How Near-Miss Events Amplify or Attenuate Risky Decision Making

Page created by Charlie Turner
 
CONTINUE READING
CREATE Research Archive
Published Articles & Papers

2012

How Near-Miss Events Amplify or Attenuate Risky
Decision Making
Robin Dillon-Merrill
Georgetown University, rld9@georgetown.edu

Catherine H. Tinsley
Georgetown University, tinsleyc@georgetown.edu

Matthew A. Cronin
George Mason University, mcronin@gmu.edu

Follow this and additional works at: http://research.create.usc.edu/published_papers

Recommended Citation
Dillon-Merrill, Robin; Tinsley, Catherine H.; and Cronin, Matthew A., "How Near-Miss Events Amplify or Attenuate Risky Decision
Making" (2012). Published Articles & Papers. Paper 93.
http://research.create.usc.edu/published_papers/93

This Article is brought to you for free and open access by CREATE Research Archive. It has been accepted for inclusion in Published Articles & Papers
by an authorized administrator of CREATE Research Archive. For more information, please contact gribben@usc.edu.
Published online ahead of print April 18, 2012

                                                                                                                                        MANAGEMENT SCIENCE
                                                                                                                                        Articles in Advance, pp. 1–18
                                                                                                                                        ISSN 0025-1909 (print) — ISSN 1526-5501 (online)                                            http://dx.doi.org/10.1287/mnsc.1120.1517
                                                                                                                                                                                                                                                            © 2012 INFORMS
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                                                 How Near-Miss Events Amplify or
                                                                                                                                                                 Attenuate Risky Decision Making
                                                                                                                                                                                 Catherine H. Tinsley, Robin L. Dillon
                                                                                                                                                                     McDonough School of Business, Georgetown University, Washington, DC 20057
                                                                                                                                                                                 {tinsleyc@georgetown.edu, rld9@georgetown.edu}

                                                                                                                                                                                            Matthew A. Cronin
                                                                                                                                                              School of Management, George Mason University, Fairfax, Virginia 22030, mcronin@gmu.edu

                                                                                                                                              I n the aftermath of many natural and man-made disasters, people often wonder why those affected were
                                                                                                                                                underprepared, especially when the disaster was the result of known or regularly occurring hazards (e.g.,
                                                                                                                                              hurricanes). We study one contributing factor: prior near-miss experiences. Near misses are events that have
                                                                                                                                              some nontrivial expectation of ending in disaster but, by chance, do not. We demonstrate that when near misses
                                                                                                                                              are interpreted as disasters that did not occur, people illegitimately underestimate the danger of subsequent
                                                                                                                                              hazardous situations and make riskier decisions (e.g., choosing not to engage in mitigation activities for the
                                                                                                                                              potential hazard). On the other hand, if near misses can be recognized and interpreted as disasters that almost
                                                                                                                                              happened, this will counter the basic “near-miss” effect and encourage more mitigation. We illustrate the robust-
                                                                                                                                              ness of this pattern across populations with varying levels of real expertise with hazards and different hazard
                                                                                                                                              contexts (household evacuation for a hurricane, Caribbean cruises during hurricane season, and deep-water oil
                                                                                                                                              drilling). We conclude with ideas to help people manage and communicate about risk.
                                                                                                                                              Key words: near miss; risk; decision making; natural disasters; organizational hazards; hurricanes; oil spills
                                                                                                                                              History: Received June 29, 2010; accepted November 27, 2011, by Teck Ho, decision analysis. Published online
                                                                                                                                                in Articles in Advance.

                                                                                                                                        Introduction                                                           When people escape an impending disaster by
                                                                                                                                        In the aftermath of Hurricane Katrina, the public and               chance, they have experienced a “near miss.” More
                                                                                                                                        media alike questioned why so many people failed                    precisely, a near miss is an event that has some non-
                                                                                                                                        to evacuate the Gulf Coast and why the government                   trivial expectation of ending in disaster but because of
                                                                                                                                        and first-responder organizations were so appallingly               luck did not (Reason 1997, Dillon and Tinsley 2008).1
                                                                                                                                        underprepared (Glasser and Grunwald 2005). The rea-                 Our natural environment produces many examples
                                                                                                                                        sons for these failures are often rooted in experiences             of near misses: a random tree pattern saves a house
                                                                                                                                        with previous hurricanes. In the lead-up to the storm,              from a mud slide or a hurricane weakens right before
                                                                                                                                        Governor Haley Barbour of Mississippi warned of                     it hits a city. Organizations experience near misses
                                                                                                                                        “hurricane fatigue”—the possibility that his constit-               as well. For example, in the deep-sea oil drilling
                                                                                                                                        uents would not evacuate because they had success-                  industry, dozens of Gulf of Mexico wells in the past
                                                                                                                                        fully weathered earlier storms; similarly, one former
                                                                                                                                                                                                            two decades suffered minor blowouts during cement-
                                                                                                                                        Federal Emergency Management Agency official said
                                                                                                                                                                                                            ing; however, in each case chance factors (e.g., favor-
                                                                                                                                        people in the agency unfortunately approached the
                                                                                                                                                                                                            able wind direction, no one welding near the leak at
                                                                                                                                        Katrina response as it had other responses, though the
                                                                                                                                        aftermath of Katrina was clearly “unusual” (Glasser                 the time, etc.) helped prevent an explosion (Tinsley
                                                                                                                                        and Grunwald 2005). Such complacency is not exclu-                  et al. 2011).
                                                                                                                                        sive to hurricanes. Citizens who survive natural disas-                We study how prior near misses influence peoples’
                                                                                                                                        ters in one season often fail to take actions that would            interpretation of similar hazards and thus influence
                                                                                                                                        mitigate their risk in future seasons (e.g., moving off             future mitigation decisions. We do this in multiple
                                                                                                                                        a Midwestern flood plain or clearing brush to prevent               contexts: a single household threatened by hurricane,
                                                                                                                                        wildfires in the West; see Lindell and Perry 2000). Our
                                                                                                                                        research demonstrates that people may be complacent                 1
                                                                                                                                                                                                             Other events have been labeled “near misses” such as last minute
                                                                                                                                        because prior experience with a hazard can subcon-
                                                                                                                                                                                                            heroic efforts to avert crisis or interventions of chance that cause
                                                                                                                                        sciously bias their mental representation of the hazard             bad rather than good outcomes (e.g., narrowly missing an air-
                                                                                                                                        in a way that often (but not always) promotes unre-                 plane departure). Thus we specify our focus here on near misses
                                                                                                                                        alistic reassurance.                                                as chance-dependent good outcomes.

                                                                                                                                                                                                        1
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        2                                                                    Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS

                                                                                                                                        a planned Caribbean cruise threatened by a hurri-             causing major flooding and wind damage. This infor-
                                                                                                                                        cane, and oil-rig operations threatened by a danger-          mation provides input to assessing probabilities and
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                        ous storm. We explain why and when a near miss                outcomes, but it also cues the retrieval of other
                                                                                                                                        produces complacency versus action and offer pre-             information that will be used to refine these esti-
                                                                                                                                        scriptions for risk communication strategies based on         mates and their relationship to each other. Krizan and
                                                                                                                                        how different types of near misses operate.                   Windschitl (2007) provide a useful conceptualization
                                                                                                                                                                                                      of this knowledge retrieval process: given a situation
                                                                                                                                        How Near Misses Influence                                     involving risk, people must assess what this infor-
                                                                                                                                                                                                      mation means in light of what they already know.
                                                                                                                                        Cognitive Processes                                           “What they already know” is the domain knowl-
                                                                                                                                        When facing an imminent hazard, people should
                                                                                                                                                                                                      edge Gonzalez and Wu (1999) spoke of as modify-
                                                                                                                                        assess the risk, which is technically a function of the
                                                                                                                                                                                                      ing assessments of probabilities and outcomes and
                                                                                                                                        probability of the event occurring and the harm that
                                                                                                                                                                                                      their combination. To select which domain of knowl-
                                                                                                                                        results from the event if it occurs (Kaplan and Garrick
                                                                                                                                                                                                      edge to use, people can use the cognitive category to
                                                                                                                                        1981, von Neumann and Morgenstern 1944). This is
                                                                                                                                        the classic subjective expected utility (SEU) model.          which a hazard event belongs (Kahneman and Miller
                                                                                                                                        For example, to decide whether or not to evacuate             1986). “Hurricane” represents a category of events
                                                                                                                                        for an impending hurricane, people should combine             that guides the retrieval of relevant knowledge from
                                                                                                                                        assessments of the likelihood of their location being         memory. So although an avalanche is also a hazard,
                                                                                                                                        hit by the hurricane and how bad the damage could             knowledge about avalanches would not be retrieved
                                                                                                                                        be. Such assessments make use of the information              based on the hurricane category (although knowledge
                                                                                                                                        at hand, but people also bring past personal experi-          of flooding, which is related to hurricanes, might).
                                                                                                                                        ences into their evaluation of the risk (Fishbein and            Near misses come into play in that they can modify
                                                                                                                                        Azjen 2010, Tierney et al. 2001). We show that a par-         the hazard category, because prior experiences with
                                                                                                                                        ticular type of personal experience, near misses, have        an event can alter the cognitive category for that event
                                                                                                                                        an undue influence on how people evaluate risk and            (Kahneman and Miller 1986). Thus, after a near miss,
                                                                                                                                        can lead to questionable choices when people face an          the knowledge people will use in assessing SEU com-
                                                                                                                                        impending hazard with which they have had prior               ponents for a future hazard will change. This explains
                                                                                                                                        near-miss experience. We show that this near-miss             why prior outcomes can strongly influence future
                                                                                                                                        effect is robust because it seems to implicitly influ-        decisions and realized outcomes tend to be seen as
                                                                                                                                        ence the thoughts people use as inputs to their deci-         deterministic (Hastie and Dawes 2001).
                                                                                                                                        sion making. This near-miss effect can be countered,             The chain of events we posit is as follows:
                                                                                                                                        but doing so needs to use the same kind of implicit           (1) upon encountering a hazard, people retrieve rel-
                                                                                                                                        mechanism.                                                    evant knowledge from memory about that hazard,
                                                                                                                                           Although an SEU model provides a strong basis for          a process that is largely implicit (Anderson 1983, 1993;
                                                                                                                                        characterizing how people decide to respond to haz-           Kahneman and Miller 1986) but results in assessments
                                                                                                                                        ards, past research (Gonzalez and Wu 1999, Tversky            of probabilities, outcomes, and how they will be com-
                                                                                                                                        and Fox 1995) has shown that the model compo-                 bined; (2) an explicit evaluation of the risk of the haz-
                                                                                                                                        nents (including the likelihood estimates for probabil-       ard is made largely using an SEU framework; and
                                                                                                                                        ity, (un)attractiveness estimates for outcomes, and the       (3) once the risk is evaluated, people must explicitly
                                                                                                                                        ways in which these can be combined into an eval-             choose what behavior to engage. In the next section,
                                                                                                                                        uation of risk) can vary based on characteristics of          we hypothesize how near misses influence this chain
                                                                                                                                        the situation such as whether the likelihood estimates        of events.
                                                                                                                                        are very large, moderate, or very small (Tversky and
                                                                                                                                        Fox 1995). More importantly for the present work,             Hypotheses
                                                                                                                                        Gonzalez and Wu (1999) demonstrated that SEU can              Dillon and Tinsley (2008) found that near misses
                                                                                                                                        vary both between and within individuals (i.e., the           in completing a space project encouraged people to
                                                                                                                                        same person may be risk averse in one situation and           choose a riskier strategy when faced with a future
                                                                                                                                        risk seeking in another) because the components are           hazard threat to the mission. Although highly contex-
                                                                                                                                        all sensitive to the domain knowledge people use              tualized and specific, their research showed that near
                                                                                                                                        when evaluating the risky event.                              misses are events that alter evaluations of risk, and
                                                                                                                                           We argue that near misses change the domain                thus a near-miss bias should generalize to many kinds
                                                                                                                                        knowledge (or cognitive category) that people use in          of hazards and be relevant to a large array of natu-
                                                                                                                                        their assessment of the SEU components, and thus              ral and man-made hazard environments. Near-miss
                                                                                                                                        can bias the judgments people make about risky sit-           events in the hazard context often highlight resiliency
                                                                                                                                        uations. For example, people may learn that a hur-            because people escape harm. For example, imagine
                                                                                                                                        ricane has a 50% chance of striking their town and            that a hurricane is being tracked in the Caribbean and
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS                                                           3

                                                                                                                                        is a concern to two neighborhoods, A and B. Peo-                     to our scenario, imagine there is another neighbor-
                                                                                                                                        ple in both neighborhoods rely on existing domain                    hood C next to neighborhood B who also experienced
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                        knowledge to direct their thinking about the risk                    the hurricane. Because neighborhood C was closer to
                                                                                                                                        of this situation and whether or not to take protec-                 the center of the storm, neighborhood C was hit with
                                                                                                                                        tive action. Assume that as the storm grows nearer,                  a stronger force, and their sandbag levees collapsed,
                                                                                                                                        it becomes clear that the hurricane will miss neigh-                 resulting in significant flooding. For the people of
                                                                                                                                        borhood A, but neighborhood B is still in danger,                    neighborhood B, seeing damage to neighborhood C
                                                                                                                                        and they create a sandbag levee around the main                      further modifies the basic near-miss information (“we
                                                                                                                                        city buildings. Fortunately, when the hurricane makes                were ok, but look what happened to them”) to alter
                                                                                                                                        landfall, the storm surge subsides before overtopping                the hurricane category away from resilience. With the
                                                                                                                                        B’s makeshift levee, and the town suffers no damage.                 stimulus of damage to a neighboring town, they may
                                                                                                                                        In this illustration, neighborhood A did not experi-                 encode that the disaster almost caused harm, and that
                                                                                                                                        ence a near miss because there was no expectation                    they were vulnerable to possible damage.
                                                                                                                                        of harm. Neighborhood B experiences a near miss                         If the near-miss effect operates through mostly
                                                                                                                                        because there was a nontrivial expectation that the                  implicit processes, then we expect that counteracting
                                                                                                                                        flooding could occur, but for good fortune (i.e., chance             the near-miss effect will require further modification
                                                                                                                                        storm characteristics) it did not. We argue that the                 of the hazard category (Kahneman and Miller 1986).
                                                                                                                                        near-miss event experienced by people of neighbor-                   When the near-miss experience also highlights the
                                                                                                                                        hood B will change the hurricane category knowledge                  harm the event could have caused, it adds informa-
                                                                                                                                        in a way that when facing a new hazard warning,                      tion to counteract the basic resilient near-miss effect;
                                                                                                                                        the domain knowledge retrieved will make the haz-                    that is, the near miss (no harm done) can alter the cat-
                                                                                                                                        ard seems less threatening, leading to complacency.                  egory to make the hazard seem less threatening, but
                                                                                                                                        Thus, near misses that emphasize resiliency will lead                new harm information counteracts this with associa-
                                                                                                                                        to riskier behavior.                                                 tions of vulnerability. In our illustration, the people
                                                                                                                                           Hypothesis 1 (H1). People with near-miss information              from neighborhood B would now encode informa-
                                                                                                                                        that highlights how a disaster did not happen will be less           tion about both resilience (from the absence of dam-
                                                                                                                                        likely to take mitigating action for an impending hazard             age) and potential harm (a neighboring town was
                                                                                                                                        than people without this information.                                severely flooded). When facing a warning about a
                                                                                                                                                                                                             future impending hurricane, people from neighbor-
                                                                                                                                           The process we have described for how near misses                 hood B should now be less swayed by the fact that
                                                                                                                                        work (change to the category knowledge) does not                     they escaped harm.
                                                                                                                                        restrict the direction in which the category may be
                                                                                                                                        modified. Near-miss experiences do have some plas-                      Hypothesis 2 (H2). People with vulnerable near-miss
                                                                                                                                        ticity in their interpretation. For example, in their dis-           information (that highlights how an event almost caused
                                                                                                                                        cussion of aviation near misses, March et al. (1991,                 harm) will be more likely to take mitigating action for
                                                                                                                                        p. 10) essentially argue that near collisions can pro-               an impending hazard than people with resilient near-miss
                                                                                                                                        duce two different types of salient associations. They               information.
                                                                                                                                        describe:                                                               How the new hazard is evaluated will depend on
                                                                                                                                            Every time a pilot avoids a collision, the event provides        the category knowledge retrieved, which in turn is
                                                                                                                                            evidence both for the threat [of a collision] and for its        dependent on how the prior near miss modified the
                                                                                                                                            irrelevance. It is not clear whether the 0 0 0 organization      hazard category. Moreover, this modification could
                                                                                                                                            came [close] to a disaster 0 0 0 or that the disaster was        produce different assessments of probabilities (P ),
                                                                                                                                            avoided.                                                         outcomes (O), or risk (R) because the information
                                                                                                                                          If people experience the near miss as a disas-                     about the particular hazard that is embedded in the
                                                                                                                                        ter that almost happened rather than a disaster that                 warning will be integrated with domain knowledge
                                                                                                                                        was avoided, then their hazard category should be                    about the hazard category, thereby influencing some
                                                                                                                                        associated with vulnerability. We distinguish these                  part of the SEU model (P , O, and/or R).
                                                                                                                                        “vulnerable” near misses (wherein a disaster almost                     We predict that near misses change the negativ-
                                                                                                                                        happened and results in the perceived vulnerability                  ity associated with a bad event rather than chang-
                                                                                                                                        of the system) from “resilient” near misses (wherein                 ing probability assessments. This is consistent with
                                                                                                                                        a disaster could have but did not happen and results                 Windschitl and Chambers’ (2004) finding that people
                                                                                                                                        in the perceived resilience of the system).2 Returning               are more likely to change their feelings about a choice
                                                                                                                                                                                                             than their explicit beliefs about the probabilities. Fur-
                                                                                                                                        2
                                                                                                                                         See also Kahneman and Varey (1990) for arguments on the critical    thermore, in the domain of near misses, Dillon and
                                                                                                                                        distinction between an event that did not occur and an event that    Tinsley (2008) showed that people changed their per-
                                                                                                                                        did not but almost occurred.                                         ceptions of risk without changing their probabilities.
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        4                                                                            Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS

                                                                                                                                        Depending on the type of near miss (resilient or                      to justify their choice (i.e., reverse causality). Study 6
                                                                                                                                        vulnerable), the valence of the information retrieved                 corroborated the findings of Studies 2–5 with actual
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                        should change, influencing risk estimates. As the risk                behavior by having participants’ decisions regarding
                                                                                                                                        estimates change, so should the resulting judgments                   a risky situation have financial consequences for their
                                                                                                                                        about what to do for an impending hazard.                             compensation.
                                                                                                                                           Hypothesis 3 (H3). Resilient near misses will decrease
                                                                                                                                                                                                              Study 1
                                                                                                                                        one’s feelings of risk more than vulnerable near misses
                                                                                                                                                                                                              This study provides evidence of the near-miss effect
                                                                                                                                        without changing perceived probabilities, and these feelings
                                                                                                                                                                                                              in actual hazard situations. It is well established that
                                                                                                                                        of risk will mediate the corresponding behavioral response.
                                                                                                                                                                                                              people’s mitigation decisions, evacuation in the case
                                                                                                                                                                                                              of hurricanes, are influenced by what relevant oth-
                                                                                                                                        Overview of Studies                                                   ers do (Tierney et al. 2001). We tested whether or not
                                                                                                                                        Our hypotheses were tested across multiple stud-                      prior near-miss experiences reduce evacuation behav-
                                                                                                                                        ies, where we sought different types of respondents                   ior beyond what is due to social cues and a house-
                                                                                                                                        and used different threats and contexts to demon-                     hold’s specific geographic location (proximity to coast
                                                                                                                                        strate that our effects are robust across various pop-                and waterways). This speaks to the importance of the
                                                                                                                                        ulations and decisions. Study 1 looked for evidence                   effect (i.e., it is not overwhelmed by people’s incli-
                                                                                                                                        of the near-miss effect using a field survey of house-                nation to do what their neighbors do), and why it
                                                                                                                                        holds in coastal counties of Louisiana and Texas                      warrants further study.
                                                                                                                                        who experienced Hurricane Lili.3 We examined how                         Participants and Procedure. In the spring of 2003,
                                                                                                                                        previous storm experience as well as prior near-                      six months after Hurricane Lili hit the Louisiana
                                                                                                                                        miss experiences (in the form of unnecessary evac-                    coastline, 1,000 households from five affected areas
                                                                                                                                        uations) influenced whether or not the individuals                    (200 each area: Vermilion and Cameron Parishes in
                                                                                                                                        surveyed evacuated for Hurricane Lili. Studies 2–6                    Louisiana and Orange, Jefferson, and Chambers coun-
                                                                                                                                        used the laboratory to discover how the near-miss                     ties in Texas) were randomly selected and mailed a
                                                                                                                                        phenomenon operates. Study 2 examined how encod-                      survey by the Hazard Reduction and Recovery Cen-
                                                                                                                                        ing near misses as resilient or vulnerable led to dif-                ter at Texas A&M asking whether they had evacu-
                                                                                                                                        ferent evacuation rates for a hypothetical hurricane                  ated. For the storm, the National Hurricane Center
                                                                                                                                        and demonstrated that the addition of vulnerability                   had issued a hurricane warning, and local officials
                                                                                                                                        information to the near-miss stimulus can counteract                  had issued an early evacuation advisory in these five
                                                                                                                                        the complacency effect. Study 3 examined the compo-                   areas. A total of 507 usable surveys were returned for
                                                                                                                                        nents of people’s SEU assessments. It probed people’s                 a response rate of 50.7%, which exceeds similar hur-
                                                                                                                                        assessments of probabilities (P ), outcome attractive-                ricane studies (Prater et al. 2000, Lindell et al. 2001).4
                                                                                                                                        ness (O), and their ultimate judgments of risk versus                 To obtain this response rate, nonresponsive house-
                                                                                                                                        safety (R) to test our hypothesized mediation. Study                  holds were sent a follow-up survey every three weeks
                                                                                                                                        4 generalizes our basic finding by changing the con-                  (until a total of three surveys had been sent).
                                                                                                                                        text from a house to a cruise ship; in doing so we                       Variables. Respondents were asked (on a 1–5 scale,
                                                                                                                                        address a concern that participants may be updating                   where 1 equaled “not at all” and 5 equaled “very
                                                                                                                                        their calculations of the risk after a resilient near miss.           great extent”) whether or not they had “previous
                                                                                                                                        Additionally, in Study 4, we examine the role coun-                   experience with an unnecessary evacuation.” This
                                                                                                                                        terfactuals have in the risky decision. Study 5 offered               was our proxy for a resilient near miss (i.e., where
                                                                                                                                        evidence that near misses do in fact change the hazard                a disaster did not happen), which was treated as
                                                                                                                                        category, and hence the knowledge associated with
                                                                                                                                                                                                              the independent variable. For the dependent variable,
                                                                                                                                        a hazard, by examining what participants’ thought
                                                                                                                                                                                                              respondents were asked whether or not they evacu-
                                                                                                                                        about a hazardous situation. This study removed the
                                                                                                                                                                                                              ated. For control variables, respondents were asked
                                                                                                                                        need to make a decision, thereby (a) providing evi-
                                                                                                                                                                                                              (on a 1–5 scale, where 1 equaled “not at all” and 5
                                                                                                                                        dence for the first (implicit) step in our sequence of
                                                                                                                                                                                                              equaled “very great extent”) about individual geo-
                                                                                                                                        how near misses affect cognitive processes and (b)
                                                                                                                                                                                                              graphic proximity including how close they lived to
                                                                                                                                        discounting a concern that people first chose what to
                                                                                                                                                                                                              the coast and how close they lived to inland water
                                                                                                                                        do and then, when forced to answer questions, gen-
                                                                                                                                                                                                              such as bays, bayous, or rivers. Also on the same
                                                                                                                                        erate assessments of probabilities, outcomes, and risk
                                                                                                                                                                                                              1–5 scale, respondents were asked about social cues
                                                                                                                                                                                                              including whether they saw businesses closing; saw
                                                                                                                                        3
                                                                                                                                         The survey was conducted six months after Hurricane Lili by the
                                                                                                                                        Hazard Reduction and Recovery Center at Texas A&M. Hurricane
                                                                                                                                                                                                              4
                                                                                                                                        Lili was the deadliest and costliest hurricane of the 2002 Atlantic    See Lindell et al. (2005) for more details of the original survey
                                                                                                                                        hurricane season.                                                     collection.
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS                                                          5

                                                                                                                                        friends, relatives, neighbors, or coworkers evacuat-                 this study was to provide empirical evidence beyond
                                                                                                                                        ing; heard announcement of a hurricane warning;                      the post hoc evaluations of highly visible disasters
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                        and heard local authorities issue a recommendation                   like Katrina that the near-miss effect happens and
                                                                                                                                        to evacuate. They were also asked whether or not                     merits further study. To understand the mechanics
                                                                                                                                        they saw storm conditions such as high wind, rain, or                of the near-miss phenomena in detail, we examine it
                                                                                                                                        flooding, and whether they had personal experience                   using a series of laboratory studies.
                                                                                                                                        with hurricane storm conditions.
                                                                                                                                           Analysis and Results. Factor analysis revealed that               Study 2
                                                                                                                                        some control variables could be averaged into scales.                We argue that near-miss information encourages
                                                                                                                                        We created the “individual geographic proximity”                     riskier behavior (H1), but that this effect can be coun-
                                                                                                                                        (alpha = 0074) scale by averaging the first two control              teracted when the near miss includes information
                                                                                                                                        variables and the “social cue” (alpha = 0081) scale by               that highlights vulnerability (H2). We tested for the
                                                                                                                                        averaging the next four control variables.                           difference between a resilient near miss and a vul-
                                                                                                                                           Binary logistic regression was used to test H1, with              nerable near miss by giving participants informa-
                                                                                                                                        evacuation (yes/no) as the dependent variable. Four                  tion about an impending hurricane and asking them
                                                                                                                                        control variables were entered in the first step (geo-               whether or not they would evacuate. We have also
                                                                                                                                        graphic proximity, social cues, see storm conditions, and            said that this process operates automatically, and that
                                                                                                                                        prior hurricane experience); our independent variable                such automaticity makes the effect robust even in the
                                                                                                                                        (prior unnecessary evacuation) was entered in the sec-               face of experience. In this study, we verify our basic
                                                                                                                                        ond step. Regression results, displayed in Table 1,                  hypotheses and test the robustness of the effect across
                                                                                                                                        show that geographic proximity, social cues, and see-                populations with varying levels of experience and
                                                                                                                                        ing storm conditions all have a positive influence on                expertise.
                                                                                                                                        evacuation behavior, whereas prior unnecessary evac-                    Participants. For Study 2, we collected data from
                                                                                                                                        uations has a negative influence. Thus, controlling for              four different samples. Participants were (1) 352
                                                                                                                                        geographic proximity, social cues, and seeing storm                  undergraduate and 47 graduate business students
                                                                                                                                        conditions, prior near-miss experiences in the sense                 from a large, private university in the eastern United
                                                                                                                                        of having evacuated when later deemed unnecessary                    States who completed a number of exercises, includ-
                                                                                                                                        lead to less protective action in the form of evacuation             ing ours, in return for class participation points;
                                                                                                                                        in the face of an impending hurricane, supporting H1.                (2) 82 upperclass undergraduate students at Tulane
                                                                                                                                           Discussion. This study shows that prior near-miss                 University in New Orleans (two-thirds of whom evac-
                                                                                                                                        experiences influence the behavior of people facing                  uated for Hurricane Katrina) who completed the short
                                                                                                                                        similar subsequent threats, even amid all the con-                   exercise at the end of a regularly scheduled lec-
                                                                                                                                        current forces that affect such behavior. Those with                 ture session; (3) 187 undergraduate business students
                                                                                                                                        resilient near-miss experiences were significantly less              from the same university as sample 1 who com-
                                                                                                                                        likely to evacuate than those without this experi-                   pleted a number of exercises online, of which ours
                                                                                                                                        ence, supporting H1. We recognize that an unneces-                   was one, in return for class participation points; and
                                                                                                                                        sary evacuation is an imperfect proxy for a resilient                (4) 102 emergency managers who averaged 13.6 years
                                                                                                                                        near-miss experience, although we think there is cor-                of experience with natural disasters, whose participa-
                                                                                                                                        respondence because unnecessary evacuations imply                    tion was solicited though email lists and newsletters
                                                                                                                                        that a disaster did not happen. However, the point of                associated with the Natural Hazard Center in Col-
                                                                                                                                                                                                             orado, and who participated in exchange for entrance
                                                                                                                                                                                                             in a lottery to win sweatshirts.
                                                                                                                                                Table 1        Study 1—Logistic Regression Results for          Procedure. Participants read that they lived in
                                                                                                                                                               Evacuation from Hurricane Lili by Near Miss
                                                                                                                                                                                                             an area subject to hurricanes and that the National
                                                                                                                                                                                                 Model 1     Weather Service was tracking a hurricane that had a
                                                                                                                                                                                                Odds ratio   30% chance of hitting their community with moder-
                                                                                                                                                Control variables
                                                                                                                                                                                                             ate force within 36 hours. They were also told that
                                                                                                                                                  geographic proximity                            1038∗∗     they lived alone, had no pets, and that evacuation
                                                                                                                                                  social cues                                     2022∗∗∗    would incur a sure loss of $2,000. However, if they
                                                                                                                                                  see storm conditions                            0080∗      stayed and the hurricane hit, the collateral damage
                                                                                                                                                  prior hurricane experience                      1000       (above and beyond damage to house, such as damage
                                                                                                                                                Independent variable                                         to one’s car, self, portable personal belongings, etc.)
                                                                                                                                                  prior unnecessary                               0085∗
                                                                                                                                                                                                             would add up to $10,000 (see Appendix A for the
                                                                                                                                                     evacuation (near miss)
                                                                                                                                                                                                             full text). After reading the vignette, they answered
                                                                                                                                                Nagelkerke R2                                     0023∗∗
                                                                                                                                                                                                             whether or not they would evacuate. For the New
                                                                                                                                                ∗
                                                                                                                                                    p < 0005; ∗∗ p < 0001; ∗∗∗ p < 00001.                    Orleans sample, participants were also asked whether
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        6                                                                                     Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS

                                                                                                                                        or not they had evacuated for Hurricane Katrina, and                 Figure 1               Study 2—Evacuation Rate by Condition for Different Sample
                                                                                                                                        two-thirds reported that they had.5                                                         Collections (See Table 2 for • 2 Test Results)
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                           Variables. We had three conditions: whether                                  100
                                                                                                                                                                                                                         90                                                              Control
                                                                                                                                        participants had resilient or vulnerable near-miss                                                                                               Resilient NM
                                                                                                                                                                                                                         80
                                                                                                                                        information or no near-miss information (control).

                                                                                                                                                                                                              % Evacuation
                                                                                                                                                                                                                         70                                                              Vulnerable NM
                                                                                                                                        Participants in the no near-miss information (control)                           60
                                                                                                                                        condition read, “You have no specific data regarding                             50
                                                                                                                                                                                                                         40
                                                                                                                                        past hurricane impacts to your property.” Participants                           30
                                                                                                                                        in the resilient near-miss condition read, “You have                             20
                                                                                                                                        lived in this house through three prior storms similar                           10
                                                                                                                                                                                                                          0
                                                                                                                                        to that forecasted, and you and your neighbors have                                       General          New Orleans            General           Expert
                                                                                                                                        never had any property damage.” Participants in the                                       students          students              students        emergency
                                                                                                                                                                                                                                (Collection 1)                          (Collection 2)    managers
                                                                                                                                        vulnerable near-miss condition for Collections 1 and 2
                                                                                                                                                                                                                                                             Collection
                                                                                                                                        read the resilient near-miss condition plus “In the last
                                                                                                                                        storm, however, a tree fell on your neighbor’s house,                Note. NM, near miss.

                                                                                                                                        completely destroying the second story. If anyone had
                                                                                                                                        been inside, they would have been seriously hurt.”                   study and the previous field study show the near-
                                                                                                                                           Participants in the vulnerable near-miss condition                miss effect, neither provides evidence of the mecha-
                                                                                                                                        for Collections 3 and 4 read the resilient near-miss                 nism by which near-misses operate. The next study
                                                                                                                                        condition plus “In the last storm, however, a tree fell              examines whether people’s evaluations of the nature
                                                                                                                                        on your neighbor’s car and completely destroyed it.” 6               of the hazard mediate their decision to evacuate.
                                                                                                                                           The dependent variable was whether or not partic-
                                                                                                                                        ipants would evacuate.                                               Study 3
                                                                                                                                           Analysis and Results. Figure 1 shows the per-                     This study looked into the hypothesized mechanisms
                                                                                                                                        centage of participants in each condition for each                   through which near-miss information works. If differ-
                                                                                                                                        collection who chose to evacuate. For all four sam-                  ent types of near misses cause changes to the hazard
                                                                                                                                        ples, participants with resilient near-miss information              category, then we would expect the different types
                                                                                                                                        chose to evacuate significantly less than those with                 of near misses to change the assessments of risk (R)
                                                                                                                                                                                                             but not the assessments of probability (P ). Moreover,
                                                                                                                                        no near-miss information, supporting H1, and partic-
                                                                                                                                                                                                             assessments of R should mediate observed mitigation
                                                                                                                                        ipants with vulnerable near-miss information chose
                                                                                                                                                                                                             choices (H3).
                                                                                                                                        to evacuate more than those with resilient near-miss
                                                                                                                                                                                                               Participants and Procedure. Participants in Study 3
                                                                                                                                        information, supporting H2 (see Table 2 for • 2 tests).
                                                                                                                                                                                                             were 236 undergraduate and graduate business stu-
                                                                                                                                           Discussion. Study 2 found that people with
                                                                                                                                                                                                             dents who completed a number of exercises online, of
                                                                                                                                        resilient near-miss information that highlights how a
                                                                                                                                                                                                             which ours was one, in return for class participation
                                                                                                                                        disaster did not happen were less likely to evacu-
                                                                                                                                                                                                             points. Participants read the same story as in Study 2,
                                                                                                                                        ate for an impending hurricane than people without
                                                                                                                                                                                                             about living in a hurricane area, yet this time they
                                                                                                                                        near-miss information (supporting H1). On the other                  answered questions about the thoughts and feelings
                                                                                                                                        hand, people with vulnerable near-miss information                   associated with the impending hazard.
                                                                                                                                        that highlights how a disaster almost happened were
                                                                                                                                        more likely to evacuate than people with resilient
                                                                                                                                        near-miss information (supporting H2). And these                     Table 2               • 2 Results for Study 2
                                                                                                                                        results were robust across participants representative                                                   • 2 (1): Control vs.             • 2 (1): Resilient vs.
                                                                                                                                        of the general population (Collections 1 and 3), those                                                   resilient near miss             vulnerable near miss
                                                                                                                                        who live in a culture highly sensitive to hurricanes                 Data collection                       (Hypothesis 1)                    (Hypothesis 2)
                                                                                                                                        (i.e., New Orleans), many of whom had prior evacu-                   1 = General                   • 2 415 = 20063, p < 00001          • 2 415 = 11098, p < 00001
                                                                                                                                        ation experience (Collection 2), and emergency man-                    population
                                                                                                                                        agement practitioners (Collection 4). Although this                    students
                                                                                                                                                                                                             2 = Tulane                    • 2 415 = 8002, p < 0001            • 2 415 = 6096, p < 0001
                                                                                                                                                                                                               students
                                                                                                                                        5
                                                                                                                                         Note that given the timing of the data collection, these students     (experienced)
                                                                                                                                        would still have been in high school during Hurricane Katrina and    3 = General                   • 2 415 = 3031, p < 0005            • 2 415 = 10086, p < 0001
                                                                                                                                        not matriculated students at Tulane.                                   population
                                                                                                                                        6                                                                      students
                                                                                                                                         We altered the wording for Studies 3 and 4 to test whether the
                                                                                                                                                                                                             4 = Emergency                 (• 2 415 = 2085, p < 0005           • 2 415 = 6068, p < 0001
                                                                                                                                        effect was robust to different types of harm information—bodily
                                                                                                                                                                                                               managers
                                                                                                                                        injury versus harm to property—which it was. We thank Howard
                                                                                                                                                                                                               (experts)
                                                                                                                                        Kunreuther for this suggestion.
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS                                                                                                      7

                                                                                                                                           Variables. We used the same three conditions                      Figure 2                            Study 3—Risk Judgments by Condition
                                                                                                                                        from Study 2 for the independent variables: control,
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                                                                                                                            8
                                                                                                                                        resilient near miss, and vulnerable near miss (with                                                     Perceived risk   Likelihood hit     Outcome unattractiveness

                                                                                                                                                                                                               Scale from factor analysis
                                                                                                                                        the wording from Collections 3 and 4). For dependent                                                7
                                                                                                                                        variables, participants answered on a 10 point scale
                                                                                                                                                                                                                                            6
                                                                                                                                        (1 equaled “not at all” and 10 equaled “extremely”),
                                                                                                                                        about the extent to which they felt worried, anxious,                                               5
                                                                                                                                        vulnerable, distressed, dread, safe, and protected, and
                                                                                                                                        whether the situation before evacuating was risky.                                                  4
                                                                                                                                        They also answered on a 10 point scale (1 equaled
                                                                                                                                                                                                                                            3
                                                                                                                                        “not at all” and 10 equaled “very much”) how much
                                                                                                                                        they agreed with the following statements: the dam-                                                 2
                                                                                                                                        age will be bad, I could experience much harm, the                                                          Control           Resilient NM          Vulnerable NM
                                                                                                                                                                                                                                                                       condition
                                                                                                                                        damage will not be a big deal, my chances of being
                                                                                                                                        hit are good, and I will likely suffer damage. Finally,
                                                                                                                                        they were asked whether or not they would evacuate.                     Following James and Brett (1984), we assessed
                                                                                                                                           Analysis and Results. We used a factor analysis                   (1) whether the mediators are a probabilistic func-
                                                                                                                                        with varimax rotation to examine the associations                    tion of the independent variables, (2) whether the
                                                                                                                                        people had with the hurricane warning and found                      dependent variable is a probabilistic function of the
                                                                                                                                        three factors: (1) estimations of probability (of being              independent variable, (3) whether the dependent vari-
                                                                                                                                        hit, chances of being hit are good and I will likely                 able is a probabilistic function of the mediators, and
                                                                                                                                        suffer damage; alpha = 0081); (2) estimations of out-                (4) how the addition of the independent variable to
                                                                                                                                        come (un)attractiveness (damage will be bad, harm                    step 3 changes the variance explained in the depen-
                                                                                                                                        will be incurred, and damage will be no big deal                     dent variables. Full mediation occurs when the addi-
                                                                                                                                        (reverse coded); alpha = 0086); and (3) perceptions of               tion of the independent variables does not explain
                                                                                                                                                                                                             any additional variance in the dependent variables
                                                                                                                                        risk versus safety (worried, anxious, vulnerable, dis-
                                                                                                                                                                                                             beyond what the mediators explained (i.e., the change
                                                                                                                                        tressed, dread, risky, safe (reverse coded), and pro-
                                                                                                                                                                                                             in R2 from step 3 to step 4 is not significant). If the
                                                                                                                                        tected (reverse coded); alpha = 0093). Three scales
                                                                                                                                                                                                             change in R2 from step 3 to 4 is significant, then par-
                                                                                                                                        (probability of hit, outcome unattractiveness, and per-
                                                                                                                                                                                                             tial mediation is a possibility.
                                                                                                                                        ceived risk) were created for each factor averaging the
                                                                                                                                                                                                                Step 1 was accomplished with the MANOVA
                                                                                                                                        above described items, and were subject to a multi-
                                                                                                                                                                                                             detailed above (again see Figure 2). For step 2, binary
                                                                                                                                        variate analysis of variance (MANOVA) with condi-
                                                                                                                                                                                                             logistic regression showed that evacuation decisions
                                                                                                                                        tion (control, resilient near miss, and vulnerable near              were significantly influenced by resilient near-miss
                                                                                                                                        miss) as the independent variable.                                   experiences (Table 3, model 1). For step 3, binary
                                                                                                                                           The multivariate F was significant (Wilks Lambda                  logistic regression showed that evacuation decisions
                                                                                                                                        F461 4625 = 2023, p < 0005), as were the univariate F val-           were significantly influenced by perceived risk and
                                                                                                                                        ues for perceived risk (F411 2335 = 3056, p = 0003) and              outcome unattractiveness (Table 3, model 2). For
                                                                                                                                        outcome unattractiveness (F411 2335 = 3004, p = 0005). The
                                                                                                                                        means for the three scales by condition are plotted
                                                                                                                                                                                                             Table 3                            Study 3—Binary Logistic Regressions on Evacuation Behavior
                                                                                                                                        in Figure 2. Planned contrasts (using Tukey’s hon-
                                                                                                                                        est significant difference) showed that for perceived                                                                           Model 1         Model 2      Model 3
                                                                                                                                        risk, the significant difference across conditions was                                                                         Odds ratio      Odds ratio   Odds ratio
                                                                                                                                        driven by the resilient near-miss condition being sig-               Independent variablea
                                                                                                                                        nificantly lower than the vulnerable near-miss condi-                  Dummy resilient near miss                                 0056∗                        0065
                                                                                                                                        tion (p < 0005), and marginally lower than the control                 Dummy vulnerable near miss                                0067                         0055
                                                                                                                                        (p = 001). For outcome unattractiveness, the signifi-                Mediators
                                                                                                                                        cant difference across conditions was again driven by                  perceived risk (R)                                                        1028∗∗       1027∗∗
                                                                                                                                                                                                               outcome unattractiveness (O)                                              1013+        1014+
                                                                                                                                        the resilient near-miss condition being significantly                  probability of hit (P )                                                   1001         1003
                                                                                                                                        lower than the vulnerable near-miss condition (p <
                                                                                                                                                                                                             Nagelkerke R2                                               0002∗           0012∗∗       0013∗∗
                                                                                                                                        0005). Probability of hit was not significantly differ-              Change in R2                                                                             0001
                                                                                                                                        ent across conditions. To test for mediation we used                   (from models 2 to 3)
                                                                                                                                        binary logistic regression on whether or not people                     a
                                                                                                                                                                                                                  For the condition variable, the control was chosen as the reference cat-
                                                                                                                                        chose to evacuate. Dummy variables were created for                  egory; thus the regression weights for the first row, for example, show the
                                                                                                                                        the resilient near-miss and vulnerable near-miss con-                effect for having resilient near-miss information.
                                                                                                                                                                                                                +
                                                                                                                                        ditions, using control as the referent category.                          p < 0007; ∗ p < 0005; ∗∗ p < 0001.
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        8                                                                     Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS

                                                                                                                                        step 4, when the condition variables were added to             might produce more downward counterfactuals than
                                                                                                                                        model 2 as additional explanatory variables, these             resilient near misses, which might impel the protec-
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                        condition variables did not contribute any unique              tive action we see participants take in this condition.
                                                                                                                                        additional explanatory power (Table 3, model 3). The           This might be particularly true if the severity of the
                                                                                                                                        nonsignificant betas in model 3 for the condition vari-        negative consequences in the near-miss information is
                                                                                                                                        ables and lack of any measurable change in R2 sug-             high (that someone could have been severely injured
                                                                                                                                        gest full mediation. Near-miss information influences          or even killed) rather than low (that someone could
                                                                                                                                        perceptions about risk and consequences, which influ-          have been inconvenienced). Thus, we test for counter-
                                                                                                                                        ence whether or not participants evacuated in the face         factual thinking in this study, and we vary the sever-
                                                                                                                                        of a hurricane warning.                                        ity of the negative consequences (high versus low).
                                                                                                                                           Discussion. Study 3 found that near-miss informa-              Another limitation of Studies 2 and 3 is that people
                                                                                                                                        tion influenced the associations people had with the           may be legitimately updating their beliefs about the
                                                                                                                                        hazard, which mediated people’s subsequent mitiga-             resilience of their house using Bayesian logic; that is,
                                                                                                                                        tion behavior. The resilient near miss tends to be asso-       perhaps people are processing the near-miss experi-
                                                                                                                                        ciated with lower risk, which explains the consistent          ence as data to recalculate the likelihood of hurricane
                                                                                                                                        lack of protective action by these participants com-           damage to their particular house, thereby reducing
                                                                                                                                        pared to those in the other conditions. The vulnerable         it from the stated 30%. Moreover, given that their
                                                                                                                                        near misses (highlighting danger albeit to someone             house is stationary, each hurricane threat is not truly
                                                                                                                                        else) tend to counteract these reassurances. Study 3           an independent event, and participants could reason-
                                                                                                                                        supports our theoretical model, that near misses are           ably infer that their particular house is at less risk
                                                                                                                                        stimuli that influence the general hazard category             than what was previously calculated.
                                                                                                                                        (Kahneman and Miller 1986) so that assessments of                 In Study 4, we asked the person to decide whether
                                                                                                                                        risk are either raised or lowered (depending on type           or not to go on a Caribbean cruise that is threat-
                                                                                                                                        of near miss) to influence behavior. However, there            ened to be interrupted by a hurricane. The survival
                                                                                                                                        are still several alternative explanations for our find-       of a prior cruise ship during a Caribbean hurricane is
                                                                                                                                        ings that need to be examined, such as near misses             completely independent from the chances of survival
                                                                                                                                        encouraging counterfactual thought or prompting                of their current Caribbean cruise ship given the con-
                                                                                                                                        legitimate Bayesian updating. Our next studies test            stantly changing ship location. We nonetheless tested
                                                                                                                                        these alternatives, and in doing so demonstrate that           for likelihood updating by asking participants what
                                                                                                                                        our behavioral results generalize more broadly.                they believe to be the percentage chance of being hit
                                                                                                                                                                                                       by a storm. We also vary the severity descriptions
                                                                                                                                        Study 4                                                        of the possible consequences from high (warning of
                                                                                                                                        An alternative to our proposed mechanism (that near            severe injury or even death) to low (inconvenience)
                                                                                                                                        misses implicitly influences the knowledge associated          and explore the role of counterfactuals in their deci-
                                                                                                                                        with the hazard category) is that near misses prompt           sion process.
                                                                                                                                        counterfactual thoughts. A counterfactual is an alter-            Participants and Procedure. For Study 4, we col-
                                                                                                                                        native to reality. Thus a counterfactual thought is            lected data from 299 undergraduate business students
                                                                                                                                        thinking explicitly about what could, should, or might         who completed a number of exercises online, of which
                                                                                                                                        have been (Kahneman and Tversky 1982). Upward                  ours was one, in return for class participation points.
                                                                                                                                        counterfactuals are thoughts about how an alternative          Participants read that they had nonrefundable tickets
                                                                                                                                        could be better than the realized outcome; downward            for a Caribbean cruise that is leaving the next day, but
                                                                                                                                        counterfactuals are thoughts about how an alternative          the National Weather Service is currently tracking a
                                                                                                                                        could be worse than the realized outcome (Roese and            hurricane in the Caribbean that they estimate has a
                                                                                                                                        Olson 1995). Counterfactual thought is more likely to          30% chance of impacting the cruise. They were also
                                                                                                                                        occur when people encounter a surprise outcome than            provided costs associated with not going on the trip
                                                                                                                                        when they encounter a routine outcome (Kahneman                and with going on the trip if a hurricane diverts the
                                                                                                                                        and Miller 1986, Miller et al. 1989), or when activated        ship. The participant then decided whether or not to
                                                                                                                                        by a problem that needs to be addressed (such as a             go on the trip. For the full text of the exercise, see
                                                                                                                                        bad outcome that someone wishes to avoid) (Epstude             Appendix B.
                                                                                                                                        and Roese 2008). Thus, if near misses either surprise             Variables. Five conditions made up the indepen-
                                                                                                                                        participants (as in the resilient near miss, that dan-         dent variables: control and resilient near miss ver-
                                                                                                                                        ger was avoided) or are represented as a problem               sus vulnerable near miss crossed by strong versus
                                                                                                                                        (as in the vulnerable near miss, that danger almost            weak prime (see Appendix B for specific wording).
                                                                                                                                        happened), one could argue that they evoke coun-               To briefly summarize, in the control condition, partic-
                                                                                                                                        terfactual thinking, and that this guides the mitiga-          ipants were told that cruises can be diverted because
                                                                                                                                        tion behavior. For example, vulnerable near misses             of hurricanes but receive no information about prior
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS                                                                                                                            9

                                                                                                                                        cruises being impacted by hurricanes. In the resilient                                           conditions and compared these to the resilient near-
                                                                                                                                        conditions, participants read that cruises can be                                                miss conditions. Significantly more people with vul-
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                        diverted by hurricanes but that they have been on                                                nerable near-miss information were willing to forgo
                                                                                                                                        three prior cruises and never experienced any prob-                                              the cruise than people with resilient near-miss infor-
                                                                                                                                        lems. In the vulnerable conditions, participants read                                            mation (• 2 415 = 8014, p < 0001), supporting H2.
                                                                                                                                        that cruises can be diverted by hurricanes but that                                                 To test whether severity of consequences had any
                                                                                                                                        they have been on three prior cruises and never expe-                                            influence on people’s travel decisions, we compared
                                                                                                                                        rienced any problems; however, they know someone                                                 the weak versus strong severity within each type
                                                                                                                                        else who has. In the weak conditions, participants                                               of near miss. Although participants whose near-miss
                                                                                                                                        were reminded that hurricane diversions can cause                                                experiences included severe (strong) consequences
                                                                                                                                        delay and cost money, and in the strong conditions                                               were slightly less likely to go on the cruise than par-
                                                                                                                                        participants were reminded that in addition to delays                                            ticipants whose near-miss experiences included weak
                                                                                                                                        and costs, people can be injured or even killed.                                                 consequences (50.8% versus 58.5% in the resilient
                                                                                                                                            For the dependent variables, participants answered                                           near-miss condition; 32.8% versus 39.0% in the vul-
                                                                                                                                        whether or not they would go on the trip. Then, par-                                             nerable near-miss condition), neither of these differ-
                                                                                                                                        ticipants were asked: “Please answer the following                                               ences were statistically significant (resilient, • 2 415 =
                                                                                                                                        statements as thoroughly as possible. ‘In making this                                            0067, p > 001; vulnerable, • 2 415 = 00501 p > 001). Thus,
                                                                                                                                        decision, I thought about if 0 0 0 1’ ‘I also thought about                                      for the following analyses, we collapse data across the
                                                                                                                                        if 0 0 0 1’ and ‘I also thought about if. 0 0 0’ ” Participants                                  severity conditions.
                                                                                                                                        then rated their belief that a hurricane would impact                                               To test whether people with near misses are updat-
                                                                                                                                        their ship (from 0%–100%).                                                                       ing their calculation of the likelihood of harm, we
                                                                                                                                                                                                                                         ran an ANOVA on participants’ belief that a hurri-
                                                                                                                                            The open-ended responses were coded by two
                                                                                                                                                                                                                                         cane would impact their ship (from 0%–100%). Across
                                                                                                                                        research assistants blind to conditions and hypothe-
                                                                                                                                                                                                                                         all conditions, participants slightly inflated their belief
                                                                                                                                        ses. They coded whether the statement contained an
                                                                                                                                                                                                                                         that a hurricane would impact their ship from the
                                                                                                                                        upward counterfactual (e.g., “If I go, I will have the
                                                                                                                                                                                                                                         given 30% (control mean, 35%; s.d., 18; resilient mean,
                                                                                                                                        time of my life”), a downward counterfactual (e.g.,
                                                                                                                                                                                                                                         33%; s.d., 17; vulnerable mean, 35%; s.d., 17), but there
                                                                                                                                        “If I were to get diverted on the cruise, I would miss
                                                                                                                                                                                                                                         were no significant differences across the conditions
                                                                                                                                        work”), a neutral counterfactual (e.g., “If I could sell
                                                                                                                                                                                                                                         (F421 2895 = 0032, p > 001).
                                                                                                                                        the cruise tickets to another party”), or no counterfac-
                                                                                                                                                                                                                                            To test whether near misses prompt counterfactual
                                                                                                                                        tual (e.g., “$2,000 is a sunk cost and I should not make
                                                                                                                                                                                                                                         thoughts, we looked at whether different near-miss
                                                                                                                                        my decision based on it”) (Nasco and Marsh 1999).                                                experiences produce different types of counterfactual
                                                                                                                                            Analysis and Results. Results for participants’                                              thoughts. Figure 4 shows the percentage of partic-
                                                                                                                                        decisions (to forgo the trip or not) are shown in                                                ipants’ counterfactual thoughts by condition. Most
                                                                                                                                        Figure 3. To test H1, we collapsed the two resilient                                             of participants’ responses contained a downward
                                                                                                                                        near-miss conditions and compared them to the no                                                 counterfactual thought, followed by no counterfac-
                                                                                                                                        near-miss experience (control). Significantly fewer                                              tual thought. Importantly, however, there were no
                                                                                                                                        people with resilient near-miss information were will-                                           systematic differences in types of thought across con-
                                                                                                                                        ing to forgo the cruise than people without near-miss                                            ditions (• 2 465 = 60451 p > 001). Thus, the explana-
                                                                                                                                        information (• 2 415 = 5099, p < 0005) supporting H1.                                            tion that near misses evoke a particular counterfactual
                                                                                                                                        To test H2, we collapsed the two vulnerable near-miss                                            thought to explain the observed differences in mitiga-
                                                                                                                                                                                                                                         tion behavior fails the first test requirement to demon-
                                                                                                                                                                                                                                         strate mediation (James and Brett 1984).
                                                                                                                                        Figure 3                              Study 4—Percentage of Participants Forgoing the Cruise,
                                                                                                                                                                              by Condition
                                                                                                                                                                                                                                         Figure 4                          Study 4—Percentage of Participants’ Counterfactual (CF)
                                                                                                                                         Percentage forgoing cruise

                                                                                                                                                                      0.8
                                                                                                                                                                                                                                                                           Thoughts, by Condition
                                                                                                                                                                      0.7
                                                                                                                                                                                                                                         Percentage of responses

                                                                                                                                                                      0.6                                                                                          0.6
                                                                                                                                                                      0.5                                                                                                                                     Control
                                                                                                                                                                                                                                                                   0.5
                                                                                                                                                                      0.4                                                                                                                                     Resilient near miss
                                                                                                                                                                                                                                                                   0.4
                                                                                                                                                                      0.3                                                                                                                                     Vulnerable near miss
                                                                                                                                                                                                                                                                   0.3
                                                                                                                                                                      0.2
                                                                                                                                                                      0.1                                                                                          0.2
                                                                                                                                                                        0                                                                                          0.1
                                                                                                                                                                            Resilient    Resilient   Vulnerable   Vulnerable   Control
                                                                                                                                                                                                                                                                    0
                                                                                                                                                                            NM weak     NM strong    NM weak      NM strong
                                                                                                                                                                                                                                                                              No          Downward CF   Neutral CF     Upward CF
                                                                                                                                        Note. NM, near miss.                                                                                                             counterfactual
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        10                                                                                                   Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS

                                                                                                                                        Figure 5                           Study 4—Percentage of Counterfactual (CF) Thoughts,        responses (the mediators) occurred in the same exper-
                                                                                                                                                                           by Cruise Decision                                         imental period. Thus, there is the possibility that there
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                                                    0.6                                                               is some reverse causality operating between how the
                                                                                                                                          Percentage of responses

                                                                                                                                                                                                                    Forgo cruise      hazard is described and the person’s decision; that is,
                                                                                                                                                                    0.5                                             Go on cruise
                                                                                                                                                                                                                                      one could decide to engage in a behavior and then use
                                                                                                                                                                    0.4
                                                                                                                                                                                                                                      that decision to shape how they characterize the situ-
                                                                                                                                                                    0.3                                                               ation, or these could coevolve. To discount this alter-
                                                                                                                                                                    0.2                                                               native, we examined the thoughts people had about a
                                                                                                                                                                    0.1                                                               hazardous situation (encoded under our different con-
                                                                                                                                                                                                                                      ditions of near-miss information) absent any decision
                                                                                                                                                                     0
                                                                                                                                                                               No        Downward CF   Neutral CF    Upward CF        about what to do. By removing the need to make a
                                                                                                                                                                          counterfactual                                              choice, we removed any potential that the assessment
                                                                                                                                                                                                                                      of the situation was based on the desire to justify a
                                                                                                                                                                                                                                      particular decision. This task has the additional bene-
                                                                                                                                           To discount the possibility that we miscoded
                                                                                                                                                                                                                                      fit of providing evidence that the biasing effect of near
                                                                                                                                        the various types of counterfactual thoughts, we
                                                                                                                                                                                                                                      misses precedes the construction of an SEU evaluation
                                                                                                                                        tested whether the different counterfactual thoughts
                                                                                                                                                                                                                                      (something assumed via our theory but not tested in
                                                                                                                                        as coded were associated with different mitigation
                                                                                                                                                                                                                                      our context). We have argued that near-miss informa-
                                                                                                                                        behaviors in reasonable ways and found that they
                                                                                                                                                                                                                                      tion changes the valence of the knowledge associated
                                                                                                                                        were. Figure 5 graphs the percentages of partici-
                                                                                                                                                                                                                                      with a type of hazard; if that is so, then we should
                                                                                                                                        pants’ counterfactual thoughts by cruise decision and
                                                                                                                                                                                                                                      expect to see different kinds of thoughts retrieved
                                                                                                                                        shows that counterfactual thoughts do influence the
                                                                                                                                                                                                                                      depending on type of near miss presented.
                                                                                                                                        decision. As would be expected, decisions to go
                                                                                                                                                                                                                                         In Study 5, we gave participants a fictitious news-
                                                                                                                                        on the cruise were associated with more upward
                                                                                                                                                                                                                                      paper article to read about cruises during hurricane
                                                                                                                                        counterfactual thoughts (upward counterfactuals ver-                                          season and then asked them to describe thoughts and
                                                                                                                                        sus other counterfactuals, • 2 415 = 2700, p < 00001),                                        feelings associated with the general category “cruises
                                                                                                                                        whereas decisions to forgo the cruise were associated                                         during hurricane season.” The task resembles Study 4,
                                                                                                                                        with downward counterfactual thoughts (downward                                               except that we removed the decision. We expected
                                                                                                                                        counterfactuals versus other counterfactuals, • 2 415 =                                       resilient near misses to be associated with more
                                                                                                                                        809, p < 0001). In sum, counterfactual thoughts, once                                         positively valenced thoughts, and vulnerable near
                                                                                                                                        evoked, can produce systematic differences in mitiga-                                         misses to be associated with more negatively valenced
                                                                                                                                        tion behavior, yet near misses do not systematically                                          thoughts. We did not make predictions about spe-
                                                                                                                                        activate any particular type of counterfactual thought.                                       cific feelings (like harm, because the person reading a
                                                                                                                                        Therefore, counterfactual thoughts do not provide a                                           newspaper article has no reason to feel any danger) or
                                                                                                                                        compelling explanation for why near misses influence                                          beliefs (e.g., hurricanes will cause damage), but rather
                                                                                                                                        mitigation decisions.                                                                         tested changes in the overall evaluations of the sit-
                                                                                                                                           Discussion. We showed that even when the situa-                                            uation (which should be guided by the information
                                                                                                                                        tion does not support updating one’s beliefs (because                                         associated with the hurricane category).
                                                                                                                                        the interaction of hurricanes and Caribbean cruises                                              Participants and Procedure. For Study 5, we col-
                                                                                                                                        are independent events), the near-miss effect still                                           lected data from 229 undergraduate business students
                                                                                                                                        operates. People who experience resilient near misses                                         who completed a number of exercises online, of which
                                                                                                                                        are more likely to ignore a hurricane warning and go                                          ours was one, in return for class participation points.
                                                                                                                                        on the cruise, whereas people who experience vulner-                                          Participants read a news story about how Caribbean
                                                                                                                                        able near misses are more likely to choose the miti-                                          cruises are deeply discounted in October and Novem-
                                                                                                                                        gation behavior, forgoing the cruise. We also showed                                          ber because of hurricane season. The story closes by
                                                                                                                                        that people with near-miss information are not revis-                                         stating that the national weather service is tracking a
                                                                                                                                        ing their calculations of the likelihood of the hazard in                                     hurricane that could impact the ship that a fictitious
                                                                                                                                        ways that might explain either their decision to go on                                        Bill Thompson is currently boarding. For the full text
                                                                                                                                        or to forgo the cruise. Finally, we showed that while                                         of the exercise, see Appendix C.
                                                                                                                                        counterfactuals are related to the ultimate choice peo-                                          Variables. For the independent variable, we used
                                                                                                                                        ple make, counterfactual thinking is not predicted by                                         same five conditions for this study as for Study 4:
                                                                                                                                        a near-miss experience.                                                                       control, resilient near miss (strong), resilient near miss
                                                                                                                                                                                                                                      (weak), vulnerable near miss (strong), and vulnera-
                                                                                                                                        Study 5                                                                                       ble near miss (weak). See Appendix C for the specific
                                                                                                                                        A potential concern with the meditational analysis of                                         wording. After reading the news article, participants
                                                                                                                                        Study 3 is that peoples’ decisions and their survey                                           wrote their general thoughts about whether or not
Tinsley, Dillon, and Cronin: How Near-Miss Events Amplify or Attenuate Risky Decision Making
                                                                                                                                        Management Science, Articles in Advance, pp. 1–18, © 2012 INFORMS                                                                                    11

                                                                                                                                        they thought fall cruises were a good idea. Partici-                 Table 4       Counts of Each Thought Type Based on Topic, Valence, and
                                                                                                                                        pants also gave two ratings on a 1–5 scale: (1) their                              Near-Miss Condition
posted on any other website, including the author’s site. Please send any questions regarding this policy to permissions@informs.org.
Copyright: INFORMS holds copyright to this Articles in Advance version, which is made available to subscribers. The file may not be

                                                                                                                                        general impression of cruises and (2) whether or not                                                                        Near-miss type
                                                                                                                                        they thought that Bill Thompson (who loves cruising
                                                                                                                                        during hurricane season) has the correct attitude.                   Thought                                       No
                                                                                                                                                                                                             type                                       near miss     Resilient   Vulnerable
                                                                                                                                           Written responses were unitized into thoughts
                                                                                                                                        (subject–verb–object). Thus, a thought could be a sen-               Fun           Valence
                                                                                                                                        tence (e.g., “Hurricanes are a big scare” has one unit).                             Negative
                                                                                                                                                                                                                                Count                     11             17            30
                                                                                                                                        There could also be multiple units in a sentence (e.g.,
                                                                                                                                                                                                                                % within near miss       4400           3708          6308
                                                                                                                                        “You save money and have fun” has the two objects
                                                                                                                                                                                                                              Positive
                                                                                                                                        and verbs with the same subject). These thoughts                                        Count                     14             28            17
                                                                                                                                        were content coded by a research assistant blind to the                                 % within near miss       5600           6202          3602
                                                                                                                                        hypotheses based on issues people seemed to think                    Safety        Valence
                                                                                                                                        about when deciding about the cruise in the prior                                    Negative
                                                                                                                                        counterfactual study. The five basic issues are fun                                     Count                      7             13            23
                                                                                                                                        (thoughts about how enjoyable or not the experience                                     % within near miss       7708           7605          7607
                                                                                                                                                                                                                             Positive
                                                                                                                                        would be; e.g., “it is not as crowded”), harm (thoughts
                                                                                                                                                                                                                                Count                      2              4             7
                                                                                                                                        about safety and getting hurt personally; e.g., “there                                  % within near miss       2202           2305          2303
                                                                                                                                        is little risk of injury”), monetary value (thoughts about           Value         Valence
                                                                                                                                        the cost/benefit ratio of the cruise; e.g., “it’s a good                             Negative
                                                                                                                                        value”), the likelihood of problems (thoughts about prob-                               Count                      6             10            15
                                                                                                                                        ability with respect to adverse events; e.g., “some-                                    % within near miss       3000           2202          3805
                                                                                                                                        thing could always go wrong”), and risk acceptance                                   Positive
                                                                                                                                                                                                                                Count                     14             35            24
                                                                                                                                        (thoughts about whether, in general, the risk/reward                                    % within near miss       7000           7708          6105
                                                                                                                                        trade-off makes sense; e.g., “why put yourself at
                                                                                                                                                                                                             Probability   Valence
                                                                                                                                        risk?”). Thoughts were also coded for whether it was                                 Negative
                                                                                                                                        attitudinally positive or negative; thus, in the same                                   Count                      4              3             2
                                                                                                                                        category of thought about the likelihood of problems,                                   % within near miss      10000           2000          1403
                                                                                                                                        a statement could be for (e.g., “I don’t mind the risk”)                             Positive
                                                                                                                                                                                                                                Count                       0            12            12
                                                                                                                                        or against (e.g., “I would not want to take the risk”)
                                                                                                                                                                                                                                % within near miss         000          8000          8507
                                                                                                                                        a cruise in hurricane season.
                                                                                                                                                                                                             Risk          Valence
                                                                                                                                           Analysis and Results. We first looked to see how                                  Negative
                                                                                                                                        our 10 codes (five topics by two valences) differed                                     Count                     15             24            20
                                                                                                                                        across conditions. As in Study 4, severity of conse-                                    % within near miss       6802           7006          8303
                                                                                                                                        quences had little effect. Only one of the 10 codes                                  Positive
                                                                                                                                        (specifically, monetary value negatively valenced)                                      Count                      7             10             4
                                                                                                                                                                                                                                % within near miss       3108           2904          1607
                                                                                                                                        reached significance, in that participants who read
                                                                                                                                                                                                                           Total
                                                                                                                                        about a strong consequence (people could have died)
                                                                                                                                                                                                                                   Count                  80            156            154
                                                                                                                                        were more likely to generate negative monetary value                                       % within near miss   10000         10000          10000
                                                                                                                                        thoughts (e.g., this cruise would not be a good value)
                                                                                                                                        than participants who read about a weak consequence
                                                                                                                                        (people could be inconvenienced or injured; p < 0001).               marginally significantly more negative-value-related
                                                                                                                                        Given the general similarities in people’s thought pat-              thoughts in the vulnerable near-miss condition than
                                                                                                                                        terns across strong versus weak consequences, we                     in the resilient near-miss condition (• 2 415 = 2065,
                                                                                                                                        collapsed across these conditions and looked at the                  p < 001, one tailed). Finally, they generated more
                                                                                                                                        influence of near-miss type (resilient versus vulnera-               negative- risk-related thoughts in the vulnerable near-
                                                                                                                                        ble versus control).                                                 miss condition than in the resilient near-miss condi-
                                                                                                                                           Table 4 shows the raw counts of each thought                      tion, although these differences were not significant.
                                                                                                                                        type based on topic, valence, and near-miss condition.                  We further tested whether near-miss information
                                                                                                                                        The overall • 2 (2) for the table was significant at 7.69            and severity affected the thoughts people had by run-
                                                                                                                                        (p = 0002), and this significance was primarily driven               ning a logistic regression using contrast coding for
                                                                                                                                        by different valence of thought across the near-miss                 conditions (Judd and McClelland 1989). As Table 5
                                                                                                                                        conditions. Participants generated significantly more                shows, the type of near miss affects the valence of the
                                                                                                                                        negative-fun-related thoughts in the vulnerable near-                thoughts associated with cruises; resilient near misses
                                                                                                                                        miss condition than in the resilient near-miss con-                  decrease the number of negative thoughts about
                                                                                                                                        dition (• 2 415 = 6024, p < 0005). They also generated               cruises, as well as increasing the ratio of positive
You can also read