Media Coverage, Forecasted Posttraumatic Stress Symptoms, and Psychological Responses Before and After an Approaching Hurricane

IMPORTANCE Exposure to disaster-related media coverage is associated with negative mental health outcomes. However, risk factors that render individuals vulnerable to this exposure are unknown. Hurricane-associated media exposure was expected to explain the association between forecasted posttraumatic stress (PTS) and adjustment after the hurricane. OBJECTIVE To examine forecasted PTS responses and media coverage as risk factors for negative mental health outcomes in the context of media coverage of an approaching disaster (Hurricane Irma). DESIGN, SETTING, AND PARTICIPANTS In a representative probability community sample of 1637 adults from Florida, respondents completed 2 online surveys: the first during the 60 hours before Hurricane Irma’s landfall (wave 1; September 8-11, 2017) and the second approximately 1 month later (wave 2; October 12-29, 2017). Poststratification weights were applied to facilitate population-based inferences. Data were analyzed from October 19 through 31, 2018.


Introduction
For coastal communities, hurricanes are increasingly common weather hazards that can cause major destruction and death. Days before a major hurricane, life is disrupted as individuals prepare their homes and evacuate if necessary. The psychological effect of these events may also be severe; exposure is often associated with posttraumatic stress (PTS) symptoms and other mental health conditions. 1,2 One study of affected populations after Hurricane Katrina found PTS prevalence as high as 30% and almost 50% prevalence for anxiety disorders. 3 However, urgent conduct of research with disaster-threatened populations is difficult, 4 so we know little about how the psychological experience of individuals anticipating a disaster may influence their subsequent responses. Given that the science surrounding global climate change anticipates increasing hurricane activity, understanding how populations at risk for hurricane exposure respond to the threat of disaster is important.
The news media is an important information source for many in the path of these storms. In the past, individuals relied heavily on local television news reports for storm-related information, 5 but online media sources now often supplement disaster reports from official sources. 6 Although information-seeking behavior may be a rational response among community members facing an evolving hazard, decades of research on media exposure to trauma suggests that extensive media consumption during a disaster event is often associated with negative consequences. Indeed, there is evidence of a link between disaster-related media consumption and negative psychological outcomes, including PTS. 7 Specifically, use of television 8 and social media 9 in the aftermath of hurricanes has been linked to increased PTS and depression. However, natural disasters account for a much smaller proportion of the literature on this topic relative to studies conducted after man-made or technological disasters. 7 Also, to date, no research on consumption of media during and shortly after an impending disaster, such as a hurricane, has been conducted.
Individuals' anticipated response to a disaster is an important factor that may influence their media consumption surrounding the threat and their subsequent responses. People often predict how they might feel in the future through a process termed affective forecasting. 10 A tendency toward negative emotional forecasts, or negative future orientation, has been associated with increased PTS 11 and psychological distress 12 in the aftermath of a community trauma. People also tend to make more negative attributions about future negative events than they do about past negative events. 13 Furthermore, pretraumatic stress, or intrusive thoughts or images about negative future events, was strongly associated with subsequent PTS symptoms in a sample of Danish soldiers from before to after deployment. 14 Together, these findings suggest that individuals' forecasted PTS responses in the days leading up to a hurricane may be associated with more negative mental health outcomes in its aftermath.
In addition, some individuals are more likely than are others to forecast greater PTS responses in anticipation of an impending disaster. Most people are not particularly accurate when predicting their future emotional responses, especially when it comes to predicting the durability of their responses to negative events. 15 However, individuals with higher levels of depression and anxiety reliably forecast more negative emotional responses to future events. 16,17 At the same time, forecasted PTS is also likely to be associated with increased exposure to hurricane-related media coverage. Prior research suggests that anxious 18 and healthy 19,20 individuals tend to orient toward stimuli that they find threatening. Similarly, through a process known as uncertainty management, individuals who are worried about a particular event may try to assuage their anxiety by seeking out information related to that event. 21 However, if individuals choose to mitigate their hurricanerelated anxiety by seeking storm-related information in the media, increased anxiety may result instead. 7 Therefore, these individuals may be vulnerable to a cycle of increased media consumption and psychological distress in the aftermath of a disaster. Thus, forecasted PTS may also be indirectly associated with outcomes through disaster-related media consumption.

The Present Study
The 2017 Atlantic hurricane season was the most active in more than a decade, producing 17 named storms, 10 of which became hurricanes. 22 These hurricanes included the first major hurricanes to hit the mainland United States in more than a decade, including Hurricane Irma, one of the strongest Atlantic Ocean hurricanes ever recorded. Hurricane Irma was a category 5 storm at its strongest but weakened to a category 3 storm before making landfall on the US mainland around 3:30 PM on September 10, 2017. The storm killed 92 persons in the United States (with 42 additional deaths in Caribbean nations) and caused approximately $50 billion of damage. 23 Media provided 24-hour sensationalized coverage, which described the possibility of "a catastrophic hit" and "worse than feared" destruction. 24 News reports featured reporters standing in high winds and rain to illustrate the dire conditions outside. 25 The media broadcasted this coverage nationally, not just locally, thus expanding the disaster's reach beyond directly affected communities.
This storm also had an uncertain path, which shifted across Florida in the days preceding landfall. Indeed, at 1 point, the entire state was threatened. This possibility presented a unique research opportunity: we studied the association between anticipated responses to an impending disaster and actual responses in its aftermath by collecting data from a representative sample of Florida residents immediately before and soon after the hurricane made landfall. This process allowed us to examine how responses to the storm evolved from before to after the hurricane across the state. We hypothesized that forecasted PTS responses to Hurricane Irma would be associated with increased hurricane-related media consumption, which in turn would be associated with poorer poststorm adjustment, after controlling for demographics, prior mental health status, and objective indicators of storm exposure. In particular, we were interested in poststorm PTS, which captures event-specific stress responses; psychological distress, which captures generalized stress responses; functional impairment, which captures mental and physical health effects on daily functioning; and worry about the future, which captures future-oriented concerns. Each of these indicators assessed a unique component of postdisaster adjustment.

Measures Demographics and Mental Health Diagnoses
Before wave 1, all panelists reported demographic characteristics (ie, age, sex, educational attainment, and ethnicity) and mental health history. Participants indicated whether a physician had ever given them a diagnosis of an anxiety or a depressive disorder; responses were coded as 0 (neither), 1 (either anxiety or depression), or 2 (both anxiety and depression).

Perceived Evacuation Zone Status
At wave 1, perceived evacuation zone status was calculated based on participants' responses to 2 questions. Participants who reported evacuating and those who believed they were in an evacuation zone were coded as 1; participants who reported not evacuating because they did not believe they were in an evacuation zone were coded as 0.

Hurricane Irma Direct Exposure
At wave 2, participants reported on a 9-item scale the degree to which they were directly exposed to Hurricane Irma. Participants could report staying in their home while under evacuation order, experiencing damage to their home or property, personal injury, or knowing someone who was injured or killed in the storm. Responses to this scale were dichotomized for analyses.

Hurricane Irma Media Exposure
At wave 2, participants reported the mean number of hours per day they spent engaging with 3 media sources "in the days during and following the recent hurricanes," including traditional media (ie, television, radio, and print news), online news sources (CNN, Yahoo, NYTimes.com, etc), and social media (Facebook, Twitter, Reddit, etc). Participants could report up to a maximum of 11 hours per day for each source; owing to the possibility of simultaneous exposure across multiple media platforms, respondents could report a maximum of 33 hours per day across all sources.

PTS Symptoms
At both waves, PTS symptoms were measured using the Primary Care PTSD (Posttraumatic Stress Disorder) Screen for Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition) (DSM-5). 27 The 5-item scale assessed the severity of symptoms corresponding to the DSM-5 posttraumatic stress disorder symptom clusters on a modified Likert-type scale ranging from 1.00 to 5.00 (with higher scores indicating more frequent symptoms). At wave 1, participants were asked: "With respect to Hurricane Irma and its aftermath, how often do you think you will experience [these symptoms] a week or two from now?" At wave 2, they were asked to report how often they had experienced these symptoms with respect to Hurricane Irma during the previous week. At both points, this scale maintained good internal reliability (wave 1 α = 0.86; wave 2 α = 0.87).

Psychological Distress
General psychological distress (ie, anxiety, depression, somatization, and anger) was measured at wave 2 using 13 items identified through factor analysis in previous studies, including 9 items drawn from the Brief Symptom Inventory 18 28

and 4 anger items from the original 53-item Brief Symptom
Inventory, 29 each assessed on a Likert-type scale ranging from 1.00 to 4.00 (greater scores indicate greater distress). The measure maintained excellent internal reliability in this sample (α = 0.92); the mean was calculated to create an index of psychological distress.

Functional Impairment
Functional impairment was assessed at wave 2 using 4 items modified from the 36-Item Short Form Health Survey. 30 These items assessed the extent to which participants' mental and physical health interfered with work and social functioning on a Likert-type scale ranging from 1.00 to 5.00 (higher scores indicated greater impairment), which maintained good internal reliability in this sample (α = 0.86).

Worry About Future Events
Worry was assessed at wave 2 using 8 items adapted from those used in prior research after September 11, 2001, 31,32 that assessed worries in the previous week about the likelihood of being exposed to natural disasters, environmental hazards, violence, and economic hardship in the future.
Scores ranged from 1.00 to 5.00, with higher scores indicating greater worry about future events.
These items maintained excellent internal consistency in this sample (α = 0.90). The mean was calculated to create a composite worry score for each participant.

Statistical Analysis
Analyses were conducted using Stata software (version 14. breakdowns were tested but did not reveal any significant differences on outcomes. Based on theoretical considerations, additional paths were drawn from prior mental health diagnoses to wave 2 PTS and from perceived evacuation zone status to hurricane-related media exposure; the latter path was not significant and was not included in the final model. Finally, an additional path from wave 2 direct hurricane exposure to wave 2 adjustment was added. Stata's structural equation model builder also allows for the inclusion of sampling weights, which were used in all models to facilitate population inferences. Analyses were conducted with and without the poststratification weights; the pattern of results remained the same. To retain sample representativeness, all statistics, including percentages reported herein, were analyzed using poststratification weights. Goodness of fit was assessed using the coefficient of determination (CD) and the standardized root mean square residual (SRMR), which are most appropriate for weighted survey data. 33 The CD is a representation of the percentage of variance in the dependent variable that the model explains and may be interpreted similarly to an R 2 value in linear regression. For the SRMR, a value of less than 0.08 indicates good model fit.  correlations ranged from r = 0.53 to r = 0.72), which was expected given that they all represent a type of negative psychological outcome. For this reason, despite the conceptual distinctness of these constructs, the 4 variables were combined into 1 latent construct of posthurricane adjustment in subsequent structural equation model analyses. All variables were also tested individually in separate path models; the patterns of responses remained identical and significant for each outcome. In the initial model, which included only the 4 observed outcome variables loading onto 1 latent variable, all factor loadings were high (Ն0.77), and model fit was good (SRMR = 0.031; CD = 0.895).

Discussion
Our findings indicate that forecasted PTS responses and storm-related media consumption before an approaching hurricane are important correlates of poststorm psychological adjustment. Forecasted PTS is also indirectly linked to poststorm outcomes via consumption of disaster-related media coverage, even when controlling for direct storm exposure. In fact, forecasted PTS responses were associated with increased media consumption, but perceived evacuation zone status was not, meaning that prestorm psychological factors appear to play a more important role in media engagement surrounding a disaster than is typically acknowledged. Given that this media engagement during a disaster has been associated with negative psychological consequences and downstream implications for physical health, 34 understanding the predictors associated with this behavior is of particular importance. Furthermore, it appears that individuals' prestorm vulnerability to distress, as measured by forecasted PTS, increases media exposure, with significant implications for postdisaster adjustment.
These results are also important because they represent the first attempt by researchers to analyze how prestorm psychological factors are associated with subsequent adjustment through prospective analyses with surveys fielded in the days leading up to a hurricane. Because we are not relying on retrospective reports of participants' psychological functioning or storm perceptions, which can be influenced by situational factors 35 or degrade over time, 36 we can be more confident in the ecological validity of our findings. Furthermore, the study design, which involved sampling from within a statewide panel and the use of sampling weights to adjust for probability of inclusion into the study, enables us to extrapolate from these findings to make population inferences.

Future Directions
Several questions remain unanswered. For example, because the use of online social media for updates during a developing crisis is associated with greater distress responses relative to other media sources, 37 perhaps greater social media use during an approaching hurricane has a stronger association with poststorm outcomes when compared with traditional media. Poststorm responses to media may also be differentially associated with prestorm psychological projections. Preliminary analyses of our data suggest this is not the case; however, this association should be examined using stronger measures of media exposure, ideally in real time.

Limitations
Despite use of a statewide panel and population weights, we acknowledge that the present sample is not necessarily representative of Florida residents. GfK sent invitations to participate in the wave 1 survey to all their Florida panelists to capture as much data on Floridians' responses as possible.
However, this sampling design precluded our ability to oversample in harder-to-recruit populations.
Furthermore, the panel is designed to recruit samples that are demographically representative of the populations from which they are drawn, but this does not include geographic representation within smaller communities. As a result, the geographic distribution of participants in our sample does not necessarily mirror that of Florida. This is important for studies of natural hazards because the geographic distribution of the sample may not be representative in terms of population hurricane exposure, objectively via strong winds and storm surge as well as subjectively via local media reports.
Sampling weights can correct for discrepancies between the sample and census benchmarks, but it would be helpful for future research to improve geographic representation as well.
We also acknowledge the possibility of a Hawthorne effect in our sample, such that asking individuals at wave 1 to attend to their expectations for future distress may have amplified reports of distress at wave 2. This effect can be a concern in longitudinal survey research because participants' continued participation in surveys introduces the possibility that their previous responses may affect subsequent behavior. However, in this case, we see no indication of the Hawthorne effect occurring.
When comparing participants who did not participate in wave 1 with those who did using bootstrapped t tests, there are no differences in wave 2 adjustment (P > .05 for all indices). As such, we can assume that the deficits in psychological adjustment over time are unlikely to be attributable to altered attention to anticipated distress responses.

Conclusions
Our results have important implications for the news media and emergency management and public health officials. That prestorm psychological factors have a stronger association than perceived evacuation zone status or direct hurricane exposure with storm-related media consumption and subsequent adjustment suggests a need to improve hurricane-related risk communications for the public. Communicating a hazard-specific appropriate level of risk could mitigate this concern by