Beyond the Privacy Paradox: Objective Versus Relative Risk in Privacy Decision Making

Abstract

Privacy decision making has been examined in the literature from alternative perspectives. A dominant “normative” perspective has focused on rational processes by which consumers with stable preferences for privacy weigh the expected benefits of privacy choices against their potential costs. More recently, a behavioral perspective has leveraged theories from decision research to construe privacy decision making as a process in which cognitive heuristics and biases predictably occur. In a series of experiments, we compare the predictive power of these two perspectives by evaluating the impact of changes in the objective risk of disclosure and the impact of changes in the relative perceptions of risk of disclosure on both hypothetical and actual consumer privacy choices. We find that both relative and objective risks can, in fact, influence consumer privacy decisions. However, and surprisingly, the impact of objective changes in risk diminishes between hypothetical and actual choice settings. Vice versa, the impact of relative risk becomes more pronounced going from hypothetical to actual choice settings. Our results suggest a way to integrate diverse streams of the information systems literature on privacy decision making: in hypothetical choice contexts, relative to actual choice contexts, consumers may both overestimate their response to normative factors and underestimate their response to behavioral factors.

Full text

PRELIMINARY DRAFT
Beyond the Privacy Paradox: Objective versus Relative Risk in Privacy Decision Making
Idris Adjerid*, Eyal Peer**, Alessandro Acquisti***
*University of Notre Dame
**Bar-Ilan University
***Carnegie Mellon University
Abstract:
Privacy decision making has been examined from various perspectives. A dominant “normative”
perspective has focused on rational processes by which consumers with stable preferences for
privacy weigh the expected benefits of privacy choices against their potential costs. More recently,
an alternate “behavioral” perspective has leveraged theories from behavioral decision research to
construe privacy decision making as a process in which cognitive heuristics and biases
predictably occur. In a series of experiments, we compare the predictive power of these two
perspectives by evaluating the impact of changes in objective risk of disclosure and the impact of
changes in relative perceptions of risk of disclosure on both hypothetical and actual consumer
privacy choices. We find that both relative and objective risks can, in fact, impact consumer
privacy decisions. However, and surprisingly, the impact of objective changes in risk diminishes
between hypothetical and actual choice settings. Vice versa, the impact of relative risk is more
pronounced going from hypothetical to actual choice settings. Our results suggest a way to
integrate diverse streams of IS literature on privacy decision making: consumers may both over-
estimate their response to normative factors and under-estimate their response to behavioral
factors in hypothetical choice contexts relative to actual choice contexts.
1
Published Version Can Be Found:
MIS Quarterly
Vol 42 Issue 2, 2018, 465-488
DOI: 10.25300/MISQ/2018/14316

PRELIMINARY DRAFT
1. Introduction
A significant amount of research across multiple disciplines (such as management and
information systems, economics, marketing, psychology, and human computer interaction) has
explored consumer privacy decision making. Historically, much of the information systems (IS)
research in this area has focused on a “normative” perspective of consumer choice. Normative
theories of consumer choice are those consistent with the classical economic view of consumers
as deliberative, utility maximizing, rational agents who possess reasonably stable, and therefore
predictable, preferences for goods (Simon 1959; Thaler and Mullainathan, 2000). Under this
perspective, privacy decisions can be construed as the result of a mental calculus that weighs the
expected benefits of privacy allowances against their resulting costs (Klopfer and Rubenstein,
1977; Milne and Gordon, 1993; Dinev and Hart, 2006).
While the importance of privacy risks and benefits in influencing consumer behavior has
been proven time and again in the literature (e.g. Marthews and Tucker, 2014), this account of
decision making has faced the challenge of explaining surprising, yet robust, empirical
occurrences in privacy contexts. This includes the dichotomy between stated privacy attitudes, or
stated privacy intentions, and actual behaviors (Spiekermann, Grossklags, and Berendt, 2001
Jensen, Potts, and Jensen, 2005), as well as seemingly contradictory reactions to privacy trade-
offs (John, Acquisti, and Loewenstein, 2011; Brandimarte, Acquisti, and Loewenstein, 2013). In
the last few decades, a growing body of work in economics and decision research has proposed
“behavioral” perspectives on decision making (Thaler and Mullainathan, 2000; Camerer,
Loewenstein, and Rabin, 2011). These perspectives are rooted in the notion that systematic, and
therefore predictable and replicable, deviations from normative (that is, rational-calculus based)
accounts of consumer decision can arise due to limitations in consumers’ cognitive ability, or
their susceptibility to behavioral heuristics and decision biases. In the privacy context, this
perspective suggests that factors independent of both objective trade-offs associated with privacy
choices and consumer preferences can still significantly influence consumer behavior. For
2

PRELIMINARY DRAFT
example, default options which are trivial to unselect may have a major impact on consumes’
privacy choices (Johnson, Bellman, and Lohse, 2002).
Both accounts of privacy decisions stem from legitimate theoretical frameworks and have
stimulated considerable bodies of empirical research. However, most of IS privacy research has
primarily focused on either the normative or behavioral perspectives of privacy decision making.
As a result, comparisons between the results produced within the two literatures are post hoc,
requiring meta-analysis across studies with diverse modeling assumptions and empirical
methodologies. Given the significant and growing social and economic implications of consumer
privacy decision making, the absence of a bridge between the two streams of work represents a
considerable gap in the literature and is thus the focus of this manuscript. We use a series of
experiments to investigate normative and behavioral perspectives within the same empirical
settings. We evaluate whether these alternative accounts of privacy decision making can both
individually account for some of the variation in observed consumer privacy choices.
Furthermore, we explore the conditions under which a normative or a behavioral account may
differentially explain privacy choices.
We focus on informational privacy, and the impact that different degrees of data
protection can have on consumers’ willingness to reveal personal information – a construct
common in both streams of literature. Across three experiments, we manipulate either normative
factors, behavioral factors, or both simultaneously. Many manipulations of either type are
available; we lean on the extant privacy and behavioral literatures to identify established
manipulations within each perspective to use in our experiments. Specifically, we manipulate a
normative factor by varying the objective levels of protection afforded to disclosures of personal
information (for instance, whether responses are identified or anonymous). We manipulate a
behavioral factor by holding the objective levels of information protection constant, but varying
the relative perception of changes in privacy protection afforded to disclosures of personal
information, based on seminal research on Prospect Theory and reference dependence
3

PRELIMINARY DRAFT
(Kahneman and Tversky, 1979). We capture the impact of these factors on both hypothetical
(self-reported) and actual disclosure behavior. This is relevant in light of existing literature
suggesting that consumers may both over-estimate their response to normative factors in
hypothetical settings due to more favorable beliefs and attitudes about protecting ones’ privacy
(Ajzen, I., Brown, T. C., and Carvajal, F., 2004) and fail to anticipate the effect of behavioral
factors when considering a choice in hypothetical settings (Loewenstein and Adler, 1995;
Liberman, Samuels, and Ross, 2004). We explore whether those tendencies result in normative vs.
behavioral perspectives having differential effects on hypothetical vs. actual behavior.
Across the experiments, we find evidence that both objective and relative differences in
privacy protection impact consumer privacy decisions, lending credence to the notion that both
perspectives (normative and behavioral) capture aspects of privacy choice. However, we also find
that objective privacy protections had a diminished effect on privacy decision making in contexts
that involve actual privacy choices relative to hypothetical privacy choices, whereas relative
changes in privacy protection had a pronounced effect in actual settings relative to hypothetical
ones. Specifically, we find that in a hypothetical context (Experiment 1), differences in objective
privacy protection result in significant differences in participants’ reported privacy concerns and
their willingness to disclose personal information. In contrast, changes in relative risk result in
smaller (although significant) differences in reported privacy concerns, and no differences in
hypothetical willingness to disclose personal information. Mirroring what participants were asked
to imagine in Experiment 1, in a context with actual disclosures (Experiment 2) we find the
opposite effect: differences in objective privacy protection have a small effect on participants’
self-disclosure, while relative changes in privacy protections strongly influence participants’
propensity for self-disclosure. In a final experiment (Experiment 3), we consider together both
normative and behavioral perspectives by manipulating all dimensions simultaneously (objective
protection vs. relative protection, and hypothetical vs. actual choice). We find results consistent
with the two prior experiments: both objective and relative changes in protection can have an
4

PRELIMINARY DRAFT
impact on privacy decision making; however, objective changes have pronounced effects in
hypothetical settings, while relative changes have pronounced effects in actual choice settings.
These finding contributes to the IS literature on the drivers and predictors of privacy
decision making. While the normative perspective for privacy decision making is now well
studied, the behavioral perspective is still developing. While our findings bolster the role of the
nascent but growing behavioral perspective, they also provide evidence of the simultaneous (yet
uneven) role of both normative and behavioral factors. Such findings have implications beyond
the privacy literature. While the study of behavioral factors (such as applications of reference
dependence and prospect theory) continues to garner interest from the broader research
community across varied contexts,1 their application to the information systems literature is
growing (Herrmann, Kundisch, and Rahman, 2014), but still relatively sparse. This is an area ripe
for exploration, considering that many technology choice contexts, related and unrelated to
privacy, are highly dynamic, bestowing on prospect theory a potentially important role in
understanding consumer decision making in these contexts.
2. Conceptual Background and Theory
A prominent focus of economic research in the last half a century has been understanding the
bounds of rational consumer choice and reconciling traditional neoclassical theory with an
accumulating body of empirical and theoretical research supporting behavioral accounts of
consumer decision making (e.g. Ho, Lim, and Camerer, 2006, Camerer, Lowenstein, and Rabin,
2011). In more recent years, behavioral research has started informing a number of other domains,
including information systems. Goes (2013), for instance, highlights the need to incorporate
insights from behavioral economics into theoretical and empirical IS research.
The interplay of rational choice and behavioral accounts of decision making is particularly
prominent in the context of consumer privacy choice. Extant literature in this area has been
1 Bartling, Brandes, and Schunk (2015) recently showed that reference dependent decision making seems to hold in the
context of coaches and players of professional soccer.
5

PRELIMINARY DRAFT
largely predicated around the notion that privacy decision making is largely a rational process
driven by what we may refer to as “normative” factors – that is, factors that are normal to
consider if an agent is attempting to maximize her utility. Such factors may include the objective
benefits and costs of information disclosure, and the agent’s stable, coherent preferences (Simon
1959; Thaler and Mullainathan, 2000). For instance, a privacy calculus view of consumer
decision posits that privacy is subject to interpretation in “economic terms” (Klopfer and
Rubenstein, 1977) and that consumer privacy choices are driven by a systematic weighing of the
benefits of information disclosures against the perceived privacy risks from such disclosures
(Milne and Gordon, 1993; Dinev and Hart, 2006). Along these lines, Westin (2000) posited that
most consumers are shrewd privacy balancers who weigh the value to them and society of various
business and government programs calling for personal information. Relatedly, a considerable
body of work has focused on identifying systematic differences in privacy concerns between
consumers (e.g. Smith, Milberg, and Burke, 1996) and has suggested that elevated privacy
concerns correspond to privacy seeking behavior, such a diminished willingness to disclose
personal information (Malhotra, Kim, and Agarwal, 2004).
A more recent theme in the IS privacy literature has been the focus on factors that ostensibly
should have little (or even no) direct impact on objective risk and benefits of disclosure, but
which nevertheless considerably impact people’s privacy concerns and personal preferences for
self-disclosure (e.g., Moon, 2000). For example, people respond more honestly and with higher
rates of disclosure to an online version, versus a paper-and-pencil version, of the same
questionnaire (Tourangeau, 2004), even though online responses are more likely to be tracked,
duplicated, disseminated, shared with or accessed by a larger number of parties than a
questionnaire filled out on a single copy of paper. Similar effects emerge when comparing online
disclosure to those made during face to face communication (e.g., Harper and Harper, 2006).
Also, people seem to rely on contextual cues, such as a survey’s look and feel or implicit social
norms, when disclosing intimate details about themselves (John, Acquisti, and Loewenstein,
6

PRELIMINARY DRAFT
2011). Or, holding objective risk constant, the mere increase in perceived control over who can
access and use online personal information can result in an increased likelihood to make sensitive,
risky disclosures (Brandimarte, Acquisti, and Lowenstein, 2013). This behavioral perspective
suggests that consumer privacy preferences may be malleable rather than stable, and that privacy
behavior is not just highly context-dependent (something that may be also predicted by rational
calculus-grounded theories of privacy decision making), but can in fact be affected by factors
with little relationship to changes in objective trade-offs from disclosure, such as order effects,
framing, and other decision heuristics.
Much of the privacy literature has studied consumer privacy decision making by evaluating
the impact of varying degrees of privacy protection and assurances on consumers’ behavior, often
with consumer disclosure or engagement with a commercial entity as outcomes of interest.
However, most of privacy IS research has focused on either the normative or behavioral
perspectives of privacy decision making. In this manuscript, we study the impact of changes in
privacy protection and assurances on consumers’ privacy behavior, but consider simultaneously
normative and behavioral perspectives. We do so by manipulating either objective changes in
privacy protection, or relative changes in perceptions of protection, in two alternative contexts
extensively examined in previous research: hypothetical self-disclosure choices (that is,
behavioral intentions), and actual disclosure decisions.
2.1. Objective Changes in Privacy Protection
A substantial body of research suggests that changes in expected privacy benefits and risks can
affect consumers’ observed privacy choices. For example, disclosing personal information can
lead to consumer benefits such as an improved experience in retail via customization of products,
promotions, and even user interfaces (Ansari and Mela, 2003), enable users to derive personal
and economic value from social networks (Ellison, Steinfield, and Lampe, 2007), and underlies
business models for online services providing free content and applications (Leontiadis et al.,
2012). Similarly, the literature has noted a number of potential risks of loss due to these
7

PRELIMINARY DRAFT
information disclosures, which include those stemming from the misuse of disclosed data
(Featherman and Pavlou, 2003), sharing of personal information with third parties, or price
discrimination as a result of information disclosures (Viswanathan et al., 2007). Prior literature
suggests that the relationship between consumers’ perceived risk and behavior is critically related
to consumer trust, both in terms of its impact on consumers’ perception of risk (Kim, Ferrin, and
Rao, 2008; Vance, Elie-Dit-Cosaque, and Straub, 2008) and as a mediator explaining why shifts in
privacy risk impact behavior. For example, Dinev and Hart (2006) show variation in the
perceived risk of a context impacts behavioral intentions both directly and through a strong effect
on consumer trust and privacy concern for a particular context.
Within this general paradigm, it follows that privacy protections have the potential to
influence consumer privacy decision making via their impact on the perceived risks of misuse of
consumers’ personal information. For example, Dinev and Hart (2006) suggest that assurances
from salespeople can mitigate perceptions of risk – which, in their model, would diminish privacy
concerns and increase trust, thus leading to changes in behavior. A number of studies test this
conjecture by shifting the objective degree of privacy protection afforded to consumers’ personal
information and find evidence for various facets of this model of consumer behavior. For instance,
Culnan and Armstrong (1999) find that the use of fair information practices by firms can
engender trust from consumers, reducing privacy concerns and perceived risks of disclosure. Xu
et al. (2009) find that self-regulation and government regulation reduce perceived risk from
participating in location-based services and increase consumers’ intention to disclose personal
information. Miyazaki and Krishnamurthy (2002) find a significant effect of privacy seals on
consumer perception of firm privacy practices and their stated willingness to disclose personal
information. And Xu et al. (2012) find that industry self-regulation and government regulation
reduce consumer privacy concerns.
Using treatments of privacy assurances similar to those employed in this literature, we
manipulate the objective degree of protection afforded to consumers’ disclosure (e.g. the breadth
8

PRELIMINARY DRAFT
of access to personal information and the anonymity of responses) afforded to participants via a
“privacy notice” (similar to firm privacy policies). We capture the role of normative factors
(objective changes in risks and benefits) on privacy behavior through the following hypothesis:
H1: Changes in objective levels of privacy protection will impact disclosure: lower levels of
privacy protection will lead to lower levels of disclosure of personal information.
2.2. Relative Shifts in Privacy Protection
As noted in the Introduction, a growing body of empirical evidence has suggested that behavioral
factors with little or no direct impact on objective risk and benefits of disclosure, can in fact
considerably impact people’s privacy concerns and personal preferences for self-disclosure.
Decision biases and heuristics can significantly affect privacy decision making (Acquisti,
Brandimarte, and Loewenstein 2015). In addition, scholars have identified significant influences
of affect and emotions on consumer privacy judgments (Li, Sarathy, and Zhang, 2008; Li, Sarathy,
and Xu, 2011). Within the behaviorally-grounded body of research on privacy, increasing
attention has been paid to the fact that privacy judgments can be relative in nature (Acquisti, John,
and Loewenstein 2012; Egelman, Felt, and Wagner, 2013): consumers may compare their
(current) situation to that of other people, or to their situation in the past, and their decisions may
be affected by phenomena such as habituation or coherent arbitrariness. This phenomenon may be
particularly salient in privacy contexts where heterogeneity in data practices abound (across firms
and over time). For instance, firms that aggregate consumer privacy information often highlight
improvements (i.e. relative changes) to consumer privacy over time2 and sometimes notify
consumers of their privacy protections in a manner that highlights the relative privacy gains from
their services compared to those of their competitors3.
2 Facebook, for example, has been known to advertise updates to privacy policies to highlight gains to consumer
privacy: https://www.facebook.com/notes/10150251867797131.
3 Microsoft’s “Scroogled” Ad campaign sought to highlight the privacy protectiveness of their services (e.g. search,
email, etc.) relative to those of Google. https://en.wikipedia.org/wiki/Scroogled.
9

PRELIMINARY DRAFT
Kahneman and Tversky (1979)’s Prospect Theory (PT) is a useful framework for
studying the potentially relative nature of consumer privacy choices. PT posits that consumers
evaluate outcomes both with respect to objective levels of consumption and with respect to a
reference point, treating outcomes above or below the reference point as gains or losses,
respectively. In other words, PT allows for both the impact of objective features of a particular
choice context that should influence choice (e.g. price of a product) and the features of a
particular context that, according to classic accounts of economically rational decision making
(e.g., Von Neumann and Morgenstern, 1944), should not have an impact on behavior. Therefore,
PT offers a framework for analyzing and compare the impact of both normative vs. behavioral
factors on privacy choice: that is, how objective changes in privacy risk can affect privacy
decision making, but also how changes in relative perceptions of privacy risk, in absence of
changes in objective risk, can still affect can affect consumer privacy decision making.
Considerable empirical evidence supports the notion of reference dependent decision
making, and rules out alternate rational explanations of reference dependence (e.g., lack of
information or consumer inexperience with a choice context). For example, Kahneman and
Tversky (1979) found that individuals are much more likely to accept a gamble when the choice
is framed as avoiding a loss compared to when the objectively equivalent choice is framed as
obtaining a gain. Moreover, seminal work on the endowment effect (e.g., Kahneman, Knetsch &
Thaler, 1991) highlights significant differences in the amount buyers are willing to pay (WTP) for
an item compared to the amount sellers are willing to accept (WTA) for the same item. Such a
WTA-WTP gap has been attributed to the difference between buyers’ and sellers’ reference point:
whereas buyers consider the purchase of a new item as a gain, sellers consider it as a loss (e.g.,
Novemsky & Kahneman, 2005). A similar WTA-WTP gap has also been found to operate in the
context of disclosure decisions (Acquisti, John, & Lowenstein, 2013). In fact, recent literature
(e.g., Koszegi and Rabin, 2007) has incorporated reference dependence in classical models of
consumer utility, allowing for consumer utility to be derived from both objective features of a
10

PRELIMINARY DRAFT
choice set and also deviations from a reference point. Simply put, Prospect Theory offers a
theoretically and empirically validated framework that offers defensible deviations from
normative models of decision making, including those that relate to normative perspectives of
privacy decision making.
We use insights from Prospect Theory and the empirical literature on reference
dependence to evaluate the impact of relative changes in privacy protection on privacy decision
making. Under normative perspectives, identical privacy notices should result, on average, in
comparable levels of disclosure irrespective of relative changes in privacy notices. However,
under an alternative account of decision making that incorporates reference dependence,
consumers would evaluate privacy notices relative to their deviation from a reference point, such
as the level of protection they had in the recent past or the one they currently use (i.e., the status
quo). We capture the role of behavioral factors (relative changes in risks and benefits) on privacy
behavior through the following hypothesis:
H2: Relative perception of the level of privacy protection will influence individual
privacy decision making: levels of privacy protection perceived to be higher relative to a
reference point will result in higher levels of disclosure of personal information.
2.3. Privacy Behavior in Actual vs. Hypothetical Choice Contexts
Given that compelling accounts and empirical evidence exist for both normative and behavioral
perspectives on privacy decision making, it is useful to consider which factors may moderate the
effect of objective and relative privacy protection on privacy decision making. One such factor
could be whether privacy decision making is being studied in hypothetical settings (and captured
in the form of attitudes or behavioral intentions), or in behavioral settings (and captured in the
form of actual behaviors and choices). Smith et al. (2011) note that “it is quite common for
researchers to measure stated intentions instead of actual behaviors” in the extant literature. On
the other hand, the behavioral literature cited previously predominantly uses actual choice as
outcomes of interest (see, e.g., John, Acquisti, and Loewenstein, 2011; Acquisti, John, and
11

PRELIMINARY DRAFT
Loewenstein 2012; Egelman, Felt, and Wagner, 2013; Brandimarte, Acquisti, and Loewenstein,
2013). Of course, one possibility is that both normative and behavioral factors influence behavior
to the same degree in hypothetical and actual choice settings, rendering this empirical distinction
between the literatures inconsequential. Alternatively, normative and behavioral factors may play
different roles when moving from hypothetical to actual choice settings. This can emerge if
consumers either misjudge the impact of normative factors between hypothetical and actual
choice settings, if they misjudge the impact of behavioral factors between hypothetical and actual
choice settings, or of course both.
We first consider the potential of consumers to misjudge their reaction to normative
factors moving from hypothetical to actual choice settings. In the psychology literature,
hypothetical bias refers to a divergence between behavioral intentions in hypothetical contexts vs.
actual behavior in real life settings (LaPiere, 1934; Champ, Bishop, and McCollum, 1996;
Murphy, 2005). For instance, Sheeran (2006) finds that individuals overstate their propensity to
use condoms, undergo a cancer screening, or exercise in hypothetical relative to the same actual
choice settings. FeldmanHall et al. (2012) find that subjects say they will give up more money to
spare others from mild electrical shocks than they actually do when the shocks are real. In
addition, Ajzen, Brown, and Carvajal (2004) find that individuals significantly overstate their
propensity to donate to a scholarship fund in hypothetical vs. actual choice settings. Ajzen,
Brown, and Carvajal (2004) explain this intention-behavior by the “activation of more favorable
beliefs and attitudes.” Specifically, they find that participants in the hypothetical choice setting
were significantly more likely to indicate that donating to the fund was a social norm (e.g. more
like to indicate that they “should” contribute and that those close to them would do the same).
They also found that hypothetical settings elicited more favorable attitude towards donating (e.g.
more likely to indicate that the behavior was “good” rather than “bad”, “right” rather than
“wrong”, etc.). If choice contexts that involve protecting ones’ privacy carry similar dynamics,
then consumers’ hypothetical evaluations of high vs. low levels of privacy protection (i.e., our
12

PRELIMINARY DRAFT
normative manipulations) may be affected by overly positive attitudes related to protecting one’s
privacy as well as elevated perceptions of it being a social norm. If these positive attitudes and
perceptions of others’ behavior do not carry over to actual choice settings (as the literature would
suggest), then we may expect consumers to overstate their response to objective changes in
privacy protection in hypothetical settings relative to actual ones.
Simultaneously, the impact of behavioral factors may vary when moving from
hypothetical to actual behavior. The behavioral economics literature evaluates behavioral factors
across hypothetical and actual choice settings and finds that, at a minimum, we can expect
behavioral factors to have an impact actual choice settings: Knetsch, Tang, and Thaler (2001) find
that the endowment effect is robust to repeat trials in actual choice settings; Lichtenstein and Slovic
(1971) show preference reversals consistent with behavioral models of choice using actual
behavior; and Pommerehne, Schneider, and Zweifel (1982) conclude that even ‘‘when the
subjects are exposed to strong incentives to make motivated, rational decisions, the phenomenon
of preference reversal does not vanish.”4 Several works, in fact, have suggested that the impact of
behavioral factors may even be pronounced in actual choice settings relative to hypothetical ones.
For instance, prior work has documented a hot-cold empathy gap between hypothetical and actual
choice settings where consumers are not able to anticipate their future “hot” states in actual
choice setting (e.g. the impacts on choosing a health entrée of being hungry) when considering
the same choices hypothetically (Loewenstein 2000; Kang and Camerer, 2011). More so,
Loewenstein and Adler (1995) find that participants consistently underestimated the impact of
being given an item on their subsequent valuation of that item (i.e. the endowment effect);
O’Donoghue and Rabin (2000) find that consumers can be naïve in their estimation of their own
susceptibility to an immediate gratification bias (i.e. time inconsistent discounting); and Liberman,
4 Relatedly, the experimental economics literature has questioned the use of hypothetical choice settings, although on
different grounds: the absence of real economic tradeoffs in hypothetical settings is seen as likely to produce behavioral
intentions that may not match actual behavior because individuals have less to lose (Smith, 1991, Conslik, 1996). This
has lead some authors to suggest that non-normative models of behavior may emerge in hypothetical choice settings
but may wane in actual choice settings (Plott and Zeiler, 2005). The conclusion has been, however, critiqued and
challenged in the literature (Fehr, Hakimov, and Kübler, 2015; Isoni, Loomes, and Sugden, 2011).
13

PRELIMINARY DRAFT
Samuels, and Ross (2004) find that participants grossly underestimate subtle framing changes to
the labels of choice contexts on their subsequent behavior. Kühberger, Schulte-Mecklenbeck, and
Perner (2002) examine differential effects of behavioral factors between hypothetical and actual
choice settings using positive vs. negative framing manipulations (manipulations rooted in
prospect theory from above) and find an “economic anomaly” in that “real decisions with large
amounts do not diminish the framing effect; it may even be stronger than is apparent from
hypothetical decisions.” Translated to privacy contexts, those findings suggest that consumers
may fail to anticipate their “hot” state or susceptibility to behavioral factors (e.g. how privacy
choice contexts are framed) when considering hypothetical disclosures relative to actual ones. In
sum, this work suggests that behavioral factors may have a pronounced effect on actual relative to
hypothetical choice.
Taken together, prior literature suggests a novel account of the influence of normative vs.
behavioral factors on consumer privacy decision making. In addition to our formal hypotheses,
which test whether both normative and behavioral factors can have an influence on privacy
decision making, we explore whether the impact of normative factors (changes in objective
protections in our context) may diminish going from hypothetical to actual privacy choices while
the impact of behavioral factors (relative changes in protection in our context) may be
pronounced going from hypothetical to actual choice contexts.
3. Methods
In three experiments, we evaluate the role of objective changes in privacy protection and of
relative judgments of privacy protections on participants’ hypothetical and actual privacy-
sensitive behaviors (self-disclosures and selections of privacy settings). Experiment 1 is a
hypothetical study in which participants are presented with questions of a personal and sensitive
nature, and graphical privacy notices are used to manipulate either the objective protection or the
relative perception of privacy protection afforded to the answers provided by subjects;
participants are asked to report both their privacy concerns and their hypothetical disclosure
14

PRELIMINARY DRAFT
behavior. Experiment 2 uses a similar context as that of Experiment 1 but captures actual
disclosure behavior, and uses textual privacy notices (similar, although simplified, to those used
by online services) to manipulate objective and relative changes in privacy protection. Finally,
Experiment 3 combines and extends Experiments 1 and 2 by manipulating simultaneously
changes in the objective and relative levels of privacy protection, and by capturing both
hypothetical and actual behavior.
We measure self-disclosure by capturing participants’ answers to questions of personal
and sensitive nature. This approach has been used successfully in the past to examine privacy-
sensitive behaviors (e.g., Moon, 2001, Acquisti et al., 2012). While the ostensible goal of the
studies is to investigate participants’ engagements in various behaviors (for instance: “Have you
ever looked at pornographic material?”; see examples in Appendix A.2), we are not interested in
those behaviors per se, but rather in whether or not participants are willing to disclose information
about their engaging in them. As all of our experiments use random assignments to the different
conditions, the distribution of participants’ actual past engagement in these activities can be
assumed to be similar across conditions. Thus, higher or lowers admittance rates across
conditions signal an impact of the manipulation on self-disclosure levels.
A strong rationale for this method is its ability to circumvent significant obstacles in
obtaining actual self-disclosures from participants. First, it avoids divulging that the goal of the
study relates to privacy and self-disclosure, allowing participants to act more naturally without
being primed by experimenter demand effects. Second, simply asking participants to disclose
sensitive information like SSN or health information may have legal or ethical implications and is
difficult to interpret since we cannot ascertain when consumers choose to mask information or
disclose it (i.e. we cannot tell the difference between a fake and an actual response if they provide
a nine digit number for SSN). Alternatively, it is unlikely that participants with privacy concerns
admit to a behavior when they haven’t engaged in it (i.e. mask their lack of engagement in a
sensitive activity). This leaves admissions in our context as the complement of the sum of (1)
15

PRELIMINARY DRAFT
people who did not engage in the behavior (which we assume to be equal across conditions) and
(2) those that did not admit to the behavior although they engaged in it. John, Acquisti, and
Loewenstein (2011) note that this approach is conservative because the impact of a given
experimental manipulation “has to rise above the noise (error variance) produced by differences
in true rates of engaging in the behavior across conditions”.
Our participants are sampled from two different but complementary online participant
pools, allowing us to test the robustness of the findings. We recruited participants for
Experiments 1 and 2 from Amazon Mechanical Turk (AMT). Prior research has validated AMT
samples as at least as representative as other internet samples, and significantly more
representative than student samples (Buhrmester, Kwang, and Gosling, 2011); furthermore,
central findings in IS and the decision sciences have been replicated using AMT samples
(Goodman, Cryder, and Cheema, 2013; Steelman, Hammer, and Limayem, 2014). Moreover,
AMT offers an effective payment and reputation management system that offers researchers the
ability to only sample participants of higher quality (Research has shown that targeting these
participants ensures high data quality; see Peer, Vosgerau, and Acquisti, 2013). In Experiment 3,
we use another crowdsourcing platform called Prolific Academic, which is similar in most
respects to AMT, except that participants on that platform only participate in academic research
(whereas AMT also offers commercial uses).
3.1. Estimation Approach
Across the three experiments in this manuscript, we evaluate the impact of randomized
manipulations on non-repeating dependent variables (e.g. measures of privacy concerns) and
repeated measures of information disclosure where a single participant is asked to make a series
of hypothetical or actual disclosure decisions. For non-repeated measures, we evaluate the impact
of our randomized treatments using the appropriate statistical tests for our variable of interest (e.g.
t test, chi-square test, etc.). Our evaluation of participants’ disclosure behavior relies on
comparably more complex tests. Because participants across all experiments were presented a
16

PRELIMINARY DRAFT
series of questions (asking them either to predict their propensity to make, or to actually make,
sensitive disclosures), we observe multiple, correlated responses from each single participant. As
a result, we use a random effects linear regression model to evaluate differences in average
disclosure between conditions.5 This model accounts for the correlation between responses from a
given participant when estimating the variance-covariance matrix of the coefficients, assuming
constant correlation ρ between any two answers within a participant (exchangeable correlation
structure: Liang and Zeger, 1986). Specifically, we estimate the following general model:
Disclosureij =
β
*Treatmenti +
δ
* Xj + α*Yi + θi + uij
Disclosureij measures a participant’s predicted or actual propensity to disclose sensitive
information or admit to sensitive behavior, i = (1,…,N participants), and j = (1,..,k questions). In
some specification, we also include Xj: a vector of controls for different features of the questions
asked to participants; for instance, Intrusivej controls for questions that differ in their
intrusiveness. Yi is a vector with controls for participant specific controls (e.g. age and gender). θi
is the participant-specific random effect and uij is the error term. Estimates on randomly assigned
treatments (Treatmenti) are unbiased as they should be uncorrelated with observed (Xj, , Yi ) and
unobserved (θi) individual differences and the error term uij. While our controls are not necessary
for the unbiased estimation of the effect of our treatments on disclosure behavior, they are
included in some specifications to rule out any breaks in randomization, and account for some of
the variation in disclosure behavior between participants.
4. Experiment 1
In Experiment 1, we manipulated, between subjects, either changes in objective levels of privacy
protection, or changes in perceived levels of privacy protection (increase or decrease over time)
5 We use a linear probability model estimation in lieu of a non-linear estimation approach (e.g. logit) for the
straightforward interperation of regression coefficients and the flexibility of OLS in analyzing both likert scale
dependent variables and binary outcomes. Angrist and Pischke (2008) have shown little qualitative difference between
the Logit and linear probability specification.
17

PRELIMINARY DRAFT
while actually holding objective privacy levels constant. We used hypothetical willingness to
disclose as our key dependent variable.
4.1. Participants
Two hundred and twenty one participants from AMT (Mfemale=37.56%; Mage = 29.16, SDage =
9.76) completed the study and were paid $0.30.
4.2. Design and procedure
Participants were asked to provide their personal opinions regarding two surveys that our research
group was, ostensibly, planning to conduct. Participants were told that our research group
conducts surveys which include sensitive questions on ethical behavior and that the
confidentiality protections for these surveys can vary depending on the study. Specifically,
participants were asked for their opinions regarding two surveys called Survey A and Survey B.
First, participants were given a description of Survey A, including the level of privacy protection
of the survey. Protection levels were described using a figure that showed, on five parameters, the
degree to which participants’ privacy would be protected during the study – for instance, whether
certain identifying information would or would not be required (and possibly linked to the
answers provided by the participants); or whether the survey offered a particular protection to the
answers provided by the subjects. In one condition, the first survey (Survey A) provided a low
overall privacy protection level with the “Less Protective” option for four of the five parameters
described (see Figure 3a), and in the other condition, the survey provided a high privacy
protection level with the “More Protective” option for four of the five parameters described (see
Figure 3b). All other details of the survey (length, purpose and payment) were the same in both
conditions. Participants were then asked a set of questions that confirmed they had evaluated and
understood each dimension of the notice provided (e.g. “Are responses kept after the study ends”)
They were then asked to report their satisfaction with the protections provided in each survey,
their perception of potential harm from disclosure in the study, and their concerns about their
privacy (see Appendix A.1). Finally, participants were asked questions gauging their hypothetical
18

PRELIMINARY DRAFT
willingness to disclose for descriptive but sensitive information (e.g. address or phone number)6,
and how often had they engaged in a set of potentially sensitive or even unethical behaviors (see
Appendix A.2). Similar to the extant literature using hypothetical or intended behavior (e.g.
Dinev and Hart, 2006, Xu et al. 2009), we measured their behavioral intention to disclose this
information using five point scales ranging from very likely to very unlikely.
Next, all participants proceeded to review a second survey (Survey B) which provided a
medium privacy level in both conditions (see Figure 3c). Participants were asked to evaluate
Survey B using the same questions as used for Survey A. The level of protections afforded in
Survey B was designed so that participants in the first condition would perceive an increase in the
privacy level from Survey A to Survey B (low to medium), whereas participants in the second
condition would perceive a decrease in the privacy level from Survey A to Survey B (from high
to medium). We evaluate participants’ perception of privacy concern, potential harm to them, and
satisfaction with protections in Survey A and B to evaluate whether they indeed perceived a
decrease or increase in protections between conditions. Notably, the actual level of privacy for
Survey B remained the same for both conditions, although the subjective level of that survey’s
privacy might have changed.
[Figure 1a: Low Protection]
[Figure 1b: High Protection]
6 These questions were not used in actual disclosure settings since, as we note in our methods section,
validation that responses were truthful is not possible.
19

PRELIMINARY DRAFT
[Figure 1c: Medium Protection]
4.3. Results
We found that participants, by and large, were able to accurately understand the notices provided
in the study. For surveys A and B, 91.85% and 94.57% correctly recalled at least four of the five
dimensions. We also found that our manipulation of objective risk using a high and low
protection notice (Figure 3a and 3b) was effective at influencing the perception of privacy
protection in the first survey (Survey A): Participants provided high protections reported being
significantly more satisfied with the protections provided (MHigh=3.36, MLow=1.56), t (219) =
12.15, p < .001, d = 1.64, significantly less concerned about privacy (MHigh=2.39, MLow=3.87), t
(219) = -12.15, p < .001, d = 1.64, and significantly less concerned that harm would come them
as a result of disclosing personal information (MHigh=2.86, MLow=4.02), t (219) = -7.46, p <.001,
d=1—see Table 1.
We evaluated the impact on participants’ predicted disclosure behavior of differences in
objective privacy protection using a random effects linear regression estimation approach.
Participants reported on a five item scale (1 indicating being “Very Unlikely” to disclose and 5
being “Very Likely” to disclose) their likelihood of disclosure for a given question. We found that
20

PRELIMINARY DRAFT
the objective differences in privacy levels in the Survey A had a significant effect on participants’
predicted behavior. Participants provided the low privacy level predicted being significantly less
likely (βLow= -.67, p=<.001) to disclose personal information (Table 2, Column 1). Moreover, we
find consistent results (βLow= -.65, p<.001) when including controls for question type (descriptive
vs. ethical) and participants’ age and gender (Table 2, Column 2). Broadly, these results provide
strong support for the hypothesis that objective risk will impact consumer privacy choice (H1
supported).
[Table 1: Experiment 1 Summary Results]
Survey A
Survey B
CONDITIONS
High
Protection
Low
Protection
p-value Increasing Decreasing p-value
Privacy
Concern
2.39 3.87 p < .001 2.76 3.29 p < .01
Protection
Satisfaction
3.36 1.56 p <.001 2.86 2.41 p < .01
Harm
Perception
2.86 4.02 p <.001 3.37 3.68 p = .04
For the second survey (Survey B), which had an objectively identical medium privacy level
(Figure 3c) for both conditions, we found that participants in the increasing protection condition
reported being significantly more satisfied with the protections provided (MInc = 2.86, MDec= 2.41),
t (219) = 2.97, p < .01, d = 0.40, less concerned about privacy (MInc = 2.76, MDec= 3.29), t (219) =
-3.48, p < .01, d = 0.47, and less concerned that their responses may be used in way that may
harm them (MInc = 3.37, MDec= 3.68), t (219) = -2.04, p = .04, d = 0.28. However, the relative
change in privacy protection in Survey B did not have a significant effect on participants’
predicted disclosure behavior. Specifically, we found that increasing privacy protection did not
have a significant effect (βIncreasing= .09, p=.451) on overall predicted disclosure levels (Table 2,
Column 3). This result is robust (βIncreasing= .11, p=.363) to including controls for question type
and participant age and gender (Table 2, Column 4). In this hypothetical disclosures setting, our
results suggest a lack of support for the hypothesis that relative perception of privacy protection
will impact behavior (H2 not supported).
21

PRELIMINARY DRAFT
[Table 2: Experiment One Regression Results]
(1)
(2)
(3)
(4)
VARIABLES
Admit
Admit
Admit
Admit
Low Protection
-0.669**
-0.650**
(0.120)
(0.118)
Increasing
0.0925
0.109
(0.123)
(0.120)
Descriptive
-0.494**
-0.565**
(0.0607)
(0.0601)
Age
-0.0132*
-0.0100
(0.00651)
(0.00680)
Gender
0.130
0.196
(0.124)
(0.129)
Constant
3.631**
4.173**
3.328**
3.772**
(0.0701)
(0.229)
(0.0784)
(0.249)
Observations
2,210
2,210
2,210
2,210
Number of id
221
221
221
221
Robust standard errors in parentheses; ** p<0.01, * p<0.05, + p<0.1
4.4. Discussion
The results of Experiment 1 suggest that differences in both objective and relative risk have some
effect on participant perceptions of protection in the study, but only objective changes in risk
influenced predicted levels of self-disclosure decisions (H1 Supported). In contrast, we don’t find
differences in predicted levels of self-disclosure after changes in relative risk (H2 Not Supported).
Experiment 1 focused on hypothetical elicitation of privacy choices, which, as discussed
previously, may be activate distinct choice processes from actual behavior. While this study only
captured hypothetical choice, it provided some initial evidence consistent with our conjecture that
the impact of normative factors may be pronounced in hypothetical settings while the impact of
behavioral factors may be diminished in hypothetical settings. The evidence provided by
Experiment 1 in this regard is of course limited, as we cannot determine whether impacts of
objective and relative risk will change or stay constant when shifting to actual self-disclosures.
Moreover, Experiment 1 used a graphical representation of privacy protection levels, including a
key that alerted participants to riskier uses of their personal information. However, privacy
22

PRELIMINARY DRAFT
protections online are often communicated in text based notices, where changes in protection may
not be as salient. Finally, the design of Experiment 1 did not allow us to identify the distinct
effect of relative increases and decreases in the privacy level. For example, it may be the case that
our results were purely driven by decreases in the privacy levels, and that increases in the privacy
level did not have an impact. We address these issues in Experiment 2.
5. Experiment 2
Experiment 1 focused on how objective and relative privacy protection impact participants’
hypothetical disclosures. In Experiment 2, we examined the role of objective and relative changes
in privacy protection on actual disclosures. In addition to evaluating H1 and H2 in actual choice
settings, Experiment 2 complements Experiment 1 in other ways. First, we use text based privacy
notices. Second, the experimental design allows us to evaluate the unique impact of relative
increase and decreases in privacy protection as well as objective changes. Specifically, we asked
participants in Experiment 2 to take part in two separate surveys about their personal behaviors.
Similarly to Experiment 1, each survey provided different stated levels of privacy protections to
participants. Between participants, we kept the objective level of privacy offered by the surveys at
the same level (and used as a simple text-based privacy notice), and manipulated whether
participants experienced a relative increase or decrease in privacy protection levels. We examined
the effects of such changes on actual disclosure behavior. By including accompanying control
conditions in which protections did not change, we were able to isolate the specific impact of
increases and decreases in privacy levels.7
7 Early analysis of Experiment 2 was included in a short paper focused on the effect of privacy notices, published as
part of the ACM proceedings from the 2013 Symposium on Usable Privacy and Security (SOUPS).
23

PRELIMINARY DRAFT
5.1. Participants
Four hundred and fifteen participants from Amazon Mechanical Turk (51.61% females, Mage =
31.27, SDage = 10.72) completed the study online. The experiment was advertised to participants
as two, ostensibly unrelated, surveys on (un)ethical behavior.8
5.2. Design and Procedure
Experiment 2 consists of a 2 (high vs. low protection in the first survey) X 2 (high vs. low
protection in the second survey) between-subject design. Thus, the study consisted of four groups
of participants whose privacy protection either increased from the first to the second survey (low
protection to high protection: LH), decreased (high protection to low protection: HL) or stayed
the same (low to low protection: LL or high to high protection: HH).
In the first survey, participants were asked demographic questions, including email
address as a mandatory question. Participants were told that we would check the validity of their
email addresses prior to approving payment for the study (we did not actually store email
addresses). Then, participants were provided with a privacy notice concerning the way their
answers to the questionnaire would be stored. To more closely model privacy protections in real
world contexts, we presented participants with text notices (as opposed to the graphical notices
presented in Experiment 1) focusing on whether their responses would be identified or
anonymous (see Appendix A.3 for full text of notices provided). Specifically, participants offered
“low” protections were informed that their answers would be linked to their email addresses.
Conversely, those offered “high” protection were informed that their answers would not be linked
to their email addresses.9 Participants were then presented with six questions relating to ethically
questionable activities (see Appendix A.5 for full set of questions). The questions included a
subset of the questions that had been judged in Acquisti et al., (2012) as highly intrusive (e.g.
“Have you ever had sexual desires for a minor?”).
8 Participants in Experiment 1 were not able to participate in Experiment 2.
9 In Experiment 1, participants commented in exist questions that they were most concerned about the propensity of a
study to require them to provide email addresses and to link their responses via their email address.
24

PRELIMINARY DRAFT
Thereafter, participants were asked to complete an additional survey that followed the same
structure as the first survey, but had a different visual design, consistent with the idea that
participants were asked to participate in two separate studies (see Appendix A.4). Also,
participants were provided two separate confirmation codes to submit in order to receive payment
for completing both surveys.10 In the second survey, participants were again asked for their
emails and demographic information. Then, they were given a privacy notice concerning the way
their answers to the questions would be stored. As in the first survey, the privacy notice signaled
either high protection (not linking responses to emails) or low protection (responses linked to
emails). Then, participants were presented with six questions, different from those presented to
them in the first survey about other ethically questionable behaviors (see Appendix A.5). Lastly,
participants responded to some exit questions that gauged both their perception of whether
privacy protections changed in each study (e.g. whether the increased, decreased, or stayed the
same, depending on the condition) and their recall of privacy notices in both surveys.
5.3. Results
We found that our manipulations of high and low protection elicited the hypothesized effect, with
participants in the high protection conditions reporting significantly higher beliefs that their
responses would be linked back to them (MHigh=.79, MLow =.14), t (411) = 18.81, p < .001, d =
1.86), relative to participants in the low protection condition. We first evaluated the disclosure
rates of participants in the first survey. We found that participants were statistically more likely to
disclose (βHigh =.05 p = .04) when they were provided with high protection in the first survey
(Table 3, Column 1). However, our results were only marginally significant (βHigh =.04 p = .07)
with the inclusion of controls for question intrusiveness, the survey’s visual design, and
participant demographics (Table 3, Column 2). In the first round, we find initial evidence in
support of the hypothesis that objective risk will impact participant behavior.
10 99.50% of the participants that completed the exit questions indicated they had participated in more than one study
and 96.59% of participants indicated that there were differences between the two studies.
25

PRELIMINARY DRAFT
We then evaluated disclosure behavior in the second survey of our experiment, where
participants were either presented an increasing, decreasing, or identical protection compared to
the first survey. A few participants (11%) were unable to accurately recall the privacy notices
(whether protections had increased, decreased, or stayed the same from the first to the second
survey) and were excluded from our second survey analysis, leaving 368 usable responses.11 First,
we compared participants that had High Protection in both surveys to participants that had Low
Protection in both surveys. For the analysis in the second round, we control for the possible
impact of disclosing more in the first survey on second survey disclosures using Survey1Sharing,
which ranges from a value of zero (for participants that did not admit to any of the behaviors in
Survey 1), to a value of six (for participants admitting to all behaviors in Survey 1). In contrast to
our results for the first survey, we found no effect of high protection vs. low protection on
disclosure (βHigh = -.003, p = .9) in the second survey (Table 3, Column 3). This result is robust
(βHigh = -.0001, p = .99) to including controls for question intrusiveness, the survey’s visual design,
and participant demographics (Table 3, Column 4). This suggests that participant sensitivity to
different levels of privacy protection diminished over a fairly short period of time (i.e. between
the time taken to complete the first and second survey); this results in mixed support for H1.
Second, we evaluated the impact of changing protection on disclosure relative to
conditions in which did not perceive an increase or decrease (participants were provided
objectively equivalent privacy notices). We found an increase in the propensity to disclose
(βIncreasing = .06, p =.04) for participants that perceived an increase in protection relative to those
whose protections stayed constant. This result is robust to including controls for question
intrusiveness, the survey’s visual design, and participant demographics (Table 3, Columns 5-6).
Conversely, we found a decrease in the overall propensity to disclose (βDecreasing =-.08, p =.006) for
participants that perceived a decrease in protection relative to those whose protections stayed
constant (Table 3, Column 7). Again, this result was robust to including controls for question
11 The results remain similar when including these participants.
26

PRELIMINARY DRAFT
intrusiveness, the survey’s visual design, and participant demographics (Table 3, Column 8).
These results suggest that participants’ relative perceptions of privacy protection had a consistent
impact on disclosure behavior (H2 Supported).
[Table 3: Experiment 2 Regression Results]
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
VARIABLES
Admit
Admit
Admit
Admit
Admit
Admit
Admit
Admit
High
Protection
0.0499*
(0.0240) 0.0423+
(0.0231) -0.00336
(0.0278) 0.0001
(0.0278)
Increasing
0.0605*
0.0604*
(0.0295)
(0.0292)
Decreasing
-0.075**
(0.0269)
-0.071**
(0.0271)
Intrusive
0.0758*
*
(0.0178)
-0.111**
(0.0271) -0.113**
(0.0259)
-0.086**
(0.0288)
Age -
0.005**
(0.0009)
0.00139
(0.0016) 0.00324*
(0.0014) 0.000579
(0.0016)
Male
0.0493*
0.0512+
0.0633*
0.0483
(0.0232)
(0.0301)
(0.0303)
(0.0307)
Survey Design 0.0379
(0.0231) -0.0120
(0.0300) 0.00729
(0.0305) -0.0234
(0.0276)
Survey 1
Sharing
0.105**
(0.0093) 0.105**
(0.0099) 0.0950**
(0.0102) 0.0973**
(0.0105) 0.110**
(0.0099) 0.109**
(0.0103)
Constant
0.444**
0.525**
0.0149
0.0206
0.0408
-0.0273
0.000929
0.0265
(0.0176)
(0.0438)
(0.0273)
(0.0693)
(0.0302)
(0.0654)
(0.0278)
(0.0700)
Observations
2,490
2,454
1,164
1,140
1,158
1,140
1,050
1,032
Number of id
415
409
194
190
193
190
175
172
Robust standard errors in parentheses; ** p<0.01, * p<0.05, + p<0.1
5.4. Discussion
As in Experiment 1, the results of Experiment 2 suggest that both objective and relative changes
in privacy protection can impact participants’ self-disclosure behavior - even in actual choice
settings. However, compared with the results we obtained in Experiment 1 (where we used a
hypothetical choice context) we observe in Experiment 2 a reversal in the prominence of effects
of relative versus objective changes in protection. Specifically, we found that objective changes
in the levels of protection had only a weak initial effect on disclosure (only significant at the 10%
27

PRELIMINARY DRAFT
level with controls) and no effect of objective differences in privacy protection on disclosure
behavior in second survey (H1 Mixed Support). In contrast, relative changes in privacy protection
had a significant impact on disclosure behavior (H2 Supported). These findings suggest that
participants’ propensity to disclose personal information can be influenced by seemingly minor
factors such as the relative, instead of the absolute, value of privacy protection; whereas, the
impact of objective changes in protection on actual behavior may be more limited. Combining
results from Experiment 1 with those from Experiment 2 provides initial evidence that objective
and behavioral factors may have differential effects in hypothetical versus actual choice settings.
Specifically, the impact of objective changes in protection seem to diminish as we shift to actual
choice contexts while the impacts of relative changes in protection seem to become more
pronounced as we shift to actual choice settings.
While comparison of effects between Experiment 1 and 2 is an informative data point for
the manuscript, the conclusions we can draw from this comparison may be limited. While many
variables were kept constant across the two experiments (such as the sampling population, the
context of the experiments, and so forth), a number of differences across experiments could limit
our ability to make direct comparisons across experiments. For example, the studies used
different privacy notices with different types of protection and were conducted at different times.
Also, both studies were conducted on Amazon Mechanical Turk, introducing some concern that
the effects may only persist with that population. These concerns are addressed in Experiment 3.
6. Experiment 3
The goal of Experiment 3 was to confirm that both behavioral and normative factors can be
predictors of privacy decision making, while also more directly comparing their respective
impacts across hypothetical and actual choice settings in the same experimental setting. Using the
same experimental setting allows us to evaluate the effect of both factors in hypothetical and
actual choice contexts while significantly reducing the likelihood of other potential explanations
of our effect being due to differences between the two prior experiments. Furthermore, we
28

PRELIMINARY DRAFT
recruited a new cohort of participants from a new online recruiting platform (Prolific Academic),
to test the robustness of the effects on populations other than AMT.
We recruited participants to take part in what was advertised as two separate studies. The
first study served to set the stage in terms of privacy protection, and offered participants either a
high or low level of privacy protection (as in Experiment 2), but did not involve any measure of
self-disclosure. The second study, as in the previous experiments, offered either high or low
levels of privacy protection. The experiment therefore consisted in a four-conditions between-
subject design. Along one dimension, we manipulated both actual levels of privacy protections
and the relative perceptions of those protections (compared to the level of protection participants
received in the previous study). This enabled us to examine how self-disclosure was affected by
both actual and relative changes in privacy protections. Along the other dimension, we
manipulated whether participants were asked to actually disclose personal information (actual
choice setting), or whether they were asked to self-report how likely they would be to answer the
self-disclosure questions (hypothetically choice setting). With this design, we can test the
robustness of the finding that both objective and behavioral factors influence behavior, and we
can directly evaluate the conjecture that objective and relative changes in privacy protection may
have differential effects across hypothetical and actual choice settings.
6.1. Participants
We recruited 739 participants (51.7% males, Mage = 29.67, SD = 10.1) from Prolific Academic
(www.prolificacademic.ac.uk) who completed the study for 1.5 GBP.
6.2. Design and procedure
Participants were invited to complete two studies (similarly to Experiment 2) which were
administered one after the other but had unrelated contexts. In the first study, participants were
first given instructions regarding the privacy protections afforded by the study. Participants
received information about the settings of the study that signaled either a high or a low protection
level for the answers provided in the study, as in Experiment 1 (see Figures 1a and 1b,
29

PRELIMINARY DRAFT
respectively). Participants were then asked to rate how high or low they considered the protection
offered in the study (on a 5-point Likert scale). In the low protection condition, participants were
required to provide their email addresses, to increase the perception that responses may be linked
to their identities. All participants then engaged in a filler task that separated the first survey from
the second (the filler task consisted in viewing a a 5minute video clip and answering open-ended
questions about it).
The second study used a different look and feel (type of font, background color, etc.) than
the first study. Participants were first given the information about the level of privacy protections
provided in the new study, which was either high or low (see Figures 1a and 1b) and were asked
to rate their view of the level of protection from “very low” to “very high.” Participants in the low
protection condition were again required to provide their email addresses. Next, participants were
randomly assigned to either the “actual” or “hypothetical” disclosure conditions. In the actual
disclosure conditions, participants were asked to answer five personal and sensitive questions
used in Acquisti et al., (2012)—see Appendix A.5. Participants were asked to provide their
answers on a 4-point scale that ranged from “never” to “many times”, with the fifth option being
“I prefer not to say.” In the hypothetical disclosure condition, the questions remained the same,
but participants were asked to imagine taking part of a study with a certain level of protection
afforded to the answers.. Similar to Experiment 1 (and again, in-line with measures used in extant
literature), hypothetical behavior participants were told that they would be presented the a set of
questions relating to (un)ethical behaviors and asked to indicate their likelihood to admit to such
behaviors using a five point scale ranging from “definitely no” to “definitely yes.” Finally,
participants indicated their age and gender to complete the study.
6.3. Results
In the first study, participants in the high protection condition rated it as offering higher privacy
protection (M=3.95 vs. 2.87, SD = 0.93, 1.19, t (737) = 13.75, p < .001). Consistent results were
found for the ratings of privacy protections in the second study (M=4.00 vs. 2.57, SD = 0.81, 1.28,
30

PRELIMINARY DRAFT
t (737) = 18.32, p < .001). We thus conclude that our manipulation worked as expected, and turn
to examine the effects on actual and hypothetical levels of self-disclosure. We first focus our
analysis on those in the hypothetical settings where we considered participants as admitting to the
behavior if they responded either with “strongly agree” or “agree” to the question of whether they
would admit to a particular behavior. We find statistically significant differences in their
hypothetical admission rates between those with objectively different (high vs. low) levels of
protection (63% vs. 53%, t(188)=2.01, p=.046). Conversely, we do not find any significant
differences in hypothetical admissions when protections are held objectively constant but
decrease in relative terms (53% vs. 50%, t(180)=.669, p=.50) or increase in relative terms (63%
in both conditions, t(181)=.006, p=.99). These results are robust to alternate measurements for
hypothetical admission, including a continuous measure (i.e. 1 to 5 on the Likert scale) and
considering those that report being uncertain (neither agree nor disagree) as also admitting to the
behavior (see Appendix A6). These results are confirmed in a random effects regression (Table 4).
We find that objective difference in protection (High Protection) had a significant effect in the
hypothetical context (Column 1) while the relative changes have no effect (Column 2 and 3).
[Table 4: Experiment 3 Hypothetical Choice Results]
Hypothetical Admissions
VARIABLES
(1)
(2)
(3)
High Protection
0.0878*
(0.0441)
Decreasing
-0.0305
(0.0459)
Increasing
0.00185
(0.0433)
Age
-0.00154
-0.000852
-0.000954
(0.00251)
(0.00231)
(0.00276)
Male
-0.105*
-0.107*
-0.0441
(0.0453)
(0.0467)
(0.0441)
Constant
0.737**
0.720**
0.718**
(0.0915)
(0.0910)
(0.0967)
Observations
950
910
915
Number of id
190
182
183
Robust standard errors in parentheses; ** p<0.01, * p<0.05, + p<0.1
31

PRELIMINARY DRAFT
Next, we consider participants in the actual disclosure conditions where participants were
shown the same exact privacy protections and asked the same questions as their counter-parts in
the hypothetical disclosure conditions. For these participants, we considered an admission as any
response to our questions that indicated that the participant engaged in a particular behavior at
least once (similar affirmative admit rates were used in prior work— John, Acquisti, and
Loewenstein, 2011). We find that objective differences (high vs. low) in protection, unlike the
hypothetical context, did not have a significant effect on disclosure behavior (65% vs. 59%,
t(178) =1.47, p=.15). Conversely, and again in contrast to the hypothetical conditions, we find
that those who perceived a relative decrease in protection disclosed significantly less than those
that did not perceive a change (49% vs. 59%, t(161)=-2.09, p=.038). Recall that these participants
were provided objectively identical protections between these conditions. Finally, similar to the
hypothetical context, we identify that the relative increases in protection did not have a significant
effect on disclosure (64% vs. 65%, t(201)=-0.35, p=.73), suggesting that the effect increases in
privacy protection identified in Experiment 2 may not be robust.
[Table 5: Experiment 3 Actual Choice Results]
Actual Admissions
VARIABLES
(1)
(2)
(3)
High Protection
0.0552
(0.0410)
Decreasing
-0.108*
(0.0476)
Increasing
-0.0126
(0.0354)
Age
0.00143
0.00248
-1.24e-05
(0.00277)
(0.00232)
(0.00192)
Male
-0.0296
-0.0826+
-0.0321
(0.0408)
(0.0480)
(0.0366)
Constant
0.594**
0.646**
0.695**
(0.0959)
(0.0987)
(0.0694)
Observations
895
810
1,010
Number of id
179
162
202
Robust standard errors in parentheses; ** p<0.01, * p<0.05, + p<0.1
32

PRELIMINARY DRAFT
These results are confirmed in our random effects regression (Table 5). We find that the objective
difference in protection (High Protection) did not have a significant effect in actual behavior
(Column 1) while relative changes (specifically, a relative decrease in protection) have a strong
observable effect (Column 2).
6.4. Discussion
In Experiment 3, we controlled for a number of factors that varied across Experiment 1 and 2, and
continued to find that H1 is supported in hypothetical choice but not supported in actual choice
settings while H2 is not supported in hypothetical settings but supported in actual choice settings.
Thus, the results bolster the findings from the previous experiments for simultaneous effects of
normative (H1) and behavioral (H2) on privacy decision making. They also provide more robust
evidence supporting the conjecture of a diminished effect of normative factors in hypothetical
relative to actual choice contexts, and a pronounced impact of behavioral factors in hypothetical
relative to actual choice contexts. Comparing the Hedge’s g (a bias corrected and normalized
measure of effect size across experiments—Hedges, 1981) for identical treatments in hypothetical
vs. actual choice settings supports the insight: Identical relative decreases in privacy protection in
hypothetical choice contexts had a treatment effect of only .09, relative to a .33 treatment effect in
analogous actual choice settings. Conversely, identical objective decreases in protection had a
treatment effect of .32 in hypothetical choice settings, but a diminished treatment effect of .21 in
actual choice settings. Summarizing results across our three experiments highlights our main
findings (see Table 6). We consistently find evidence that both normative and behavioral factors
can simultaneously influence consumer perceptions of privacy risk and actual privacy choices,
but that these effects may emerge differentially across hypothetical vs. actual choice contexts.
[Table 6: Overview of Results]
Experiment 1 Experiment 2 Experiment 3
Hypothetical Choice Actual Choice Hypothetical Choice Actual Choice
33

PRELIMINARY DRAFT
7. Discussion and Conclusions
Our work builds on the IS literature on consumer privacy decision making and the behavioral
economics literature on reference dependence and relative judgment.Leaning on proposed models
of reference-dependent utility, which account for both the utility from absolute levels of
consumption and deviations from a reference point, we present some evidence suggesting that, in
the context of privacy decision making, relative changes may have an increasingly important
impact on decision making, particularly in actual choice contexts, relative to absolute or objective
level of protection provided.
Our results have implications for both theories of consumer privacy behavior and policy
efforts. More generally, our results suggest that the behavioral factors we evaluate may be
underappreciated by consumers when anticipating their privacy concerns and behavior in
hypothetical situations, but may actually be more influential on, and more consistent drivers of
behavior in contexts that involve actual privacy behavior. These findings are consistent with the
broader psychology and behavioral economics literature (e.g., Lowenstein and Adler, 1995;
Gilbert and Ebert, 2002), in that they imply that people may overestimate the impact of normative
factors on their hypothetical behavior while underestimating the sometimes powerful impact of
decision biases on actual decision making. Our results are consistent with the growing literature
on how privacy decision making may be particularly susceptible to deviations from economically
rational models of decision making, by not only presenting additional evidence of these
deviations but starting to identify the conditions under which these effects are most likely to
materialize. Finally, our results start to reconcile some of the dissonance in the privacy literature
and help explain early results from the privacy paradox literature (e.g., Spiekermann et al., 2001)
H1: Objective
Privacy
Protection
Supported Mixed Support Supported Not Supported
H2: Relative
Privacy
Protection
Not Supported Supported Not Supported Supported
34

PRELIMINARY DRAFT
by substantiating a critical link in how limitations in consumer (ir)rationality may be driving the
observed dissonance between consumer concerns and hypothetical behavior and actual decision
making. We clarify however that pragmatic considerations can lead to the choice of using
hypothetical vs. actual choice (e.g. whether it is feasible to observe actual behavior in ones’
context of interest) so our focus is not to suggest that one approach is always preferred to the
other, but simply to suggest that this distinction could be relevant to the tension between
normative and behavioral models of privacy decision making. In fact, our results substantiate that
both perspectives are predictive of consumer behavior across both choice settings but that they
may simply emerge differentially between them.
Our work has some important limitations however. First, we evaluate specific
manipulations of the normative and behavioral perspective, which introduces the question of
whether these effects would extend to other manipulations of normative vs. behavioral
perspectives. We alleviate some of these concerns by using experimental treatments that vary in
how they manipulate these factors (e.g. visual vs. text notice) and focus on manipulations that are
well rooted in their respective literatures—Prospect Theory, for example, is a seminal theory in
the behavioral economics literature and informs numerous behavioral phenomena. More so, we
do not claim that the impact of normative factors are always stronger predictors of behavior in a
given contexts. Clearly this claim is problematic since one can arbitrarily alter the strength of any
normative or behavioral manipulation such that one dominates in a given setting. Rather, we
suggest that the effect of similar or identical manipulation of normative or behavioral factors
differ as they are assessed in hypothetical vs. actual choice contexts.
Another limitation consists in the experimental nature of the work, with constraints in
terms of external validity, due to context and sample selection. In some sense, however, this
tension is part of our empirical testing, since we evaluate choice across hypothetical and the more
realistic actual choice setting. In any case, we leverage diverse online samples that provide more
diverse participant pools to address residual concerns in this regard.
35

PRELIMINARY DRAFT
Disentangling these diverging perspectives has implications beyond the academic
discourse on consumer privacy behaviors. Hoofnagle and Urban (2014) argue that prominent
policy intended to protect consumer privacy are predicated on the notion that consumers are able
to consistently and predictably react to changes in normative factors within privacy contexts (e.g.
the objective benefits and costs of data allowances and disclosures). Such protections are
provided across various contexts and through distinct mechanisms. In some cases, protections are
mandated via regulation that covers some subset of personal information such as HIPAA in the
U.S. or a more general swatch of consumer personal information such as the EU Data Directive in
the European Union. Beyond regulation, protections can be provided to consumer personal
information via self-regulatory mechanisms including company privacy notices detailing how use
of consumer personal information will be limited and the also choice mechanisms that provide
consumers the option to choose their desired degree of privacy protection.
Moreover, the evidence in support of reference-dependent privacy decision making
presented in this manuscript has, in itself, considerable implications for firms and policy makers.
If consumers’ judgments of privacy protections in actual choice contexts are relative rather than
absolute, market choices might not necessarily capture or reflect “objective” privacy preferences.
For example, if privacy protection is increased from a very low (absolute) level of protection,
consumers might consider that as a gain, even though the resulting privacy protection might still
be low; and consumers might be more inclined to choose privacy protections that seem more
protective (in relative terms) but actually may not be. Conversely, if the level of privacy
protection is decreased from a high (absolute) level of protection, consumers might consider that
as a loss, and be less willing to use the offered service or disclose personal information, even
though the actual level of privacy has remained quite high. These results suggest that policy
maker goals of consumer privacy protection through transparency and control mechanisms may
not be realized if firms choose to highlight gains and downplay losses to privacy protection over
time and among their competitors. However, these results could also present an opportunity for
36

PRELIMINARY DRAFT
policy makers to bring attention to high relevance privacy contexts by mandating that firms
clearly highlight changes in data practices over time, including decreases in protection. This
approach may be particularly effective given that, over time, relative changes in protection in our
experiments impacted privacy decision more than the objective risk that participants faced.
The implications for firms seeking value from innovations rooted in the collection of
consumer personal information is less clear. Firms that benefit from increased disclosure and
allowances by consumers may find some short-term value in presenting notices and choices as
relatively “more protective.” However, if actual data practices violate consumer expectations for
privacy, troublesome and costly privacy incidents may persist, leading to less disclosure and trust
by consumers and decreased use in the long term. Moreover, if firms highlight the privacy
protective nature of their services relative to their competitors, consumers may have an elevated
expectation for privacy which may be inconsistent with actual firm data practices. The
increasingly dynamic nature of data practices over time and the heterogeneity of data practices
between firms suggests that the relative perception of privacy protection will continue to be an
important predictor of consumer privacy decision making, and will thus have significant
implications for the effectiveness of tools mandated by policy makers and the mechanisms by
which firms solicit privacy-relevant choices.
References
1. Acquisti, A., & Grossklags, J. (2005). Privacy and rationality in individual decision
making. IEEE Security & Privacy, 2, 24-30.
2. Acquisti, A., Brandimarte, L., & Loewenstein, G. (2015). Privacy and human behavior in the
age of information. Science, 347(6221), 509-514.
3. Acquisti, A., John, L. K., & Loewenstein, G. (2012). The impact of relative standards on the
propensity to disclose. Journal of Marketing Research,49(2), 160-174.
4. Ajzen, I., Brown, T. C., & Carvajal, F. (2004). Explaining the discrepancy between intentions
and actions: The case of hypothetical bias in contingent valuation. Personality and social
psychology bulletin, 30(9), 1108-1121.
5. Ansari, A., & Mela, C. F. (2003). E-customization. Journal of Marketing Research, 40(2),
131-145.
6. Bartling, B., Brandes, L., & Schunk, D. (2015). Expectations as reference points: Field
evidence from professional soccer. Management Science.
7. Bélanger, F., & Crossler, R. E. (2011). Privacy in the digital age: a review of information
privacy research in information systems. Mis Quarterly, 35(4), 1017-1042.
37

PRELIMINARY DRAFT
8. Bhatia, S. (2013). Associations and the accumulation of preference.Psychological
review, 120(3), 522.
9. Brandimarte, L., Acquisti, A., & Loewenstein, G. (2013). Misplaced Confidences Privacy and
the Control Paradox. Social Psychological and Personality Science, 4(3), 340-347.
10. Buhrmester, M., Kwang, T., & Gosling, S. D. (2011). Amazon's Mechanical Turk a new
source of inexpensive, yet high-quality, data?. Perspectives on Psychological Science, 6(1),
3-5.
11. Camerer, C. F., Loewenstein, G., & Rabin, M. (Eds.). (2011). Advances in behavioral
economics. Princeton University Press.
12. Camerer, C. F., Loewenstein, G., & Rabin, M.. (2011). Advances in behavioral economics.
Princeton University Press.
13. Culnan, M. J., & Armstrong, P. K. (1999). Information privacy concerns, procedural fairness,
and impersonal trust: An empirical investigation.Organization Science, 10(1), 104-115.
14. Dinev, T., & Hart, P. (2006). An extended privacy calculus model for e-commerce
transactions. Information Systems Research, 17(1), 61-80.
15. Egelman, S., Felt, A. P., & Wagner, D. (2013). Choice architecture and smartphone privacy:
There’sa price for that. In The economics of information security and privacy (pp. 211-236).
Springer Berlin Heidelberg.
16. Ellison, N. B., Steinfield, C., & Lampe, C. (2007). The benefits of Facebook “friends:” Social
capital and college students’ use of online social network sites. Journal of Computer‐
Mediated Communication, 12(4), 1143-1168.
17. Featherman, M. S., & Pavlou, P. A. (2003). Predicting e-services adoption: a perceived risk
facets perspective. International journal of human-computer studies, 59(4), 451-474.
18. Fehr, D., Hakimov, R., & Kübler, D. (2015). The willingness to pay–willingness to accept
gap: A failed replication of Plott and Zeiler. European Economic Review, 78, 120-128.
19. FeldmanHall, O., Mobbs, D., Evans, D., Hiscox, L., Navrady, L., & Dalgleish, T. (2012).
What we say and what we do: the relationship between real and hypothetical moral
choices. Cognition, 123(3), 434-441.
20. Gilbert, D. T., & Ebert, J. E. (2002). Decisions and revisions: the affective forecasting of
changeable outcomes. Journal of personality and social psychology, 82(4), 503.
21. Goes, P. B. (2013). Editor's comments: information systems research and behavioral
economics. MIS quarterly, 37(3), 3-8.
22. Harper, V. B., & Harper, E. J. (2006). Understanding student self-disclosure typology
through blogging. The Qualitative Report, 11(2), 251-261.
23. Hedges, L. V. (1981). Distribution theory for Glass's estimator of effect size and related
estimators. Journal of Educational and Behavioral Statistics,6(2), 107-128.
24. Herrmann, P. N., Kundisch, D. O., & Rahman, M. S. (2014). Beating Irrationality: Does
Delegating to IT Alleviate the Sunk Cost Effect?.Management Science, 61(4), 831-850.
25. Ho, T. H., Lim, N., & Camerer, C. F. (2006). Modeling the psychology of consumer and firm
behavior with behavioral economics. Journal of marketing Research, 43(3), 307-331.
26. Hoofnagle, C. J., & Urban, J. M. (2014). Alan Westin's Privacy Homo Economicus. Wake
Forest Law Review, 47, 102-316.
27. Hui, K. L., Teo, H. H., & Lee, S. Y. T. (2007). The value of privacy assurance: an
exploratory field experiment. Mis Quarterly, 19-33.
28. Isoni, A., Loomes, G., & Sugden, R. (2011). The willingness to pay—willingness to accept
gap, the “endowment effect,” subject misconceptions, and experimental procedures for
eliciting valuations: Comment. The American Economic Review, 101(2), 991-1011.
29. Jensen, C., Potts, C., & Jensen, C. (2005). Privacy practices of Internet users: self-reports
versus observed behavior. International Journal of Human-Computer Studies, 63(1), 203-227.
30. John, L., Acquisti, A., & Loewenstein, G. (2011). Strangers on a plane: Context dependent
willingness to divulge personal information. Journal of Consumer Research, 37(5), 858-873.
38

PRELIMINARY DRAFT
31. Johnson, E. J., Bellman, S., & Lohse, G. L. (2002). Defaults, framing and privacy: Why
opting in-opting out1. Marketing Letters, 13(1), 5-15.
32. Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under
risk. Econometrica, 263-291.
33. Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1991). Anomalies: The endowment effect,
loss aversion, and status quo bias. The journal of economic perspectives, 193-206.
34. Kim, D. J., Ferrin, D. L., & Rao, H. R. (2008). A trust-based consumer decision-making
model in electronic commerce: The role of trust, perceived risk, and their antecedents.
Decision support systems, 44(2), 544-564.
35. Klopfer, P.H. & Rubenstein, D.I. (1977). “The Concept Privacy and its Biological Basis,”
Journal of Social Issues, 33(3), 52-65.
36. Knetsch, J. L., Tang, F. F., & Thaler, R. H. (2001). The endowment effect and repeated market
trials: Is the Vickrey auction demand revealing?. Experimental Economics, 4(3), 257-269.
37. Kőszegi, B., & Rabin, M. (2006). A model of reference-dependent preferences.The Quarterly
Journal of Economics, 1133-1165.
38. Leontiadis, I., Efstratiou, C., Picone, M., & Mascolo, C. (2012). Don't kill my ads!: balancing
privacy in an ad-supported mobile application market. In Proceedings of the Twelfth
Workshop on Mobile Computing Systems & Applications (p. 2). ACM.
39. Liberman, V., Samuels, S. M., & Ross, L. (2004). The name of the game: Predictive power of
reputations versus situational labels in determining prisoner’s dilemma game
moves. Personality and social psychology bulletin,30(9), 1175-1185.
40. Loewenstein, G. (2000). Emotions in economic theory and economic behavior. The American
Economic Review, 90(2), 426-432
41. Loewenstein, G., & Adler, D. (1995). A bias in the prediction of tastes. The Economic
Journal, 929-937.
42. Malhotra, N. K., Kim, S. S., & Agarwal, J. (2004). Internet users' information privacy concerns
(IUIPC): the construct, the scale, and a causal model. Information Systems Research, 15(4),
336-355.
43. Marthews, A., & Tucker, C. (2014). Government surveillance and internet search behavior.
Available at SSRN 2412564.
44. Milberg, S. J., Burke, S. J., Smith, H. J., & Kallman, E. A. (1995). Values, personal
information privacy, and regulatory approaches. Communications of the ACM, 38(12), 65-74.
45. Milne, G. R., & Gordon, M. E. (1993). Direct mail privacy-efficiency trade-offs within an
implied social contract framework. Journal of Public Policy & Marketing, 206-215.
46. Miyazaki, A. D., & Krishnamurthy, S. (2002). Internet seals of approval: Effects on online
privacy policies and consumer perceptions. Journal of Consumer Affairs, 36(1), 28-49.
47. Moon, Y. (2000). Intimate exchanges: Using computers to elicit self-disclosure from
consumers. Journal of consumer research, 26(4), 323-339.
48. Norberg, P. A., Horne, D. R., & Horne, D. A. (2007). The privacy paradox: Personal
information disclosure intentions versus behaviors. Journal of Consumer Affairs, 41(1), 100-
126.
49. Novemsky, N., & Kahneman, D. (2005). The boundaries of loss aversion.Journal of
Marketing Research, 42(2), 119-128.
50. O'Donoghue, T., & Rabin, M. (2000). The economics of immediate gratification.Journal of
Behavioral Decision Making, 13(2), 233-250.
51. Smith, H. J., Dinev, T., & Xu, H. (2011). Information privacy research: an interdisciplinary
review. MIS quarterly, 35(4), 989-1016.
52. Smith, H. J., Milberg, S. J., & Burke, S. J. (1996). Information Privacy: Measuring
Individuals' Concerns About Organizational Practices. MIS quarterly,20(2).
39

PRELIMINARY DRAFT
53. Spiekermann, S., Grossklags, J., & Berendt, B. (2001). E-privacy in 2nd generation E-
commerce: privacy preferences versus actual behavior. In Proceedings of the 3rd ACM
conference on Electronic Commerce (pp. 38-47). ACM.
54. Steelman, Z. R., Hammer, B. I., & Limayem, M. (2014). Data collection in the digital age:
innovative alterantives to student samples. Mis Quarterly, 38(2), 355-378.
55. Stutzman, F., Gross, R., & Acquisti, A. (2013). Silent listeners: The evolution of privacy and
disclosure on facebook. Journal of Privacy and Confidentiality,4(2), 2.
56. The Federal Trade Commission (FTC). (2012). Protecting consumer privacy in an era of rapid
change: recommendations for businesses and policy makers.
http://www.ftc.gov/os/2012/03/120326privacyreport.pdf.
57. Tourangeau, R. (2004). Survey research and societal change. Annu. Rev. Psychol., 55, 775-
801.
58. Tsai, J. Y., Egelman, S., Cranor, L., & Acquisti, A. (2011). The effect of online privacy
information on purchasing behavior: An experimental study. Information Systems
Research, 22(2), 254-268.
59. Vance, A., Elie-Dit-Cosaque, C., & Straub, D. W. (2008). Examining trust in information
technology artifacts: the effects of system quality and culture. Journal of Management
Information Systems, 24(4), 73-100.
60. Viswanathan, S., Kuruzovich, J., Gosain, S., & Agarwal, R. (2007). Online infomediaries and
price discrimination: Evidence from the automotive retailing sector. Journal of
Marketing, 71(3), 89-107.
61. Westin, A. F. (2000). Intrusions: Privacy tradeoffs in a free society. Public Perspective, 11(6),
8-11.
62. Xu, H., Teo, H. H., Tan, B. C., & Agarwal, R. (2009). The role of push-pull technology in
privacy calculus: the case of location-based services. Journal of Management Information
Systems, 26(3), 135-174.
63. Xu, H., Teo, H. H., Tan, B. C., & Agarwal, R. (2012). Research Note-Effects of Individual
Self-Protection, Industry Self-Regulation, and Government Regulation on Privacy Concerns:
A Study of Location-Based Services. Information Systems Research, 23(4), 1342-1363.
64. Zeger, S. L., & Liang, K. Y. (1986). Longitudinal data analysis for discrete and continuous
outcomes. Biometrics, 121-130.
40

PRELIMINARY DRAFT
APPENDICES
A.1 Experiment 1- Summary Questions
Measure
Description
Privacy Concern
I would be concerned about my privacy if I was a participant in this
upcoming study.
Protection Satisfaction
I am satisfied with the protections provided in this upcoming study.
Harm Perception
I would be concerned that my responses in this upcoming study could
be used to harm me.
A.2 Experiment 1- Disclosure Questions
Question
Text
Category
1
What is your annual income?
Descriptive
2
What is your sexual orientation?
Descriptive
3
What is your address?
Descriptive
4
What is your phone number?
Descriptive
5
What is your view on gay rights?
Descriptive
6
Have you every downloaded a pirated song?
Ethical
7
Have you ever flirted with someone other than your
partner or spouse?
Ethical
8
Have you ever used drugs of any kind (e.g. weed, heroin,
crack)?
Ethical
9
Have you ever looked at pornographic material?
Ethical
10
Have you ever made up a serious excuse, such as a grave
illness or death in the family, to get out of doing
something?
Ethical
A.3 Experiment 2 Privacy Notice
Privacy Notice
Notice Text
High Protection
The analysis for this study requires that your responses are stored using a
randomly assigned ID. All other information that could potentially be used to
identify you (email, zip code, etc.) will be stored separately from your
responses. As such, your responses to the following set of questions cannot be
directly linked back to you.
Low Protection
The analysis for this study requires that your responses are stored using your
email. As such, your responses to the following set of questions may be
directly linked back to you.
41

PRELIMINARY DRAFT
A.4 Attention Check and Study Designs
Design 1 and Attention Check:
Design 2 and Attention Check:
42

PRELIMINARY DRAFT
A.5 Experiment Disclosure Questions (Highly Intrusive in Bold)
Question
Text
1*
Have you ever downloaded a pirated song from the internet?
2*
While in a relationship, have you ever flirted with somebody other than your
partner?
3
Have you ever masturbated at work or in a public restroom?
4
Have you ever fantasized about having violent nonconsensual sex with
someone?
5
Have you ever tried to gain access to some else's (e.g. a partner, friend,
or colleague's) email account?
6*
Have you ever looked at pornographic material?
7*
Have you ever used drugs of any kind (e.g. weed, heroin, crack)?
8
Have you ever let a friend drive after you thought he or she had had too
much to drink?
9*
Have you ever made up a serious excuse, such as grave illness or death in
the family, to get out of doing something?
10
Have you ever had sex in a public venue (e.g. restroom of a club,
airplane)?
11
Have you ever while an adult, had sexual desires for a minor?
12
Have you ever had a fantasy of doing something terrible (e.g. torture) to
someone?
*Questions used in Experiment 3
A.6 Analysis using revised measure of Hypothetical Disclosure
Allowing our measure to be both continuous and also include admissions as those indicating they
are unsure do not qualitatively change our results: Objective changes have a significant impact on
hypothetical behavior while relative changes do not.
Continuous (Likert)
Revised Binary Measure
Objective Decrease
*3.5 vs. 3.2, t(188)=-1.9, p=.06
*.6 vs. .69, t(188)=-2.05, p=.04
Relative Decrease
3.1 vs. 3.2, t(180) =-.69, p=.49
.56 vs. .6, t(180)=-.77, p=.44
Relative Increase
3.5 vs. 3.5, t(181) =-.26, p=.79
.7 vs. .69, t(181)=-.13, p=.91
43