Despite the widespread interest in and practical importance of risk-taking behaviors, there
is substantial disagreement about theoretical conceptualizations and the appropriate measurement
of risk-taking propensity. This study will examine the construct and criterion-related validity of
scores produced by one widely used measure of risk-taking propensity: the Domain Specific
Risk-Taking Scale (DOSPERT). The structure of the DOSPERT, its relation to other risk-taking
2

measures, and its utility in predicting risk-taking behaviors will be analyzed. This study will
highlight DOSPERT scores’ unique relations with concrete outcomes, after accounting for
similar measures, and the inclusion of alternative risk-taking propensity measures will elucidate
possible shortcomings of the DOSPERT structure or content. In the following section, I will
briefly introduce theoretical and measurement approaches germane to examinations of the
DOSPERT against alternative measures.
Theoretical Conceptualizations of Risk-Taking Propensity
Risk-taking propensity has been conceptualized in three distinct theoretical ways: a) as a
stable trait, b) as a domain-specific evaluation, c) as a descriptive versus normative decision-
making process.
The Trait Approach. Early developments of the risk-taking propensity construct focused
on trait-like tendencies: sensation seeking, impulsiveness, and venturesomeness (Eysenck &
Eysenck, 1978; Zuckerman, Eysenck, & Eysenck, 1978). These constructs describe risk-taking
propensity as a set of behaviors that reflect habitual dispositions to approach or avoid risk-related
decisions (Knowles, Cutter, Walsh, & Casey, 1974; Yechiam & Ert, 2011). If risk-taking
propensity is a trait, levels of risk-taking propensity will be reflected by a desire for greater
frequency and intensity of risky situations (DeYoung, 2010). Sensation seeking captures
information about people who take risks frequently rather than those who engage in the same
activities infrequently (Desrichard & Denarié, 2005). Individual differences in risk-taking
propensity also reflect differences in the need for novel, complex, intense, or varied situations
(Arnett, 1994; Lauriola & Levin, 2001). Prior research on the temporal stability and rank-order
consistency of risk-taking propensity and between-person neurobiological differences are also
broadly supportive of the trait conceptualization.
3

For example, risk-taking “acceptance,” considered broadly as attitudes toward risk
compared to certainty, has been shown to be stable over time (Yechiam & Ert, 2011). Risk-
taking propensity has also displayed rank-order stability similar to major personality facets: risk-
seeking peeks in late adolescence and declines with age (Josef et al., 2016). The motivations for
risk-taking propensity should also be expected to be consistent over time. The need for thrills or
the aversion to boredom that characterize sensation-seeking are rehearsed evaluations of
particular situations, and these evaluations become habitual (Zuckerman & Kuhlman, 2000).
Low need for cognition or low systematic processing of uncertain and unfavorable outcomes
should also be considered habitual until a strong negative outcome forces re-evaluation of a
similar situation in the future.
Risk-taking propensity is also associated with neurobiological differences (Zuckerman &
Kuhlman, 2000). The interaction of genes and shared environment explain, more than genes
alone, tendencies to engage in risk-relevant behaviors like seatbelt use, birth control use, or
riding a motorcycle (Miles et al., 2001). The stability of risk-taking propensity as a habitual set
of behaviors supports conceptualizations of sensation seeking, impulsiveness, and
venturesomeness as traits.
Domain-specific evaluation. Trait conceptualizations of risk-taking propensity have been
questioned due to evidence for cross-situational instability of risk-related behavior. Individuals
who are highly risk-averse in some domains seek out risk in other domains (e.g., “insurance-
buying gamblers” and “skydiving wallflowers;” Johnson, Wilke, & Weber, 2004). This may, in
part, reflect differences in how individuals evaluate the riskiness of activities (Althaus, 2005;
Aven & Renn, 2009; Johnson et al., 2004; Weber et al., 2002). Risk-taking propensity may
therefore be better understood as a series of positively or negatively valenced associations, which
4

can be less stable over time and across domains. For example, an individual may be very
comfortable with financial risk but have a deep fear of being disliked by others. Later in life, that
same individual may have financial obligations (e.g., children) that make him or her far less
willing to engage in financially risk activities but also make him or her less concerned about the
perceptions of others.
These differences in willingness to engage in a risk-related activity can be understood
through Vroom’s expectancy theory of motivation (Vroom, 1964). The motivation to engage in
or avoid a certain action is a multiplicative function of valence, instrumentality, and expectancy
(Landy & Becker, 1989). In considering risk-related decisions, valence and expectancy are the
key drivers of individual differences. Difference in expectancy of a risk-related behavior is a
difference in estimation of the likelihood that a negative outcome will occur. Numerically,
expectancy ranges from 0, the estimation that a negative outcome is impossible, to 1, the
estimation that a negative outcome is certain (Lunenburg, 2011). Valence refers to a person’s
preference for a certain outcome of a risk-related behavior. Numerically, valence ranges from -1
(extremely negative) to 1 (extremely positive); 0 indicating indifference to the outcome
(Lunenburg, 2011). For the “skydiving wallflower,” social embarrassment is too intensely
negative, but skydiving is not perceived as a high-risk activity, perhaps due to high perceived-
controllability (Johnson et al., 2004). This person may also have a low expectancy of the
negative outcomes of skydiving, and high expectancy of negative social outcomes. This
combination of valance and expectancy makes this person likely to engage in a recreational risk-
taking behavior, but avoid a social risk-taking behavior.

Concepts of risk-taking propensity as multidimensional, reflecting differences across
domains, emphasize individual differences in the perception of riskiness and the perceived
5

importance of positive and negative outcomes in different domains. Measures developed out of
this theoretical framework emphasize the influence of domains and situations on risk-taking
behavior.
Descriptive versus normative construct. The first two conceptualizations consider risk-
taking propensity a reflection of the tendency to engage in behaviors in which the probabilities of
outcomes and even the outcomes themselves are unknown. However, in behavioral economics,
risk-taking propensity generally describes “decisions made under risk,” meaning the respondent
knows the possible positive and negative outcomes, and their exact probabilities (Tversky &
Kahneman, 1981; Weber, 2010; Weber, Blais, & Betz, 2002). Unlike trait and domain-specific
propensity approaches, this field is primarily interested in financial behavior. Empirical tests
with real, non-trivial, monetary consequences are considered optimal measures of risk-taking
propensity by behavioral economists.
This field also distinguishes between normative models and descriptive models of
decision-making. Normative models assume rational, consistent principles guide behavior,
whereas descriptive models like prospect theory empirically explore how individuals deviate
from normative expectations (Zaleskiewicz, 2001). Normative models require multiplying the
outcome of an action with its probability and subtracting this value from the cost, and the
“correct answer” in a normative value equation is indicated by a positive solution. For example,
a $1 lottery ticket with only one $1,000,000 winner should only be purchased if fewer than
999,999 other people are participating ($1 – 1/999,999 * $1,000,000 = .000001). Measures like
the Iowa Gambling Task and the Balloon Analogue Risk Task are descriptive instruments with
normative responses in mind. The Iowa Gambling Task has been used as a clinical instrument to
capture decision-making impairment (Upton, Bishara, Ahn, & Stout, 2011), and it was originally
6

validated with a sample of patients with damage to the prefrontal cortex (Bechara, Damasio,
Damasio, & Anderson, 1994). Tasks derivative of the Iowa Gambling Task do not require
clinical samples to be considered normative measures. Descriptive measures involve a different
interpretation of similar tasks. Descriptive models empirically explore how individuals deviate
from normative expectations by taking decision-making biases, error, and the relative utility of
money into account (Weber, 2010; Zaleskiewicz, 2001). All tasks involving money or other
easily quantified outcomes based on clear probabilities can be interpreted through both
descriptive and normative models. The “rational” response can be calculated, and quantified
deviations from this ideal set of responses can be interpreted in a descriptive model.
When risk-taking propensity is used as a label for differences in deviation from a norm,
this approach can be treated as similar to trait conceptualizations. A “risk premium” or “risk
perception” term within an expected utility calculation is an assumed, “typically…stable
construct, i.e. a personality trait” (Weber & Johnson, 2009, p. 139). However, nothing about the
normative vs. descriptive approach or an expected utility calculation necessitates a term
accounting for differences only due to traits. Trait or relatively stable domain-specific
differences may be parts of the risk-taking propensity difference term in these models, but not all
apparent risk-taking behavior may be due to a “risk attitude” term (Weber & Johnson, 2009). In
this way, a normative vs. descriptive approach may be broader in its identification of individual
differences in risk-taking propensity, but measures derived from this approach are expected to
converge partially with both trait and domain-specific measures.
Approaches to the Measurement of Risk-Taking Propensity

These three broad theoretical ways of conceptualizing risk-taking propensity are also
reflected in three broad ways in which risk-taking propensity is operationalized and measured.
7

Risk-taking propensity has been measured primarily in the following ways: a) trait self-report
inventories b) domain-specific evaluations c) decision-making proxy tasks involving
hypothetical money and d) decision-making proxy tasks involving real money.
The majority of risk-taking propensity measures rely on self-reports. Self-report measures
result in scores that have been shown to be predictive of concrete risk-related behaviors like
counterproductive workplace behaviors, unhealthy habits, unsafe sexual practices, and reckless
driving (Renner & Anderle, 2000; Van Iddekinge, Roth, Raymark, & Odle-Dusseau, 2012;
Zuckerman, 2007).
Self-report measures either assess a single global risk-taking propensity facet (e.g.,
Arnett, 1994), or assess risk-taking propensity in different domains. Broadly, risk-taking
propensity trait self-report inventories separate the construct into sensation-seeking (or
“venturesomeness”) and impulsiveness facets. Sensation-seeking refers to willingness to risk
negative outcomes in pursuit of a new, complex, or intense experiences, and impulsiveness is
conceptualized as the tendency to rapidly respond to cues for potential rewards without
considering possible negative consequences (Zuckerman & Kuhlman, 2000). These two
tendencies are not mutually exclusive (e.g. new experiences may be perceived as so rewarding, a
person seeks them impulsively), and together they describe different motivations habitually
approaching or avoiding risk-related decisions.
One of the most widely used domain-specific assessment of risk-taking propensity is
Weber and colleagues’ (2002) Domain Specific Risk Taking (DOSPERT) scale that examines
five domains of risk-taking: a) health & safety, b) financial, c) recreational, d) ethical, and e)
social. Other domains have been suggested to expand the DOSPERT: driving, occupational risk-
taking (entrepreneurship or changing careers often), self-harm (including suicidal ideation and
8

attempt), or aggression (Josef et al., 2016; Nicholson, Soane, Fenton-O’Creevy, & Willman,
2005; Sadeh & Baskin-Sommers, 2016), but the DOSPERT five-domain framework is the most
widely used among these measures.
The DOSPERT was developed by reviewing concurrent literature and existing risk-taking
measures capturing a wide variety of domains encountered by young adults in Western cultures
(Weber et al., 2002). The original DOSPERT scale was developed, from an initial set of 101
items, by selecting 10 items per domain with the highest item-total correlations to their own
subscale, and the scale was reduced to the final 40-item original scale using ordinary least-
squares exploratory factor analyses (Weber et al., 2002). Certain items were revised to be more
culturally generalizable (e.g. “Disagreeing with your father…” was revised to “Disagreeing with
an authority figure…”), and the scale was shortened to 30 items by conducting exploratory and
confirmatory factor analyses on random halves of a novel sample (Blais & Weber, 2006).
Though the original 40-item DOSPERT consisted of six factors (ethical, social, health/safety,
recreational, gambling, and investment), the short DOSPERT scale consists of five, by merging
investment and gambling into the “financial” domain (Blais & Weber, 2006).
Despite the popularity of self-report inventories, some have argued that scores on self-
report measures lack validity unless participants are performing objective tasks incentivized with
real money for appropriately high stakes (Dislich et al., 2010; Holt & Laury, 2002; Lejuez et al.,
2002). These researchers have argued that responses to survey items are limited in their
predictive validity and construct validity because survey questions are not “incentive
compatible,” i.e. linked to any tangible gain or loss (Dohmen et al., 2011). Additionally, there is
concern that self-report risk measures cannot predict self-reported socially undesirable behaviors
(Sackett, Burris, & Callahan, 1989).
9

The alternatives to self-report questionnaires are decision-making proxy tasks, meant to
capture the characteristics of real-world risk-related decisions. Tasks like the Iowa Gambling
Task and Balloon Analogue Risk Taking (BART) task vary in form but share key components:
forcing approximation of future outcomes, capturing the expectancy of outcomes, and capturing
aversion to loss (Bechara, 1994; Lejuez et al., 2002). Participants in these proxy tasks act in a
simple game environment in which they can choose between options with uncertain outcomes.
Participants can respond differently to the experience of a negative outcome (losing money) by
choosing conservative or reckless options in future rounds. Decision-making tasks appear to
reflect components of risk-related traits like sensation seeking and impulsiveness, and there is
evidence for decision-making tasks as an appropriate measure of risk-taking preference as
conceptualized by risk-related personality constructs (Lauriola et al., 2014; Mishra & Lalumière,
2011; Upton et al., 2011). Scores on these proxy tasks and self-report questionnaires of risk-
taking propensity have been observed to be only modestly correlated (Lauriola et al., 2014;
MacCrimmon & Wehrung, 1985; Mishra & Lalumière, 2011; Skeel, Neudecker, Pilarski, Pytlak,
2007; Szrek et al., 2012).
While decision-making proxy tasks are more objective than self-report measures of risk
behavior, they are not without limitations. Specifically, the validity of responses to decision-
making proxy tasks with monetary incentives are likely to be influenced by both the
characteristics of the specific task (e.g., the interestingness of the task, the degree of attention
required; see Camerer & Hogarth, 1999), and the characteristics of the participant (e.g., mental
arithmetic skills). Importantly, most decision-making proxy tasks are difficult to interpret when
there is large variability in individual wealth in the sample, i.e. large differences in the relative
utility of the task’s payout.
10

These limitations of decision-making proxy tasks suggest that many researchers who are
interested in risk-taking propensity will continue to rely on self-report measures. It is the goal of
this study to examine the validity of scores on the most widely used domain-specific assessment
of risk-taking propensity, the DOSPERT Scale (Blais & Weber, 2006). Specifically, I aim to
examine: a) the validity of the hypothesized five-factor structure of scores on the DOSPERT b)
the convergent and discriminant validity of scores on the DOSPERT domains vis a vis scores on
measures that operationalize risk-taking differently, c) the criterion validity of DOSPERT scores
with respect to risky behaviors, and d) the incremental validity of DOSPERT scores over and
above scores on other measures of risk-taking propensity for the prediction of risky behaviors.
11

CHAPTER 2
METHOD

Subjects
Data were collected from 399 participants. Participants were recruited through the subject
pool of a large, public Midwestern land-grant institution, and received 2 course credits as
compensation. Three participants were excluded for excessive missing data (> 15%
missingness). An additional 13 participants were excluded for failing 2 of 3 attention checks
(final N = 383). Participants were permitted one failed attention check because stricter exclusion
thresholds often screen data that is not indicative of random or careless responding (Kim et al.,
2017) The sample was primarily female (57.1%) and Caucasian (76%). Participants’ age ranged
from 18 to 26, mean being 19.08.
Measures
Eysenck & Eysenck Impulsiveness/Venturesomeness
The Impulsiveness scale consists of 19 items reflecting quick decision-making without
considering or perceiving risks. Sample items of the Impulsiveness scale include “Do you
usually make up your mind quickly?” and “Do you mostly speak before thinking things out?”
(Eysenck, Pearson, Easting, & Allsopp, 1985). The Venturesomeness scale is a 16-item scale
capturing decisions made while being aware of risks but acting anyway. The Venturesomeness
scale includes items like “Would you like to learn to fly an airplane?” and “Would you enjoy fast
driving?”. Because both scales were developed in the United Kingdom, mild revisions were
made to replace terms and phrases uncommon in American English. For example, “to go pot-
holing” has been revised as “to explore a cave.”
12

Sensation Seeking Scale-Form V
The Sensation Seeking Scale-Form V is a 40-item forced choice scale reflecting
preference for novel and exciting experiences. For example, the sensation-seeking attitude “the
worst social sin is to be a bore” is compared to “the worst social sin is to be rude.” Though
developed initially for clinical populations, the Sensation Seeking Scale was revised to be
generalizable to typical adult samples, and the resulting SSS-V encompasses four factors of
sensation seeking behavior (Zuckerman et al., 1978). Thrill and Adventure Seeking captures
desires to engage in extreme sports or dangerous recreational activities. Experience Seeking
reflects desire for new sensory experiences through travel and other aspects of a “non-
conforming lifestyle.” Disinhibition represents the desire for social and sexual freedom.
Boredom Susceptibility captures distaste for routine and tedium.
The Impulsiveness/Venturesomeness and SSS-V scale were revised from forced-choice
format to a 5-point Likert scale response. This revision facilitates the use of factor analytic
techniques based on maximum likelihood estimation. Dichotomously scored items violate
maximum likelihood estimations’ assumption of multivariate normality and distort the
correlation matrices by which factor analyses are conducted (Reise, Waller, & Comrey, 2000).
This revision also eliminates the possibility of artificially low correlations between forced choice
format scales and Likert response scales in used by most other measures in the study.
Arnett Inventory of Sensation Seeking
The 20-item Arnett Inventory of Sensation Seeking is similar to the SSS-V, though it
uses preference for novelty and intensity of situations as theoretical components of the latent risk
trait (Arnett, 1994). Items like “I would like to travel to places that are strange and far away”
capture novelty. Intensity items include “When I listen to music, I like it to be loud.” Participants
13

rate each item using a 4-point Likert-type scale, ranging from describes me very well to does not
describe me at all.
Domain Specific Risk Taking Scale
The 30-item Domain Specific Risk-Taking Scale (DOSPERT) is intended to assess five
domains of propensity to favor risk (Blais & Weber, 2006). Participants assess their “likelihood
of engaging in each activity or behavior if [they] were to find [themselves] in that situation”
using a 7-point Likert-type scale, ranging from extremely unlikely to extremely likely. Items
assess domains of Health & Safety, Ethical, Financial, Social, and Recreational risk.
Representative items of those domains include, respectively: “Riding a motorcycle without a
helmet,” “Passing off someone else’s work as your own,” “Betting a day’s income on the
outcome of a sporting event,” “Disagreeing with an authority figure on a major issue,” and
“Taking a skydiving class.”
The scale was validated originally by comparing participants’ likelihood of engaging in
40 activities to their ratings of how risky each activity seems, as well as their scores on
Zuckerman’s Sensation Seeking Scale (Weber et al., 2002). The original DOSPERT was revised
in order to be more applicable to diverse adult populations and to shorten the scale to 30 items
(Blais & Weber, 2006).
Iowa Gambling Task
Participants in the Iowa Gambling Task draw “loss cards” and “gain cards” with the goal
of earning as much money as possible. The task involves four different decks, and participants
are told a specific number of cards they must draw. Participants choose which decks they would
like to draw from. Two decks have large gains and large losses. The other two have a smaller
range of losses and gains. Playing the smaller range decks will lead to an overall gain, but
14

playing the “more extreme” decks will lead to an overall loss. Risk-seeking people tend to draw
more from the more extreme decks because the “prospect of delayed punishment is outweighed
by that of immediate gain” (Bechera et al., 1994).
Columbia Card Task
The Columbia Card Task is a risk-taking behavioral-proxy card game. Like other
gambling tasks, the object is to maximize profit. Participants are presented 32 cards. Some cards
denote gains, and some losses. Information about the number of “loss cards” in a deck, the range
of amounts earned for drawing a “gain card”, and the range of amounts deducted from the bank
for drawing a “loss card” are all explicitly stated. For each trial, participants choose the number
of cards they would like to draw, then draw that number of cards all at once. Risk-averse people
are more likely to choose fewer cards, because they recognize the fewer cards draw, the fewer
opportunities for loss (Figner, Mackinlay, Wilkening, & Weber, 2009).
Balloon Analogue Risk Task
The Balloon Analogue Risk Task (BART) is a computer simulation designed to reflect
real-world situations where riskiness pays off up to a breaking point where further action incurs
losses (Lejuez et al., 2002). During the task, participants are asked to inflate a virtual balloon or
stop to collect their winnings. Participants receive a small amount of money for every pump, but
if the balloon pops, participants lose the money they accumulated for that balloon. The
“capacity” of each balloon is decided by chance, and the probability of popping the balloon
increases with each successive pump (e.g. the balloon has a 1/30 probability of popping on pump
1, 1/29 probability of popping on pump 2, etc.). Participants can see the amount of money
accumulated in a “bank” on screen. Participants more comfortable with risk will continue
15

pumping a given balloon, weighing the potential to win more money as more attractive than the
potential to lose the money already earned in the trial (Lujeuz et al., 2003).
Risky, Impulsive, and Self-Destructive Behavior Questionnaire
The Risky, Impulsive, and Self-Destructive Behavior Questionnaire (RISQ) measures
risky behaviors in the domains of drug use, aggression, gambling, sexual behavior, alcohol use,
self-harm, impulsive eating, and general reckless behaviors (Sadeh & Baskin-Sommers, 2016).
The RISQ asks participants to report for each risky behavior: 1) how many times they have
engaged in the behavior in their life 2) how many times they have engaged in it recently 3) how
old they were the first time 4) whether the behavior caused legal, social, or serious health
problems 5) how much the behavior relieves stress and 6) how much the behavior is thrilling.
For the purposes of this study, the total number of times a participant has engaged in a behavior
(1) was treated as the primary indicator of risky behavior. For ethical reasons, suicidal self-harm
questions were excluded as well as items describing assaults and threats made with a deadly
weapon.
Attention check items
Three attention check items were included to screen for random responding. All three
were of instructional manipulation check style, for example “If you believe you are paying
attention to this survey, please select ‘describes me very well.’” Passing an attention check was
scored as simply selecting the instructed response item.
Procedure
Participants were randomly assigned to six different versions of the survey. These
versions differed only in the order in which the measures were presented. Measures were
16

grouped into six “blocks” which were randomized according to a Latin Square design. The
survey materials were administered via Qualtrics.
The decision-making tasks were administered at the same computer, but run through
Inquisit. The Qualtrics survey was interspersed, according to the randomization design, with
pages prompting the participant to let the research assistant open the Inquisit tasks. Before the
start of the study, one of the three tasks was randomly chosen to count for real cash payout
according to the “bank” amount they accrued during that task. Participants were told explicitly
before the start of the randomly assigned cash task that their responses counted for real money.
The tasks were designed to pay an average of $5, with possible payments ranging from $1 to
$10.
After completing the decision-making tasks and the other risk-taking measures,
participants completed a social desirability measure and a demographics questionnaire. Finally,
participants received their cash payment and were dismissed.
17

CHAPTER 3
RESULTS
DOSPERT Structure

Confirmatory factor analyses were conducted to explore the structure of the DOSPERT.
Fit and modification statistics were calculated through Mplus using robust maximum likelihood
estimation. Fit indices for all four models are provided in Table 1. The five-factor model
assumed by the DOSPERT does not fit the data well (χ² = 1030.94, df = 395, CFI = .74, RSMEA
= .07). An examination of the modification indices and residuals suggested that the financial risk
taking items should be split into two factors: investment (e.g. “Investing 10% of your annual
income in a new business venture”) and gambling (e.g. “Betting a day’s income on the outcome
of a sporting event”).
This alternative six-factor solution fits the data better than the five-factor solution (χ² =
821.14, df = 390, CFI = .82, RMSEA = .05). Comparisons between non-nested models can be
made by calculating differences in the Bayesean Information Criterion (BIC), with differences
greater than six indicating better fit by the model with the smaller BIC (Credé & Harms, 2015;
Rafferty, 1995). Using this statistic, the six-factor model displays better fit than the DOSPERT’s
assumed five-factor model (ΔBIC = 193.60).
Two additional models were tested to explore the possibility of a higher-order factor
structure, reflecting a “general risk-taking” construct. The hierarchical model consisted of six
factors loading on one higher-order general factor. This fit of this model is similar to the six
factor solution (χ² = 880.46, df = 399, CFI = .80, RMSEA = .06). Because the chi-square value
calculated by maximum likelihood cannot be used to compute a chi-square difference without
accounting for the scaling correction factor, the difference between the hierarchical model and
18

the nested six-factor model were calculated using the Satorra & Bentler (2010) scaled chi-square
difference statistic. This difference test indicates that the hierarchical model is displays
significantly worse fit than the six-factor solution (Δχ² = 59.83, df = 9). A bi-factor model was
tested as an alternative higher-order to the hierarchical model. The bi-factor solution describes a
structure in which a general factor exists, but is not formed by the six factors. A bi-factor model
is a common higher-order alternative to a hierarchical model (Rindskopf & Rose, 1988). The fit
a bi-factor model, is notably worse than the hierarchical model (χ² = 1064.36, df = 382, CFI =
.72, RMSEA = .07). The difference in BIC confirms the better fit of the hierarchical model
relative to the bi-factor solution (ΔBIC = 281.04). No calculated model meets cutoff criteria for
acceptable fit: CFI and TLI > .95, RMSEA< .05 (Hu & Benlter, 1999). In these analyses, RMSEA and SRMR indices are more indicative of good fit than CFI or TLI. RMSEA and SRMR reflect the difference between the observed and predicted covariance and correlation matrices, respectively, whereas CFI and TLI reflect the difference between the examined model and the null model, in which no components are related (Cook, Kallen, & Amtmann, 2009). It is possible that random responding, which resembles uniform distribution of variables that are uncorrelated, has artificially decreased the fit as estimated by CFI and TLI. Low estimates of fit may also have been a product of the estimation method used in Mplus. Robust maximum likelihood estimation may not have been adequate to address severity of multivariate normality violation in the sample. The distribution of DOSPERT scores is clearly not multivariate normal (Henze-Zirkler = 1.50, p < .001), so analyses were rerun in LISREL to compare results derived from an asymptotic distribution free estimator. Fit indices calculated through LISREL are reported in Table 2. Fit improves with this estimation method. The six- factor solution (χ² = 861.34, df = 390, CFI = .91, RMSEA = .06) appears to fit the data better 19 than the five-factor solution (χ² = 1128.39, df = 395, CFI = .85, RMSEA = .07). The improved fit of both the hierarchical model (χ² = 942.31, df = 399, CFI = .89, RMSEA = .06) and the bi- factor model (χ² = 803.14, df = 375, CFI = .91, RMSEA = .05) suggests there is support for both a general risk-taking propensity factor and a six-factor solution. Because no model tested through confirmatory factor analysis met thresholds of acceptable fit, an exploratory factor analysis was conducted to explore a less constrained, but still plausible structure of the DOSPERT. The exploratory factor analysis was calculated with Mplus using maximum likelihood estimation and an oblique rotation allowing correlation between factors. A parallel analysis was conducted to compare the eigenvalues of the observed matrix to those derived from 100 iterations of random data with the same sample size and number of variables. Six factors were retained (observed eigenvalue = 1.29; average parallel eigenvalue = 1.28) because the retention of seven factors (observed eigenvalue = 1.16; average parallel eigenvalue = 1.25) produces factors formed due to sampling error (Hayton, Allen, & Scarpello, 2004). The six-factor solution does not fit the data according to Hu & Bentler’s (1999) criteria, (χ² = 517.57, df = 270, CFI = .91, TLI = .85, RMSEA = .05, SRMR = .04). Factor loadings are presented in Table 3. Factor 1 appears to capture gambling, as all three DOSPERT gambling items load strongly on this factor. Factor 2 appears to capture recreational risk-taking, as all DOSPERT recreational risk-taking items load on this factor. Factor 3 consists of the DOSPERT investment items and the ethical item describing risky tax deductions, so this factor seems to describe non- gambling financial risk-taking. Factor 4 appears to capture ethical risk taking. All ethical items, as well as health/safety items with possible ethical components (e.g. drinking heavily, unprotected sex) load on this factor. Factor 5 appears to capture social risk taking, though only 20 four of six social items load on this factor. Factor 6 seems to capture information from DOSPERT items that is not described by any original DOSPERT domain. The strongest loadings on factor 6 are from the following items: “walking home alone at night in an unsafe area” “moving to a city far away from your extended family,” “starting a new career in your mid- thirties” and “leaving your young children alone at home while running an errand.” The items making up factor 6 are ethical, health/safety, social, and recreational risk-taking activities. The common element in these items may be risk-taking behavior that results from a strong sense of independence and possible overestimation of self-efficacy. Communalities were calculated by subtracting the estimated residual variances from 1. Some item’s communalities are quite low (e.g. “admitting that your tastes are different from those of a friend” communality = .08; “sunbathing without sunscreen” communality = .12; “starting a new career in your mid-thirties” communality = .18), suggesting that the exploratory factor structure cannot account for a sufficient amount of variance in all DOSPERT items. Correlations between DOSPERT and non-domain specific measures Sensation-seeking was measured by two self-report inventories: the Arnett Sensation Seeking scale and the Sensation Seeking Scale – Form V. In this sample, correlations between the Arnett Sensation-Seeking scale and all other risk-taking measures are negative or near zero. These associations should not indicate that the Arnett Sensation-Seeking scale does not measure risk-taking or that sensation seeking is negatively associated with similar constructs. This scale uses a response format worded in the opposite direction of all other scales. It is probable that a non-trivial proportion of participants in this sample assumed the response format was in the same direction as others, or too few participants read the response options carefully. Because the Arnett scale format was potentially confusing, “sensation-seeking” in the following analyses