University of Southern California Dissertations and Theses

Psychological distance in the public’s response to terrorism: An investigation of the 2016 Orlando nightclub shooting

(USC Thesis Other)

Psychological distance in the public’s response to terrorism: An investigation of the 2016 Orlando nightclub shooting

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Running head: PSYCHOLOGICAL DISTANCE 1

Psychological Distance in the Public’s Response to Terrorism: An Investigation of the 2016
Orlando Nightclub Shooting

Matt Baucum
Department of Psychology
Master of Arts, Psychology
University of Southern California
August 15, 2018

PSYCHOLOGICAL DISTANCE 2

Table of Contents
ABSTRACT 3
INTRODUCTION 4
STUDY 1: METHODS 11
STUDY 1: RESULTS 17
STUDY 1: DISCUSSION 21
STUDY 2: METHODS 23
STUDY 2: RESULTS 34
STUDY 2: DISCUSSION 39
GENERAL DISCUSSION 44
REFERENCES 46
TABLES 51
FIGURES 55

PSYCHOLOGICAL DISTANCE 3

ABSTRACT
Individuals experience and react to disaster events at varying levels of psychological
distance; disasters can be near or far, recent or long-passed, and impact those who are similar or
dissimilar to the self. Using the high-profile Orlando shooting of 2016 as a case study, we
employ both a panel sample and a social media dataset to study how United States residents’
expressions of fear and risk perception in the aftermath of the attack varied based on their
physical, temporal, or social distance from the attack (as defined by Construal Level Theory),
and whether such effects were mediated by the use of concrete language when describing the
attack. Survey data suggested that the Orlando shooting had some effect on respondents’ self-
reported risk perception, but with no moderating effects of psychological distance, while
expressions of fear on social media did depend on users’ temporal, physical, and social distance
from the attack. Physical and temporal distance also predicted the use of more concrete language
when describing the shooting, with the most concrete language used by those who lived close to
Orlando and posted soon after thee attack’s occurrence. Interestingly concrete language use and
fear expressions were negatively correlated in the social media sample. We discuss implications
for how Construal Level Theory and text analytic methods can shed light on the mechanisms of
public risk perception.
PSYCHOLOGICAL DISTANCE 4

INTRODUCTION
Understanding public reactions to highly publicized terror attacks is crucial to appraising
the overall risk that terrorism poses to society. Terror attacks have the potential to cause
psychological and economic damage that can far outlast their immediate effects (Blalock,
Kadivali, & Simon, 2009; Fairbrother et al., 2003; Galea et al., 20 03; Marshall et al., 2007), and
it is thus important that researchers and public officials can anticipate the contours of such
effects as best as possible. Of course, as with any other public risk or hazard, public reactions to
any one specific terror attack will vary widely between individuals and between groups, and it is
worth investigating the mechanisms and processes that govern why certain groups and
individuals react the way that they do in the wake of salient terrorist events.
One potentially relevant factor in predicting an individual’s response to a terror event is
how psychologically “close” or “distant” the individual feels from the attack. This notion of
“psychological distance” stems from Construal Level Theory (Trope & Liberman, 2010), which
stipulates that when individuals process any event, idea, or phenomenon that is removed from
their direct and immediate experience, their “psychological distance” from the event is
determined across four dimensions: spatial distance (how near or far the event is), temporal
distance (how far in the past/future the event takes place), social distance (how similar/familiar
one feels with those involved), and hypotheticality (how likely the event is to take place). It has
been shown that these seemingly disparate types of distance may share a common neural locus in
the intraparietal lobule (Parkinson, Liu, & Wheatley, 2014), and that each of the distance
dimensions has a similar effect on whether the event/phenomenon in question is construed in an
abstract or concrete manner. Events that are more proximal are generally construed concretely
PSYCHOLOGICAL DISTANCE 5

(for instance, “a mass shooting”), while more distal events tend to be construed abstractly (e.g.,
“an attack on our country”; Bar-Anan, Liberman, & Trope, 2006; Trope & Liberman, 2010).
The psychological distance framework has clear implications for the study of public
terror reactions. It is reasonable to expect that an individual’s or group’s reactions to a terror
event will be heightened for attacks that are recent (temporal distance), occur nearby (physical
distance), or target victims who are perceived as similar to the self (social distance). Indeed,
many researchers have specifically investigated the interplay between psychological distance and
the psychological effects of terror attacks. Reactions to the September 11
th
terrorist attacks have
been shown to be most severe for those who lived closest to New York City). Paris residents’
negative reactions to the coordinated shootings that occurred in 2015 quickly diminished in the
first week after the attack’s onset (i.e., increasing temporal distance; Lin, Margolin, & Wen,
2017). Lin and Margolin (2014) even found that city-wide levels of fear after the Boston
Marathon Bombings were best predicted by the frequency with which its residents visited Boston
(which could be conceptualized as a form of “social distance”). Thus, existing literature suggests
that the dimensions of psychological distance (with the possible exception of “hypotheticality”,
which is arguably not relevant for events that have already occurred) can impact public terror
response in the expected direction.
The present study seeks to address two limitations in past work on psychological distance
and public risk perception. First, no studies have employed Construal Level Theory to examine
the effects of multiple psychological distance dimensions on public disaster reactions in a single
study; existing studies have either focused on single distance dimensions (e.g., Lin, Margolin, &
Wen, 2017) or simply employed other theoretical frameworks (Cardeña et al., 2005; Gaissmaier
& Gigerenzer, 2012; Lin & Margolin, 2014). Thus, there have been no studies examining 1) the
PSYCHOLOGICAL DISTANCE 6

relative effects of different psychological distance dimensions, controlling for one another, or 2)
potential interactions between different forms of psychological distance. Second, it is unclear
whether the effects of psychological distance on public reactions depend on abstract/concrete
construal as a mediating process (which is a central tenet of Construal Level Theory); spatial,
temporal, and social distance from an attack might simply impact public reactions through
mechanisms other than abstract/concrete construal. Magee, Milliken, and Lurie (2010)
specifically investigated how individuals’ social power (a hypothesized form of social distance),
such as their occupational status, predicted their use of abstract or concrete language in the
aftermath of the September 11
th
attacks, and found that higher-power individuals tended to use
more abstract language that lower-power individuals. Yet no other studies have investigated
event construal in the wake of terror attacks, and how such construal might mediate the
relationships between psychological distance variables and other reaction variables (such as risk
perception, negative affect, etc.). The present study aims to fill both of these research gaps by 1)
investigating the impacts of multiple psychological distance dimensions (spatial, temporal, and
social) on public terror reactions, as well as their interactions, and 2) whether such effects can
plausibly be mediated by individuals’ abstract or concrete construal of the event itself. In doing
so, this study will serve as the most comprehensive investigation of the impact of psychological
distance on real-world public terrorism responses to date. Furthermore, by studying
abstract/concrete construal as a mediating process, it will serve to clarify whether some of the
psychological distance effects seen in previous literature (effect of proximity, diminishing
reactions over time, etc.) can truly be explained from a Construal Level Theory standpoint.

PSYCHOLOGICAL DISTANCE 7

Hybrid Sample Approach: Traditional Panel and Social Media Data
To fully investigate the impacts of psychological distance on public terror response, and
whether such effects are mediated by abstract and concrete construal, we employ a two-sample
approach that draws on both “traditional” survey data as well as a social media sample, both
focusing on United States residents’ reactions to the 2016 shooting at the Pulse Nightclub in
Orlando, Florida. This attack was the deadliest mass shooting in the United States at the time of
its occurrence, which we classify as a “terror attack” for the purposes of this study given that the
shooter had personally pledged allegiance to the terror group ISIS (Islamic State in Iraq and
Syria) before carrying out the assault. Furthermore, the Orlando shooting is prototypical of the
kinds of “soft-target” terror attacks that have come to dominate the terrorism landscape in recent
years (Barnett, 2015). We specifically choose to focus on this attack because it occurred during
data collection for the terrorism-focused panel survey analyzed here, providing subsamples of
participants who responded to the same items before and after the attack’s occurrence and
allowing us to examine the attack’s impact on responses.
We use this survey data to test hypotheses regarding the impact of psychological distance
dimensions (spatial, temporal, social) on 1) terrorism-related fear, and 2) terrorism-related risk
assessment. We then attempt to replicate the results from the survey on a sample of Twitter posts
made in the days following the Orlando shooting, and investigate whether such effects were
mediated by the level of abstractness or concreteness in users’ language. Other researchers have
used similar hybrid research approaches (e.g., Dehghani et al., 2016; Hoover, Johnson, Boghrati,
Graham, & Dehghani, 2017) as a way of capitalizing on the strengths of both “real-world”
language data (via social media) and more controlled, traditional research methods. The use of a
panel survey allows us to employ simple, self-report measures of dependent variables, and
PSYCHOLOGICAL DISTANCE 8

allows us to benefit from explicitly queried demographic and location data for each participant.
Meanwhile, the use of social media sample allows for a real-world measurement of
abstract/concrete construal, which is a difficult construct to measure via self-report (thus, why it
is typically manipulated or measured with non-survey based methods, such as implicit
association tests; Bar-Anan, et al., 2006; Magee et al., 2010; Snefjella & Kuperman, 2015).
Hypotheses
In line with past research on the role of time (Lin et al., 2017), spatial distance
(Gaissmaier & Gigerenzer, 2012; Cardeña et al., 2005; Fischhoff, Gonzalez, Small, & Lerner,
2003), and various forms of social distance (Bourque, Mileti, Kano, & Wood, 2012; Lin &
Margolin, 2014) in shaping public disaster response, we expect psychological distance across
these three dimensions to negatively predict self-reported fear and risk perceptions after the
Orlando shooting. In line with Construal Level Theory, we also expect lower values of
psychological distance to predict the use of more concrete language when discussing the attack
on social media, and higher values to predict abstract language. Lastly, we hypothesize that
construing the attack in a concrete manner will heighten individual fear and risk perceptions,
such that any relationships between the psychological distance dimensions and fear/risk should
be at least partially mediated by abstract/concrete construal. In addition to testing for the main
effects of all psychological distance variables, we also test for interactions between each variable
with the elapsed time since the attack (temporal distance). Past work suggests that negative
sentiments quickly diminish in the days following small-scale terror attacks (Lin et al., 2017),
and it is possible that such attacks are only “top of mind” for the public in a relatively short
window after their occurrence. If this is the case, the effects of other psychological distance
predictors might be most pronounced in the days immediately following the Orlando attack, but
PSYCHOLOGICAL DISTANCE 9

diminish in magnitude as time passes. For study 1 (panel survey data), we thus make the
following hypotheses:
• H1: Greater social, physical, and temporal distance from the Orlando shootings will
predict greater self-reported fear regarding the threat of terrorism (along with interactions
between temporal distance and the other predictors).
• H2: Greater social, physical, and temporal distance from the Orlando shootings will
predict self-reported risk perceptions of future terror attacks. (along with interactions
between temporal distance and the other predictors).
For study 2 (social media dataset), our hypotheses are as follows:
• H3: As in H1, greater social, physical, and temporal distance from the Orlando shootings
will predict fear-related language in posts discussing the attacks (along with interactions
between temporal distance and the other predictors).
• H4: Greater social, physical, and temporal distance from the Orlando shootings will
predict the use of more abstract/less concrete language in posts discussing the attack
(along with interactions between temporal distance and the other predictors).
• H5: The use of abstract/concrete language in posts discussing the attack will at least
partially mediate any effects found in H3.
Psychological Distance Variables
Spatial and temporal distance can be simply defined in the context of a terror attack: each
respondent’s physical distance from where the attack took place, and the elapsed time between
the attack and collection of their data (survey or social media). We choose not to study the
“hypotheticality” dimension of psychological distance due to the difficulty of applying it to
already-past events. Social distance is a more ambiguous construct, so we conceptualize it in
PSYCHOLOGICAL DISTANCE 10

multiple ways to more fully capture the construct’s potential effect on our dependent variables of
interest.
Because the Orlando shooting occurred in an urban setting, it is reasonable to expect that
urban residents would feel more socially close with those affected than rural residents. Thus, the
population density of each respondent’s city of residence partly serves as a social distance
measure (with higher values indicating less distance from the attack victims), with the
assumption that greater population densities correspond to more urbanized areas. Furthermore,
because the Orlando shooting victimized a specific minority community (the LGBT community),
we also hypothesized that social closeness with the LGBT community would predict
respondents’ reactions to the attack. While this construct was not measured directly (and would
be difficult to estimate for the sample of Twitter users), we assumed that, on average,
participants who identified as relatively conservative would feel less socially close to the LGBT
community than participants who identified as relatively liberal, and thus employ political
orientation as a proxy construct. Lastly, research has shown that status and power serve to
socially distance individuals from others (Magee et al., 2010; Magee & Smith, 2010), with
higher-power individuals more likely to construe events abstractly and lower-power individuals
more likely to construe them concretely. Thus, in the survey and social media samples, we use
respondents’ income levels as a proxy for social power and use this as a third conceptualization
of social distance from the attack (with higher-income individuals assumed to occupy positions
of greater power or influence, and thus greater social distance the terror event).

PSYCHOLOGICAL DISTANCE 11

STUDY 1: METHODS
Study 1 employed a panel survey to investigate whether various psychological distance
dimensions impacted respondents’ reactions to the 2016 Orlando shootings in the direction
predicted by Construal Level Theory. Since the panel survey did not explicitly measure
abstract/concrete event construal (a difficult variable to measure via self-report), the mediating
role that construal plays in these effects is addressed in study 2. For study 1, we anticipated that
those who were psychologically closer to the Orlando attack would experience greater self-
reported fear regarding terrorism, and would see future terror attacks as more likely (given that
psychological closeness has been shown to increase subjective probability estimates; Wakslak &
Trope, 2009). In this way, we attempt to capture the impact of psychological distance on
participants’ affective (fear) and cognitive (likelihood/risk perceptions) reactions to the event.
Participants
Participants were contacted via a panel run by the University of Oregon’s Decision
Research institute and asked to participate in a survey regarding their attitudes towards terrorism.
The panel includes 1669 residents of the United States and Canada, originally recruited through
various online advertisements. The 1508 panelists who reside in the United States were
originally asked to participate in a six-wave longitudinal study on their attitudes towards
terrorism (though we only focus on the first wave of data collection, during which the Orlando
nightclub shooting took place). There were 700 members who participated in the first survey
wave, of whom n=607 had non-missing values for our dependent and predictor variables of
interest (62.8% female, median age 42). Of these n=607 participants, 98.5% had graduated high
school, 75.9% had completed some college or vocational training, 45.9% had graduated college,
and 14.8% had more than a college degree.
PSYCHOLOGICAL DISTANCE 12

Procedure
Panel members were contacted and asked to participate in a six-wave survey assessing
their thoughts and attitudes towards terrorism, which took place entirely online. We focus
entirely on data collected from the first survey wave, which was open to online participation
from June 10-16, 2016. The Orlando nightclub shooting occurred during this data collection
period, on the early morning of June 12, 2016; by then, 370 participants (61.0%) had already
taken the wave 1 survey, while the remaining 237 users (39.0%) had not yet taken it. To confirm
that the Orlando shooting was salient enough to viably impact respondents’ fear and risk ratings,
respondents were asked at wave 2 whether they remembered the shooting, with 97.8% of our
sample (with non-missing data for wave 2) indicating that they did.
Materials
Dependent variables. The original survey included many items not analyzed here, and
we thus report only those survey items directly relevant for our analytical purposes. For
dependent variables, we chose to focus on both affective and cognitive reactions to the terror
attacks, and thus analyze items regarding terror-related fear/anxiety and terror-related risk
perceptions as dependent variables. Negative affect was measured with three items asking
participants how often they had felt “anxious”, “worried”, or “frightened” about terror attacks
over the past six months; we specifically chose to measure fear due to research suggesting that
other negative emotions, such as anger or sadness, can have conflicting effects with fear in a risk
perception context (see Lerner & Keltner, 2000; Lerner, Gonzalez, Small, & Fischhoff, 2003).
Response options for the fear items were “never”, “not very often”, “sometimes”, “very often”,
and “I don’t know” (note that there were only five “I don’t know” responses in the dataset, which
we counted as missing for simplicity).
PSYCHOLOGICAL DISTANCE 13

To measure cognitive perceptions of future risk, we asked participants how likely it was
that terrorists would carry out either an “armed attack on civilians” or detonate a “bomb in a
public place” sometime over the next six months. These specific attacks were chosen to
operationalize terrorism risk perceptions as they are the most common methods used in “soft-
target” terror attacks (low-tech attacks on vulnerable public targets) such as the Orlando
nightclub shooting (in contrast with rarer, higher-consequence terror methods such as nuclear
detonations or airplane hijackings). Response options for both attack types were “not likely”,
“slightly likely”, “somewhat likely,” and “very likely.”
We split the data into subsamples of participants who completed the wave 1 survey either
before or after the Orlando shooting, and subjected each subsample to separate longitudinal
confirmatory factor analyses for the risk and fear constructs. This analysis uses data from all six
survey waves to ensure that the items consistently measure the same latent construct at all time
points. We chose to model the before/after Orlando subsamples separately for the sake of sample
homogeneity, since we hypothesized that the Orlando attack would impact respondents’ levels of
risk perception and fear (there were n=305 complete cases for the “before Orlando” subsample
and n=159 complete cases for the “after Orlando” subsample). For both the fear and risk
constructs, we fit models that assumed equal item loadings and equal item intercepts across all
waves (known as “strong measurement invariance”), which fit the data well (all CFI<0.999; all
RMSEA<0.055; all SRMR<0.04) and produced loadings greater than 0.8 (with standardized
latent factors). These models suggest that our dependent variables of interest were well-captured
by these items throughout the course of the survey. Having established this, we fit polytomous
Item Response Theory models (for fear/anxiety and terrorism risk perception) to participants’
responses and use their standardized factor score estimates for all further analyses.
PSYCHOLOGICAL DISTANCE 14

Spatial distance measure. Respondents were asked to provide their home ZIP code; the
distance between this ZIP code and the ZIP code of the Pulse nightclub in Orlando (the location
of the shooting, ZIP code 32806) was used as a measure of each respondent’s spatial distance
from the terror attack.
Temporal distance measure. For respondents who completed the survey after the
Orlando shooting, we coded their temporal distance from the attack as the number of days
(including partial days) between the attack and survey completion. We counted the attack as
occurring at 2:06 a.m. Eastern Time on June 12, 2016, the time at which police were first
notified of the incident (Stapleton, 2016).
Social distance measures. As aforementioned, social distance was conceptualized across
multiple dimensions, each representing an aspect of how participants might be able to identify
with the victims of the nightclub shooting.
Urban/rural residence. The population density of each respondent’s city of residence (as
determined by their self-reported ZIP code) was used to measure social distance in terms of
urban/rural residence, with the assumption that greater population densities would correspond to
more urbanized areas (and thus, less social distance from the attack, which took place in an urban
area). Population density data was obtained from the United States Census Bureau (2010).
Social closeness with LGBT community. As aforementioned, the survey did not
specifically include items regarding participants’ attitudes towards the LGBT community, and
instead employ political orientation as a proxy construct. Conservatives generally demonstrate
less tolerant attitudes towards the LGBT community than liberals (partly due to conservatives’
increased focus on moral ‘purity’; Graham, Haidt, & Nosek, 2009), and political conservatism
has been shown to partially mediate the relationship between religiosity and attitudes towards
PSYCHOLOGICAL DISTANCE 15

homosexuality (Hill, Moulto n, & Burdette, 2004). We thus assumed that conservatives would
generally have less contact with members of the LGBT community than liberals, and used a 7-
point political ideology item (ranging from “extremely conservative” to “extremely liberal”,
coded from -3 to 3) as a proxy for social distance from the LGBT community. Note that there are
many reasons why political orientation might correlate with responses to the Orlando shooting
other than its supposed correlation with closeness to the LGBT community; thus, this specific
variable is a liberal test of our social distance hypotheses that should be interpreted with caution.
Social power. We chose to operationalize power as each participant’s income level; while
not a perfect proxy for power, those with higher incomes are more likely than those with lower
incomes to possess authority and influence in their professional lives, which has been shown to
correlate with more abstract construal of events (Magee et al., 2010). Furthermore, income is an
easily measured variable with an interpretable quantitative basis. The survey item measuring
annual income included 11 categories: less than $10,000, $10,000 to $14,999, $15,000 to
$24,999, $25,000 to $29,999, $30,000 to $39,999, $40,000 to $49,999, $50,000 to $59,999,
$60,000 to $74,999, $75,000 to $84,999, $85,000 to $99,999, and greater than $100,000. Values
on this item were coded as the lowest value included in that income bracket (given that the
highest income category was unbounded), and thus represents each respondents’ minimum self-
reported income.
Analysis
Study 1 uses Bayesian regression to test for reliable effects of the psychological distance
variables on self-reported fear and risk perceptions. Bayesian regression produces a posterior
probability distribution over each coefficient of interest, which simplifies the interpretation of
each effect’s magnitude. Bayesian regression also allows for model comparisons in terms of
PSYCHOLOGICAL DISTANCE 16

“Bayes factors” (the ratio of the likelihood of the sample data under two competing models),
which can quantify the strength of evidence for null models in the absence of any reliably
nonzero effects (note that this contrasts with traditional null hypothesis methods, which cannot
support null models over alternative models). All regression coefficients were assigned standard
normal priors; this type of prior specification is desirable in the case of many predictor variables,
as it shrinks the estimated regression coefficients towards zero (thus avoiding inflated estimates
for coefficients due to the sheer number of estimations being made). Interaction terms were set
with a smaller prior standard deviation of 0.5, given that interaction terms are typically smaller
than main effects.

PSYCHOLOGICAL DISTANCE 17

STUDY 1: RESULTS
Descriptive Statistics
Of the n=607 respondents, 370 (61.0%) completed the survey before the Orlando attack,
with mean (standardized) fear and terrorism likelihood scores of -0.04 (SD=1.03) and -0.10
(SD=1.00), respectively. The other 237 respondents (39.0%) completed the survey after the
attack, and reported mean (standardized) fear and terrorism likelihood scores of 0.06 (SD=0.95)
and 0.16 (SD=0.99). Across all participants, fear and terrorism likelihood ratings were correlated
at r=0.35. On average, respondents were slightly left of center in their political views (M=-0.27,
SD=1.44), lived 1144.6 miles away from the location of the Orlando shooting (SD=657.6), and
lived in cities with population densities of 4569 people per square mile (SD=5714.1). Note that
the initial regression results demonstrated no evidence of excessive multicollinearity between
predictors (all VIF<10). Table 1 presents bivariate correlations between each of the
psychological distance variables, and confirms the lack of severe multicollinearity between the
predictors.
Bayesian Regression Analyses
Before constructing a regression model, we confirmed that all variables were
approximately symmetrically distributed and that there were no serious deviations from linearity
in the bivariate plots. Population density demonstrated considerable positive skewness, and was
thus log-transformed for further analyses. This transformation also reflects the theoretical
assumption that, from a social distance standpoint, differences in population density on the lower
end (distinguishing urban from rural areas) likely matter more than similar-sized variations at the
high end (distinguishing between urban areas of different density). Take, for instance, Galena,
AK (a small town with population 5,700), Dallas, TX, and New York, NY, each with population
PSYCHOLOGICAL DISTANCE 18

densities of 431, 3,518 and 27,012 people per square mile, respectively. Using the raw value for
population density would assume that the difference in “urban-ness” between Dallas and New
York (both major cities) is roughly 7.5 times the difference in “urban-ness” between Dallas and
Summertown, while employing log-scaled population density sets these differences as
approximately equal, which seemed to better represent the distinction between urban and rural
areas.
Thus, the following predictors were entered into separate Bayesian regression models
predicting respondents’ self-reported negative affect and terrorism likelihood ratings:
• Before/after Orlando attack (Before=-1, After=1)
• Distance from Orlando nightclub (standardized)
• Log-transformed residence population density (standardized)
• Political Orientation (-3 to 3, standardized)
• Income (lowest value of income bracket; standardized)

Since our hypotheses specifically deal with whether psychological distance predicted
reactions to the Orlando shooting (which in turn are represented by the main effect of the
before/after Orlando contrast), each predictor was entered as an interaction term with the
before/after Orlando contrast. We also entered gender (Female=-1, Male =1) as a main effect,
given that males and females generally demonstrate differing levels of risk perception (Finucane,
Slovic, Mertz, Flynn, & Satterfield, 2000; Flynn, Slovic, & Mertz, 1994).
For both fear/anxiety and ratings, neither the main effect of the Orlando attack nor its
interactions with any of the psychological distance predictors were reliably nonzero (the middle
95% of each coefficient’s posterior density included zero). The Orlando attack did cause a
PSYCHOLOGICAL DISTANCE 19

reliable increase in risk ratings ( β=0.13; all 4000 posterior sample draws lied above zero), such
that ratings of future terror attack likelihood increased by roughly 0.13 standard deviations after
the attack. However, none of the interactions between the Orlando attack contrast and the
psychological distance predictors were significant, suggesting that psychological distance did not
generally moderate individuals’ responses to the attack.
To investigate how strongly our data supported the lack of an effect of psychological
distance, we examined the Bayes factor comparing this model to a main-effects-only model
(which did not specify any interactions between the Orlando attack and psychological distance
predictors). The main effects only model was supported by Bayes factors of >1000 for predicting
both fear and terrorism likelihood ratings, strongly suggesting that the impact of the Orlando
attack did not depend on individuals’ physical or social distance from the attack (as
operationalized here).
We also re-ran the regression analyses only on those participants who responded to the
survey after the shooting, to examine the role of the elapsed time between the attack and each
response. For this model, we specified main effects for each of the psychological distance
predictors, gender, and time since the attack, as well as all two-way interactions between time
and the other psychological distance predictors, but found no reliably nonzero effects predicting
fear/anxiety or terrorism likelihood ratings.
In the whole-sample analysis, gender did have a reliably negative impact on fear/anxiety
( β=-0.24), such that men’s factor scores were 0.24 SD lower than those of women, respectively.
Population density (log-transformed) also had a reliably positive impact on fear/anxiety ratings
( β=0.16); translating this into original measurement units suggest that respondents’ fear/anxiety
factor scores increased by 0.16 SD for every tripling of residence population density. However,
PSYCHOLOGICAL DISTANCE 20

neither of these effects (of gender or population density) pertained to respondents’ reactions to
the Orlando attack, but merely described their overall attitudes and emotional reactions to the
threat of terrorism (both before and after the attack).

PSYCHOLOGICAL DISTANCE 21

STUDY 1: DISCUSSION
The hypothesized variables had surprisingly little predictive value for respondents’ self-
reported fear and risk perceptions. The main effects of population density and gender were in the
expected directions (greater fear of terrorism for those in urban areas, where attacks are more
likely to take place, and higher risk perceptions and fear/anxiety for females; see Finucane,
Slovic, Mertz, Flynn, & Satterfield, 2000; Flynn, Slovic, & Mertz, 1994), but do not suggest that
these variables played a role in respondents’ reactions to the Orlando attack specifically. In fact,
a model that omitted all effects of psychological distance on reactions to the Orlando attack
was strongly supported over the original model, providing firm evidence that psychological
distance played little role in the attack’s effect on survey responses.
From a Construal Level Theory standpoint, this finding might simply mean that
respondents did not construe the Orlando attack differently based on their psychological distance
from it, and subsequently provided similar fear and risk ratings regardless of psychological
distance. It is also possible that psychological distance did play a role in respondents’
abstract/concrete construal of the attack, but that their construal level did not impact their self-
reported fear or likelihood estimates. While this question cannot be addressed with our survey
data alone (since it lacks any measure of abstract/concrete event construal), the analysis of social
media data (study 2) can provide a better picture of how individuals at varying psychological
distances from the attack construed it in an abstract versus concrete manner.
Of course, we acknowledge that reactions to terror attacks can take many forms,
including avoidance behaviors and policy attitudes, that span beyond the dependent variables
measured here. It was an implicit assumption of our method that differences in individuals’
responses to the attack would likely stem from either an affective reaction to the event, or a
PSYCHOLOGICAL DISTANCE 22

cognitive reappraisal of future risk. While it is possible that other variables might have shown
reliable effects where our chosen dependent variables did not, our analysis focused on what
seemed to be two of the most psychologically “simple” ways to measure reactions to a terror
attack: how fearful it made people, and whether they thought it was likely to happen again. By
these measures, such reactions were largely independent of psychological distance.

PSYCHOLOGICAL DISTANCE 23

STUDY 2: METHODS
The purpose of study 2 was to investigate whether the lack of effects from study 1
replicated in a natural language, social media setting. Furthermore, because study 2 focuses on
text rather than survey responses, it also allows for the measure of abstract and concrete
construal through text analytic techniques, to study the mediating role of abstract and concrete
construal on any effects of psychological distance. After collecting a dataset of Twitter posts
made in response to the Orlando shooting, we investigate whether fear-related language in the
posts correlates with the same psychological distance predictors in study 1. We then test 1)
whether concrete/abstract language in the posts correlates with psychological distance predictors,
and 2) whether posts with more fear-related terms tended to use more concrete language. Note
that, because we focus on Twitter posts that specifically reference the Orlando attack (through
the use of hashtags), there was no pre-attack data available for comparison as there was in study
1. Thus, study 2 focuses on whether post-attack sentiments varied according to users’
psychological distance.
Sample
We collected a sample of Twitter posts made between June 11, 2016 and June 19, 2016
(the week following the Orlando nightclub shooting) that included one or more of the hashtags
“#OrlandoShooting”, “#Orlando”, or “#pulseshooting”; this criteria was necessary given that we
were only interested in users’ language that directly related to the terror attacks (and initial web
search suggested that these were the most common hashtags used on social media to refer to the
event). The original search yielded over 4 million posts.
We excluded from the sample all retweets (those that were written by one user and
simply re-posted by another) and posts that only contained hashtags or links. Posts had to be in
PSYCHOLOGICAL DISTANCE 24

English (see automatic language detection function in R package “cld2”; Ooms, 2017), authored
by non-verified Twitter accounts (where “verified” accounts are official accounts for
organizations or celebrities), and posted by users in the United States for whom location data (at
the city level) was available. Where there were multiple posts written by the same user, we
include only their earliest post, and ensured that all posts were made after the onset of the
Orlando shooting (defined as 2:06 a.m. Eastern Time, June 12, 2016). Further inspection of the
data revealed some tweets posted by news sites (rather than individuals) that were not screened
out with the original criteria, which were subsequently removed. Filtering the dataset by these
criteria yielded a final sample of 36,259 posts.
Procedure
Text preprocessing. As social media data is often quite noisy, we implemented multiple
cleaning steps to convert each Twitter post into a readable text object for further analysis. Each
post was stripped of all non-punctuation/non-alphanumeric characters (which removes special
characters such as Emojis), as well as all links to other content (such as webpages or pictures).
Because some users often use hashtags as parts of their post’s syntax (e.g., “Our thoughts are
with the #pulseshooting victims”), we chose to retain all hashtags that were directly followed by
non-hashtag words while removing all others. Thus, hashtags embedded in the middle of
sentences are assumed to serve some grammatical function and are kept as part of the post’s
content, while those that appear at the end (where users often place multiple hashtags in a row)
are removed. Thus, the tweet “Our thoughts are with the #pulseshooting victims #Orlando
#OrlandoShooting” would be shortened to “Our thoughts are with the pulseshooting victims.”
While these procedures cannot guarantee that each post will perfectly reflect the semantic
PSYCHOLOGICAL DISTANCE 25

content intended by the author, it reduces much of the noise introduced by social media
conventions.
Temporal distance measure. For each post, temporal distance from the Orlando attack
was measured as in study 1: the number of days (including partial days) between 2:06 a.m. ET
on June 12, 2016 and the date/time at which the post was created.
Physical distance measure. Physical distance was measured as the number of miles
between the latitude and longitude coordinates of the Pulse nightclub and the central
latitude/longitude coordinates of each user’s nearest ZIP code, which was retrieved through the
Bing Maps API. Because not all users choose to report their city of residence in their profile (or
might simply include broader location information, such as state or country), location data was
only retained for users in the United States where the API could identify a single location profile
at the city level (i.e., those returning multiple possible matches, or returning larger geographic
units such as states or counties, were excluded). The central latitude/longitude coordinates for
each user’s API-returned city location were then matched to the nearest U.S. ZIP code, and we
calculated the distance between this ZIP code and the Pulse nightclub using the “Imap” R
package (Wallace, 2012) and ZIP code location data from the United States Census Bureau
(United States Census Bureau, 2017).
Social distance measures. We operationalized social distance in study 2 in a similar
manner to study 1.
Political Orientation. As in study 1, we assumed that more conservative Twitter users
would (on average) feel more socially distant from the LGBT community than more liberal
users. While political orientation is not directly available from Twitter profiles, we employ a
method introduced by Golbeck and Hansen (2011) to estimate the political sentiments of each
PSYCHOLOGICAL DISTANCE 26

user. In this procedure, Golbeck and Hansen (2011) estimated the average political ideology of
the users who followed various Twitter news accounts (e.g., @foxnews, @latimes), based on the
Congress members that their audiences also followed. We use these ideology estimates for 20
news accounts to estimate each participants’ political orientation as the average ideology score of
all of the news accounts they follow. Table 2 presents the 20 news accounts used for estimation,
and their political ideology score ranging from 0 (conservative) to 100 (liberal).
Income. As in study 1, we employ estimated annual income as a proxy construct for
social power (with higher-income individuals assumed, on average, to hold professional
positions of greater power or influence). Culotta, Kumar, and Cutler (2015) identified the Twitter
accounts that best predicted the annual income of the users that followed them; we apply the
regression model from their study to each of our user’s friends list to estimate which of four
income categories ($0-$50,000; $50,000-$100,000; $100,000-$150,000; $150,000+) the user
most likely belongs to, and used the lower bounds of each user’s most likely income bracket as
proxy construct for social power. Table 3 presents the Twitter accounts used, and their respective
regression weights (Culotta et al., 2015). Each user’s “score” for each income bracket was
calculated as the sum of the weights for the accounts they followed that corresponded to that
income bracket, and they were assigned to the income bracket for which they had the highest
score.
Population density. As aforementioned, we employed the Bing Maps API to match each
user’s self-described location with a U.S. city; those that could be correctly matched were cross-
referenced with data from the 2010 Census (United States Census Bureau, 2010) to produce city-
level population density for each user.
PSYCHOLOGICAL DISTANCE 27

Risk- and fear-related language. Just as study 1 employed measures of risk estimates
and self-reported fear, we attempted the estimated the degree to which these variables were
reflected in each Twitter post. We initially employed a word embeddings method, in which each
word in a post is represented by a vector in high-dimensional space, with the distance between
points/words representing their contextual similarity (e.g., “dog” and “cat” would be nearer than
“dog” and “book”). Using this approach, a Twitter post can be represented as the average vector
of all its included terms, and one can compute the “distance” between the post and a collection of
chosen terms (in our case, terms representing fear and risk). This approach is termed “distributed
dictionary representations” (Garten et al., 2018), and is typically well-suited for short pieces of
text such as social media posts. We initially measured each Twitter post’s “loadings” for the
concepts of “fear” and “risk” as its distance from a set of seed terms representing each construct
(e.g., “danger”, “avoid”, “careful” for risk, and “fearful”, “nervous”, “scared” for fear).
However, fear and risk measures for each Tweet were highly correlated (r=.93), likely because
the seed words for fear and risk are semantically and contextually similar (all pertaining to some
degree of negative sentiment). Thus, word embedding measures lacked sufficient discriminant
validity for use as dependent variables.
We instead chose to employ a simple word count method to quantify each post’s use of
fear-related language. To do so, we simply count the number of words in each post that belong to
the “fear” dictionary from LIWC 2007, a popular and well-validated collection of terms that
reflect various psychological constructs (Pennebaker, Booth, & Francis, 2007). The “fear”
dictionary includes terms such as “scared”, “vulnerable”, “stunned”, and “uneasy” (note that
words beginning with “terror,” though included in the dictionary, were not counted given that
they could simply reflect users discussing “terrorism”). Because a pre-validated “risk” dictionary
PSYCHOLOGICAL DISTANCE 28

is not available, we chose only to focus on the use of fear-related language as a proxy for users’
overall affective reactions to the Orlando shooting. Still, because this method relies on terms
from a previously-validated dictionary rather than contextual similarities of words, it likely
produces a more discriminating measure of user fear than a word embeddings approach.
Language Abstraction. Abstraction is a difficult linguistic variable to identify using
automated methods, which is perhaps why other work on linguistic abstraction has primarily
relied on hand-coded methods (Magee et al., 2010; Coenen, Medebouw, & Semin, 2006;
Liberman & Trope, 1998). Yet this approach, although useful, is infeasible with large samples
such as the one used here. To address this issue, we first attempted to quantify language
abstraction in each Twitter post based on the work of Johnson, Boghrati, Wakslak, and Dehghani
(2017), who developed a method for automatically quantifying text abstraction based on its
syntactical features. This method is conceptually based on the Linguistic Category Model
(Coenen et al., 2006), which stipulated that parts of speech differ in their level of abstraction;
adjective are considered highly abstract (as they represent context-invariant features of their
objects), while verbs differ in their abstraction depending on their role as state verbs (e.g.,
“love”, considered abstract), interpretative action verbs (e.g., “helping”, “working”, considered
somewhat abstract), or direct action verbs (e.g., “walking”, “running”, considered concrete).
Johnson et al. (2017) built on this framework by identifying the syntactical features (e.g.,
“compound noun”, “present participle verb”) that best predicted Linguistic Category Model
abstraction scores, and use these syntactic features to generate an abstraction score for a given
piece of text. Readers are directed to Johnson et al. (2017) for more detail on this coding scheme.
We do not provide further detail on this method because, while it holds great promise for
researchers’ ability to study construal level in large natural language samples, it did not
PSYCHOLOGICAL DISTANCE 29

adequately correlate with coders’ ratings of abstractness for our own sample. Four coders read a
random subsample of 40 posts whose syntax abstraction scores were in the top or bottom 20% of
the sample (i.e., posts that were highly abstract or concrete), and rated whether each post was
scored as highly abstract or highly concrete; however, while there was moderate inter-rater
agreement (average r=0.311), there was very little agreement between the coders’ ratings and the
syntax abstraction score (average r=-0.03). This result is not terribly surprising given the specific
context of the Orlando shooting, and the specific ways that abstract and concrete thinking likely
manifested themselves in discourse about the attack. Those who processed the attack in an
abstract manner might, for instance, express their views on terrorism or firearms regulation (i.e.,
issues to which the Orlando shooting was conceptually related), despite using “concrete”
syntactical features when expressing such views.
Thus, we developed an event-specific dictionary of terms to measure the degree to which
each Twitter post discussed concrete or abstract features of the Orlando nightclub shooting. We
first generated a list of words that could feasibly be related to the abstract or concrete construal
of the event. We hypothesized that concrete construal of the event would predict references to 1)
the actual scene of the attack, 2) the people involved in the attack (the shooter or the victims), or
3) specific actions that could be taken to aid the victims of the attack (e.g., donating money).
Thus, hypothesized concrete terms included “fifty” (the approximate number of deaths in the
attack), “club” (the location where the attack took place), and “donate” (actions that people can
take to aid the victims). We also hypothesized that abstract construal might predict discussions of
its underlying causes or its comparison to similar shootings or public terror attacks. Thus,
hypothesized abstract terms included words such as “prevent” (related to the issue of future mass
shootings), “deadliest” (a comparison between the attack and other mass shootings), and
PSYCHOLOGICAL DISTANCE 30

“terrorism” (a category of events to which the attack could belong). The full list of tested terms
appears in Table 4. Note that this list is not exhaustive of all possible terms that might indicate
concrete or abstract descriptions of the attack; our goal was simply to capture some of the
variance in construal-related language based on a short list of terms.
We then eliminated those terms that occurred in less than 1% of all posts, then computed
the polychoric correlations between the use of the remaining terms in a random 50% subsample
of the dataset. We first checked whether there were any substantial negative correlations between
related words (e.g., “died” and “killed”), which would indicate a tendency for words to serve as
synonyms or substitutions for one another, though none of the remaining word pairs exhibited
this pattern.
We then submitted the polychoric correlation matrix to an exploratory factor analysis to
examine whether the use of certain words could be explained by an underlying factor structure
(note that the use of polychoric, rather than Pearson, correlations is advisable when dealing with
non-continuous data; Holgado–Tello, Chacón–Moscoso, Barbero–García, & Vila–Abad, 2010).
If, as hypothesized, certain terms arise from concrete or abstract construal of the event, then the
use of those terms should be strongly correlated in the dataset, with factors that are easily
interpretable as reflecting either concrete or abstract word usage. After removing cross-loaded
terms, a two-factor structure emerged (factor correlation of r=-.11): the terms “50”, “injured”,
“dead”, “killed”, “police”, “club” and “mass shooting” loaded onto the first factor, while the
terms “violence”, “again”, “gun”, “guns”, and “stop” loaded onto the second factor. Based on its
included terms, we interpret the second category as relating to users’ perceptions of a national
pattern of gun violence (given the prevalence of firearm-related words and pattern-related words
such as “stop” and “again”).
PSYCHOLOGICAL DISTANCE 31

The first factor seemed to represent references to the specific details of the attack, given
its relation to the number of attack victims, descriptions of the attack victims, and descriptions of
the attack scene. Interestingly, we initially hypothesized “mass shooting” to function as a general
category of events, rather than a concrete reference to the attack. Yet given its loading onto the
same factor as specific detail terms, we chose to treat it as a concrete description of the attack
given that it references the specific type of weapon used (a “shooting”), compared to more
ambiguous terms such as “attack” or “violence.” We then submitted the other half of the dataset
to a confirmatory factor analysis, specifying the use of the terms “50”, “injured”, “dead’,
“killed”, “police”, “club”, and “mass shooting” as resulting from a single latent factor of
concrete construal. The terms “dead” and “killed” exhibited considerable negative residual
covariance, suggesting they may have served as substitutes for one another. Given their semantic
similarity, we combined these terms into a single term category and re-ran the model, which fit
the data well (CFI=0.995, SRMR=0.095, RMSEA=0.016). We thus treated the use of the terms
“50”, “injured”, “dead/killed”, “police”, “club”, and “mass shooting” as indicators of users’
tendency to describe the attack in concrete, specific details (average polychoric correlation of
0.43). However, to ensure that any findings are not overly dependent on any one of these terms,
we re-run all models with each of the six terms omitted to check that all effects are robust against
the exclusion of single terms (see Results section).
We initially operationalized “semantic concreteness” as each respondent’s factor score
from a two-parameter Item Response Theory models (as in study 1), with each of the six
concrete terms treated as dichotomous items, but found that this measure correlated strongly with
a simple count of the number of word categories (“50”, “injured”, “dead/killed”, “police”, “mass
shooting”, “club”) used in each post (r=0.93, r=0.76 for non-zero values). Thus, for simplicity,
PSYCHOLOGICAL DISTANCE 32

we used this count data (ranging from 0 to 6) as a measure of a post’s focus on concrete attack
details.
Analysis
We rely on Bayesian negative binomial regression to test for relationships between each
of our psychological distance predictors and our dependent variables of interest (fear-based and
concrete language use). Such models use a log-link function to predict the conditional mean of a
count variable (such as word counts) from the weighted sum of predictor values. Negative
binomial regression is a generalization of Poisson regression that can accommodate
overdispersion in the data (i.e., mean larger than variance; Bliss & Fisher, 1953; Lawless, 1987),
which was present for concrete term use ( 𝑥 ̅=0.20, s
2
=0.35). Furthermore, both fear-based term
use and concrete term use were well approximated by negative binomial distributions; see Tables
5 and 6 for the actual and expected counts for each variable under a Poisson and Negative
Binomial distribution fit. Note that, unlike concrete term use, fear-based term use was not
severely overdispersed ( 𝑥 ̅=0.072, s
2
=0.068) and did not fit a Negative Binomial distribution
drastically better than a Poisson distribution (see Table 5). However, for the sake of consistency,
we use negative binomial regression for all analyses given that it is merely a generalization of
Poisson regression that can easily accommodate properly-dispersed dependent variables.
Furthermore, because we use raw word counts rather than term frequencies (i.e., raw word
counts divided by the text’s length), we include Tweet word count as a covariate in all analyses.
Given that income and political ideology estimates were only available for just under half
of our sample (n=16,492), we first test for the effects of time, distance, and population density
(along with their interactions with time) on the entire sample. We then add income and political
ideology as predictors and test their effects only on the subsample (n=16,492) with values for all
PSYCHOLOGICAL DISTANCE 33

variables. Note that the users with valid income and political ideology estimates did not
appreciably differ on their mean levels of fear-based term use, concrete term use, or any of the
predictor variables compared to the remainder of the sample (all ds<0.04).

PSYCHOLOGICAL DISTANCE 34

STUDY 2: RESULTS
Descriptive Statistics
Dependent variables. Regarding the use of fear-related language, most users (93.6%)
used none of the “fear” terms from the LIWC 2007 dictionary, while 6.1% used one term, 0.34%
used two terms, 0.01% used three terms, and 0.006% (only two users) used four terms (M=0.068,
SD=0.27). Concrete language use was somewhat more prevalent; the majority of the sample
(86.3%) used none of the aforementioned concrete language categories, while 9.6% used one
category, 2.3% used two categories, 1.3% used three categories, 0.4% used four categories, and
0.07% used all five categories (M=0.20, SD=0.59). Thus, the overall prevalence of both of these
language variables was relatively low.
Predictor variables. Time since the Orlando attack was virtually uncorrelated with
distance and population density (Spearman ρ<0.03), while distance and density were somewhat
correlated (Spearman ρ=0.26, likely reflecting larger, distant cities such as Los Angeles and New
York). For the subsample of users for whom income and political orientation estimates were both
available (n=16,492), all independent variables (distance from attack, time since attack,
population density, income estimates, political orientation estimates) were relatively uncorrelated
(all | 𝑟𝑠 |<0.15).
Figure 1 presents kernel density plots for each predictor. As can be seen from the plots,
most variables exhibited some degree of skew, with some bimodality present in the population
density variable (largely due to the high calculated population density of New York City, which
comprises the righthand mode of the distribution). As aforementioned, we conduct all analyses
using the logarithm of population density rather than its raw value (also presented in Figure 1).
Also of note is the time variable’s multiple “peaks” that occur at fixed intervals, which this likely
PSYCHOLOGICAL DISTANCE 35

arises from daily rhythms in Twitter post activity (e.g., more posts made in the evening than in
the morning).
Psychological Distance and Fear
Use of fear-related terms reliably decreased as a function of time since the attack
(e
b
=0.815; all 4000 posterior distribution samples lied below zero). Interpreting the
exponentiated regression coefficient suggests that the mean number of fear-related terms used in
each post decreased by 18.5% (1-0.815=0.185) for each day elapsed since the shooting. Use of
fear-related terms also depended on an interaction between time and distance from the attack
(e
b
=1.06, 99.9% of the posterior density lied above zero). This interaction suggested that, for
each additional 1000 miles of distance from the Orlando attack, the effect of time (an 18.1%
decrease in fear terms per day) diminished by roughly 5%. Thus, the effect of time was most
pronounced at locations closer to Orlando, and was essentially nonexistent at roughly 3500 miles
of distance from the Orlando attack. The whole-sample regression model found no reliable main
effect of population density or distance from the attack.
Analysis of the subsample of users with income and political orientation estimates
suggested a reliable effect for both. An increase in one income bracket corresponded to a 6.9%
decrease in fear-related term use (e
b
=0.931, 98.1% of the posterior below zero), while a one
standard deviation change towards more liberal political orientation resulted in a 7.3% increase
in term use (e
b
=1.07, 97.5% of the posterior above zero). For a visualization of these effects, see
Figure 2 for histograms of fear-related term use at the highest and lowest levels of each reliable
predictor.

PSYCHOLOGICAL DISTANCE 36

Psychological Distance and Concrete Language Use
Concrete term use decreased by 41.8% each passing day since the attack (e
b
=0.582, all
4000 posterior samples lied below zero), and was lower for users physically more distant from
the attack (e
b
=0.897, 99.9% of the posterior density below zero), such that term use decreased by
10.3% for each 1000 miles of distance from Orlando. As with the analysis of fear-related
language, there was a small interaction between time and physical distance (e
b
=1.043, 98.0% of
the posterior density above zero), such that for each 1000 miles of distance from Orlando, the
effect of time (a 41.7% drop in term use per day) decreased by roughly 2.5%. There were no
reliable main effects or interactions involving population density, income, or population density.
See Figure 3 for histograms of concrete term use at the highest and lowest levels of each reliable
psychological distance predictor.
To demonstrate that these effects were robust against the specific terms chosen to
represent “concrete language,” we re-ran the above regression model six times, each time
dropping one of the concrete terms (“50”, “dead/killed”, “injured”, “police”, “club”, “mass
shooting”) from the dependent variable count. We then compare the resulting coefficients for
time, distance, and their interaction (the reliably nonzero effects from the full model) to a null
distribution generated under the hypothesis that all six terms were perfectly exchangeable. This
distribution was generated by randomly selecting one term to omit from the concrete language
“dictionary” for each user in the data set, regressing this concrete language variable on all
predictors, and repeating this process for 1000 iterations. The resulting distributions of regression
coefficients reflect variation in the regression weights under the null assumption that each of the
six terms serve as equivalent measures of concrete thinking, and are thus exchangeable at the
level of the user. If any one term is artificially inflating the regression effects in our main model
PSYCHOLOGICAL DISTANCE 37

(e.g., if “mass shooting” is more highly related to the predictors than any of the other terms),
then the regression coefficient from the model that omits that term should be substantially
smaller than most of the regression coefficients generated from this null hypothesis. Models for
the whole sample suggested that the effects of time, distance, and their interaction on concrete
term use were not significantly attenuated when removing any one of the six concrete terms (all
ps>0.05). The one exception was the effect of time, which dropped to -0.52 when omitting the
term “injured”, compared with the null distribution mean of -0.55, though this difference was
small in comparison with the magnitude of the effect.
Relationship Between Concrete Construal and Fear-Related Language
Figure 4 presents a side-by-side comparison of the effect sizes for each psychological
distance predictor on both fear-related and concrete term use. To test whether concrete language
use meaningfully mediated any of the relationships between the psychological distance variables
and fear-related language, we re-ran the regressions for fear terms including concrete term use as
a covariate, to examine whether any of the previously meaningful regression coefficients
decreased in value. Instead, contrary to our hypothesis, there was a reliably negative relationship
between concrete term use and fear term use in the whole-sample model (e
b
=0.896, 99.5% of the
posterior density below zero)., such that each additional concrete term predicted a 10.4%
decrease in the use of fear-related terms. The other regression coefficients in the whole-sample
model (btime=-0.21, bdistance=-0.04, bdensity=0.02, btime*dist=0.06) were similar in magnitude to those
obtained without including concrete term use as a covariate (btime=-0.20, bdistance=-0.04,
bdensity=0.02, btime*dist=0.06). For the n=16,492 subsample with all data available, the income and
political regression coefficients remained unchanged when including concrete language use as a
covariate (bincome=-0.07, bpolitical=0.07). Thus, rather that concrete language use mediating the
PSYCHOLOGICAL DISTANCE 38

relationship between psychological distance and fear, for some predictors it exhibited a
suppression effect, in which concrete language use negatively predicted fear-based language
while simultaneously positively correlating with some of its predictors.

PSYCHOLOGICAL DISTANCE 39

STUDY 2: DISCUSSION
The data supported some of our hypotheses regarding the role of psychological distance
in Twitter users’ construal of the 2016 Orlando attack. There was an especially pronounced
effect of time on users’ language surrounding the attack; as time passed, users became much less
likely to use a collection of terms that referred to the specific details of the event. This finding
makes intuitive sense—it would seem more appropriate to communicate specific details of the
attack in its immediate aftermath than days afterward, once the details of the attack became more
widely known. This effect was coupled, however, with a tendency for users closer to the site of
the attack to use more concrete terms than more distant users. If interpreted from a Construal
Level standpoint, these results seem to suggest that, as physical and temporal distance from the
attack decreased, users were generally less inclined to recount concrete details of the attack. We
also observed an interactive effect between these two variables, such that the effect of physical
distance diminished in the days following the attack. It seems that time served to “homogenize”
concrete term use across the United States, with location-based variations in concrete term use
diminishing over the course of the week following the event. Said another way, users closer to
the attack demonstrated a more pronounced decrease in language use over time than users further
from the attack, whose initial levels of concrete construal were lower to begin with. While we
hypothesized that time might synergistically interaction with other psychological distance
predictors, such interactions have not been the focus of much study within the Construal Level
Theory framework, and further research may be needed to understand the cognitive processes
underlying this effect.
Users’ expression of fear shared some of the same predictors as concrete construal,
generally decreasing with time since the attack, and showing a more pronounced effect of time at
PSYCHOLOGICAL DISTANCE 40

locations closest to Orlando. However, fear-based language was also predicted by users’
estimated income and political orientation, in the expected directions; fear terms were more
common among lower-income and more liberal users (as estimated by the accounts they
followed). We initially hypothesized that higher incomes and more conservative political
orientations would predict greater psychological distance from the attack, in the form of 1)
greater social power (operationalized by income), and 2) less identification with the LGBT
community (operationalized by political orientation). However, given that neither of these
variables reliably predicted concrete term use, their correlations with fear-based language might
better be explained through other plausible mechanisms. Liberal users, with generally stricter
views on gun control than conservative users, may simply have been more prone to experience
fear after a mass shooting of any kind (regardless of the population it victimized). Lower-income
individuals have previously been shown to perceive greater risks from societal hazards than
higher-income individuals (Flynn, Slovic, & Mertz, 1994; Fothergill & Peek, 2004), though this
effect may be attributable to feelings of vulnerability or the differential impacts of disaster events
on populations with low socioeconomic status (Flynn, Slovic, & Mertz, 1994; Fothergill & Peek,
2004). Thus, while our findings regarding temporal and physical distance seem to support a
Construal Level account of reactions to the Orlando shooting, more research is needed to clarify
whether the same effects can be observed for social distance, of which we found no evidence for
a construal effect.
Yet any support for a Construal Level account of users’ reactions should be qualified by
the unexpected suppression effect observed between concrete term use and fear expressions. As
hypothesized, both fear and concrete language use shared some of the same predictors, yet the
uses of fear-related and concrete terms were negatively correlated in the sample. It seems that,
PSYCHOLOGICAL DISTANCE 41

while being temporally and physically close to the event predicted greater expressions of fear
and concrete construal, the presence of one made the other less likely. Two explanations seem
plausible. The first is simply that concrete construal is negatively predictive of fear, and that fear
depends on temporal and physical distance through mechanisms besides concrete/abstract
construal. For instance, focusing on the concrete details of a disaster event might prevent one
from ruminating on its underlying causes or consequences (e.g., “this was an unpredictable terror
attack”, “these shootings will only continue to get worse”), which might otherwise lead to
increases in fear and risk perception. The positive correlations between fear and
temporal/physical distance might then be simply explained by mechanisms that do not involve
abstract/concrete construal (e.g., the availability heuristic, assuming that “it would never happen
here,” etc.). The other possibility is that concrete construal generally predicts greater expressions
of fear, and their negative correlation is an artifact of Twitter’s limits on post length. Users can
only include so many words in their Twitter posts, and even if concrete construal and fear
positively covary, each user may simply have to “choose” which sentiment or idea to express in
their short character allotment. This effect could be especially pronounced in our analysis given
that we control for total word count in all regressions. For a given total post length, it may be
relatively unsurprising the presence of a word count variable would negatively predict the
presence of another, even if they are conceptually related. We find some support for this notion
from the fact that, when omitting total word count as a covariate in the whole-sample prediction
of fear-related term use, the effect of concrete language is almost null (e
b
=0.983) and not reliably
below zero, suggesting that part of the suppression effect may simply be due to word count
restrictions. Thus, the relationship between fear and terror attack construal may require further
PSYCHOLOGICAL DISTANCE 42

study, to investigate whether the effect observed here is indicative of a true negative relationship,
or merely an artifact of studying multiple word count variables in fixed-length texts.
There are, of course, many limitations of any psychological investigation involving social
media. We invariably selected participants by outcome, by only including individuals who 1) had
Twitter accounts, and 2) chose to post in response to the Orlando attack. One particular
limitation of text-analytic studies in general, especially those involving wordcount variables, is
small effect sizes, given the noise inherent to studies of natural language. Note that the effect of
time as a predictor of fear expression and concrete language was fairly substantial, but all other
effects are relatively miniscule (as shown in Figures 2, 3, and 4). It is important to emphasize
that such effects are much more valuable from a theory-building standpoint than a predictive
standpoint. We argue that this investigation has served as a useful and insightful test of Construal
Level Theory’s implications for public disaster response, and even a small but reliably nonzero
effect with data as noisy as natural language samples can serve to validate theoretical claims. Yet
any subsequent attempts to predict out-of-sample concrete term use from the predictors reported
here would likely perform relatively poorly.
Of course, the theoretical explanations for our reported findings are only as valid as our
chosen conceptualization of “concrete construal” in the context of the Orlando attack. Our
collection of terms was carefully constructed through exploratory and confirmatory factor
analyses, but this is not sufficient to guarantee that such terms signal concrete construal. We did
confirm that all observed effects were robust against the omission of any one term (further
supporting the reliability of the term dictionary), but future studies could develop even more
rigorous methods of measuring construal in a public disaster context. Another limitation of our
concrete term measure is its unipolar nature—we were unable to identify a collection of terms
PSYCHOLOGICAL DISTANCE 43

that signaled abstract processing of the event. The theoretical assumption underlying Construal
Level Theory is that concrete and abstract construal inhabit opposite ends of a one-dimensional
spectrum, and so it may be feasible to think of “more concrete construal” as signaling “less
abstract construal”—but this should be verified in future natural language studies. If studying
event construal through natural language is to make productive use of automated methods (rather
than human coding), it will likely require validated ways of measuring concrete and abstract
construal, something this investigation stopped short of accomplishing. Still, our data confirm
that, in line to Construal Level Theory, physical and temporal distance from a highly publicized
terror attack reliably predicted individuals’ tendencies to reference specific details of the event.
Our results also suggest a role for interactive effects between psychological distance dimensions,
previously unstudied in real-world investigations of psychological construal, and highlight a
need to clarify the relationship between construal and fear in the wake of public disasters.

PSYCHOLOGICAL DISTANCE 44

GENERAL DISCUSSION
Results from study 1 suggested that the Orlando nightclub shooting increased
respondents’ risk ratings for similar types of terror attacks, but that risk and fear ratings did not
depend on individuals’ psychological distance from the attacks. In contrast, study 2 suggested
that psychological distance from the Orlando attack did predict individual construal levels and
fear responses, but that these two outcome variables may not correlate in the expected direction.
This disparity between the samples might simply be due to their composition—demographic
differences between the panel sample and Twitter users who chose to respond to the Orlando
shooting might simply moderate the effects of psychological distance variables on fear.
However, it is more likely that the differing results stem from the inherent selection bias in the
Twitter sample—only users who felt extreme enough about the attack to post about it were
included for analyses. This is a considerable limitation of any social media study of isolated
events, as it effectively selects users based on the outcome measure of study. Certain methods
can get around this issue (see Lin et al., 2017), but our theoretical interest in language directed
towards the Orlando attack necessitated selecting a sample of users who actively chose to
respond to the event.
Of course, our hypotheses regarding the role of psychological distance in predicting
Twitter users’ fear- and construal-related language do not inherently depend on sample
representativeness (as they specifically pertain to users who chose to post about the attack). Yet
such results can only viably be compared to results from the panel survey if the findings from
both studies agreed (which might suggest the effects’ robustness against sample selection).
Given that psychological distance played different roles in each study, it is difficult to know
whether this disparity is due to sample selection, the demographics of the sample, the dependent
PSYCHOLOGICAL DISTANCE 45

measures analyzed (linguistic variables versus explicit self-report), or some combination of
these.
Still, results from the Twitter analysis suggest fruitful avenues for further research on
public risk perception from a psychological distance standpoint. Future studies can continue to
clarify the role of physical and temporal distance in expressions of fear, mechanisms that explain
the effects of income and political affiliation, and whether time interacts with other individual
differences besides physical distance. We argue that this study’s greatest contribution lies in its
use of automated text analysis methods to study abstract and concrete construal in a public
disaster context. While other language-based measures of construal have focused on human
coding, this study is the first to identify psychometrically-sound groups of terms that seem to
reflect individual construal levels for a specific event. While the unidimensionality underlying
our list of concrete terms is not sufficient to establish its complete validity, we argue that its
internal consistency, robustness against specific term omissions, and its correlations with
psychological distance predictors make a convincing case for its use as an efficient and
automated measure of concrete construal, and similar methods could be used to develop “event-
specific dictionaries” for other hazards of interest. In the domain of public risk perception,
techniques such as this can allow for natural language samples to complement and even augment
studies of lab or survey data, and to provide theoretically useful insights in their own right.

PSYCHOLOGICAL DISTANCE 46

REFERENCES
Bar-Anan, Y., Liberman, N., & Trope, Y. (2006). The association between psychological
distance and construal level: evidence from an implicit association test. Journal of
Experimental Psychology: General, 135(4), 609.
Barnett, A. (2015). Has successful terror gone to ground?. Risk Analysis, 35(4), 732-740.
Blalock, G., Kadiyali, V., & Simon, D. H. (2009). Driving fatalities after 9/11: A hidden cost of
terrorism. Applied Economics, 41(14), 1717-1729.
Bliss, C. I., & Fisher, R. A. (1953). Fitting the negative binomial distribution to biological
data. Biometrics, 9(2), 176-200.
Bourque, L. B., Mileti, D. S., Kano, M., & Wood, M. M. (2012). Who prepares for
terrorism? Environment and Behavior, 44(3), 374-409.
Cardeña, E., Dennis, J. M., Winkel, M., & Skitka, L. J. (2005). A snapshot of terror: Acute
posttraumatic responses to the September 11 attack. Journal of Trauma &
Dissociation, 6(2), 69-84.
Coenen, L.H.M., Hedebouw, L., & Semin, G.R. (2006). Measuring language abstraction: The
Linguistic Category Model (LCM). Retrieved January 10, 2017, from
http://www.cratylus.org/Text/1111548454250-3815/pC/1111473983125-
6408/uploadedFiles/1151434261594-8567.pdf
Culotta, A., Ravi, N. K., & Cutler, J. (2015). Predicting the demographics of Twitter users from
website traffic data. In Proceedings of the 29th AAAI Conference on Artificial
Intelligence, AAAI 2015 and the 27th Innovative Applications of Artificial Intelligence
Conference, IAAI 2015 (Vol. 1, pp. 72-78). AI Access Foundation.
PSYCHOLOGICAL DISTANCE 47

Dehghani, M., Johnson, K., Hoover, J., Sagi, E., Garten, J., Parmar, N. J., Vaisey, S., Iliev, R., &
Graham, J. (2016). Purity homophily in social networks. Journal of Experimental
Psychology: General. Advance online publication. http://dx.doi.org/10.1037/xge0000139
Fairbrother, G., Stuber, J., Galea, S., Fleischman, A. R., & Pfefferbaum, B. (2003). Posttraumatic
stress reactions in New York City children after the September 11, 2001, terrorist attacks.
Ambulatory Pediatrics, 3(6), 304-311.
Finucane, M. L., Slovic, P., Mertz, C. K., Flynn, J., & Satterfield, T. A. (2000). Gender, race,
and perceived risk: The 'white male' effect. Health, Risk & Society, 2(2), 159-172.
Fischhoff, B., Gonzalez, R. M., Small, D. A., & Lerner, J. S. (2003). Judged terror risk and
proximity to the World Trade Center. Journal of Risk and Uncertainty, 26(2), 137-151.
Flynn, J., Slovic, P., & Mertz, C. K. (1994). Gender, race, and perception of environmental
health risks. Risk Analysis, 14(6), 1101-1108.
Fothergill, A., & Peek, L. A. (2004). Poverty and disasters in the United States: A review of
recent sociological findings. Natural Hazards, 32(1), 89-110.
Fujita, K., Henderson, M. D., Eng, J., Trope, Y., & Liberman, N. (2006). Spatial distance and
mental construal of social events. Psychological Science, 17(4), 278-282.
Gaissmaier, W., & Gigerenzer, G. (2012). 9/11, Act II: A fine-grained analysis of regional
variations in traffic fatalities in the aftermath of the terrorist attacks. Psychological
Science, 23(12), 1449-1454.
Galea, S., Vlahov, D., Resnick, H., Ahern, J., Susser, E., Gold, J., Bucuvalas, M., & Kilpatrick,
D. (2003). Trends of probable post-traumatic stress disorder in New York City after the
September 11 terrorist attacks. American Journal of Epidemiology, 158(6), 514-524.
PSYCHOLOGICAL DISTANCE 48

Garten, J., Hoover, J., Johnson, K. M., Boghrati, R., Iskiwitch, C., & Dehghani, M. (2018).
Dictionaries and distributions: Combining expert knowledge and large scale textual data
content analysis. Behavior Research Methods, 50(1), 344-361.
Golbeck, J., & Hansen, D. (2011). Computing political preference among twitter followers.
In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp.
1105-1108). ACM.
Graham, J., Haidt, J., & Nosek, B. A. (2009). Liberals and conservatives rely on different sets of
moral foundations. Journal of Personality and Social Psychology, 96(5), 1029.
Hill, T. D., Moulton, B. E., & Burdette, A. M. (2004). Conservative Protestantism and attitudes
toward homosexuality: Does political orientation mediate this relationship? Sociological
Focus, 37(1), 59-70.
Holgado–Tello, F. P., Chacón–Moscoso, S., Barbero–García, I., & Vila–Abad, E. (2010).
Polychoric versus Pearson correlations in exploratory and confirmatory factor analysis of
ordinal variables. Quality & Quantity, 44(1), 153-166.
Hoover, J., Johnson-Grey, K., Boghrati, R., Graham, J., & Dehghani, M. (2017). Moral Framing
and Charitable Donation: Integrating exploratory social media analyses and confirmatory
experimentation. Unpublished manuscript. Retrieved from https://osf.io/a2qjm/
Johnson, K. M., Boghrati, R., Wakslak, C., & Dehghani, M. Measuring Abstract Mindsets
through Syntax: Improvements in Automating the Linguistic Category Model.
Unpublished manuscript. Retrieved from
https://mindmodeling.org/cogsci2017/papers/0441/paper0441.pdf
Lawless, J. F. (1987). Negative binomial and mixed Poisson regression. Canadian Journal of
Statistics, 15(3), 209-225.
PSYCHOLOGICAL DISTANCE 49

Lerner, J. S., & Keltner, D. (2000). Beyond valence: Toward a model of emotion-specific
influences on judgement and choice. Cognition & Emotion, 14(4), 473-493.
Lerner, J. S., Gonzalez, R. M., Small, D. A., & Fischhoff, B. (2003). Effects of fear and anger on
perceived risks of terrorism: A national field experiment. Psychological science, 14(2),
144-150.
Liberman, N., & Trope, Y. (1998). The role of feasibility and desirability considerations in near
and distant future decisions: A test of temporal construal theory. Journal of Personality
and Social Psychology, 75(1), 5.
Lin, Y. R., & Margolin, D. (2014). The ripple of fear, sympathy and solidarity during the Boston
bombings. EPJ Data Science, 3(1), 31.
Lin, Y. R., Margolin, D., & Wen, X. (2017). Tracking and analyzing individual distress
following terrorist attacks using social media streams. Risk Analysis, 37(8), 1580-1605.
Magee, J. C., & Smith, P. K. (2013). The social distance theory of power. Personality and Social
Psychology Review, 17(2), 158-186.
Magee, J. C., Milliken, F. J., & Lurie, A. R. (2010). Power differences in the construal of a crisis:
The immediate aftermath of September 11, 2001. Personality and Social Psychology
Bulletin, 36(3), 354-370.
Marshall, R. D., Bryant, R. A., Amsel, L., Suh, E. J., Cook, J. M., & Neria, Y. (2007). The
psychology of ongoing threat: relative risk appraisal, the September 11 attacks, and
terrorism-related fears. American Psychologist, 62(4), 304.

Ooms, J. (2017). cld2: Google's Compact Language Detector 2. R package version 1.1. https://C
RAN.R-project.org/package=cld2
PSYCHOLOGICAL DISTANCE 50

Parkinson, C., Liu, S., & Wheatley, T. (2014). A common cortical metric for spatial, temporal, a
nd social distance. Journal of Neuroscience, 34(5), 1979-1987.
Pennebaker, J. W., Booth, R. J., & Francis, M. E. (2007). LIWC2007: Linguistic inquiry and
word count. Austin, TX.
Snefjella, B., & Kuperman, V. (2015). Concreteness and psychological distance in natural
language use. Psychological Science, 26(9), 1449-1460.
Stapleton, A.C. (2016). Timeline of Orlando nightclub shooting. CNN. Retrieved from
https://www.cnn.com/2016/06/12/us/orlando-shooting-timeline/index.html
Trope, Y., & Liberman, N. (2010). Construal-level theory of psychological
distance. Psychological Review, 117(2), 440.
United States Census Bureau. (2010). 2010 Census Summary File 1. [Comma separated values
file]. Retrieved from
https://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?src=CF
United States Census Bureau. (2017). 2017 Census U.S. Gazetteer files. [text file]. Retrieved
from https://www.census.gov/geo/maps-data/data/gazetteer.html
Wakslak, C., & Trope, Y. (2009). The effect of construal level on subjective probability
estimates. Psychological Science, 20(1), 52-58.
Wallace, J.R. (2012). Imap: Interactive Mapping. R package version 1.32. https://CRAN.R-
project.org/package=Imap

PSYCHOLOGICAL DISTANCE 51

TABLES
Table 1
Bivariate correlations for psychological distance predictors (study 1)
1 2 3 4 5
Distance from Orlando 1.00
Time before/after
attack
.09 1.00
Political Orientation -.08 -.06 1.00
Income .14 .00 -.02 1.00
Population Density .10 .06 -.10 .13 1.00

Table 2
Political ideology score for Twitter news accounts (study 2)

News Account Twitter Handle Ideology Score
Fox News @foxnews 26
Drudge Report @Druge_Report 27
Washington Times @washtimes 29
Wall Street Journal @WSJ 41
U.S. News @usnews 49
Los Angeles Times @latimes 51
U.S.A. Today @USATODAY 51
Good Morning America @gma 51
News Hour @NewsHour 52
CBS News @CBSNews 53
Newsweek @Newsweek 54
Washington Post @Washingtonpost 54
Today Show @todayshow 55
The Early Show @theearlyshow 55
TIME @TIME 57
ABC World News @abcworldnews 57
CNN Breaking News @Cnnbrk 58
NBC Nightly News @nbcnightlynews 58
New York Times @nytimes 60
Morning Edition @MorningEdition 66
PSYCHOLOGICAL DISTANCE 52

Table 3
Twitter accounts used for estimation of user income (study 2)
Username $0-$50,000
$50,000-
$100,000
$100,000-
$150,000
$150,000+
YouTube 130 -47 -45 -37
PlayStation 120 -55 -37 -25
IGN 110 -52 -34 -23
RockstarGames 110 -52 -33 -22
KevinHart4real 93 -32 -29
Xbox 91 -43 -28
Ubisoft 91 -45 -28
Drake 89 -40 -29
EA 88 -43
steam_games 87 -42
WSJ -180 43 67 68
nytimes -130 40 48 48
TheEconomist -120 49 53
washingtonpost -120 43 40
Forbes -110 43 45
AP -110 44 36
business -100 39 44
AdamSchefter -98 40 36
cnnbrk -95 56
espn -90 46
SportsCenter 44
ErinAndrews 43
jimmyfallon 43
MarthaStewart 40
gamespot -37
HarvardBiz 36 40
TechCrunch 34 36
katyperry -28 -26
instagram -27 -28
NewYorker 41
imbeizama 34
TheEllenShow -34
taylorswift13 -24
Pink -22
NOTE: User scores for each income bracket were calculated as the sum of that bracket’s
coefficients for the accounts the user followed; each user’s income was assigned as the
category with the maximum score
PSYCHOLOGICAL DISTANCE 53

Table 4
Terms originally hypothesized to reflect concrete/abstract construal
Terms Description Hypothesized Construal
50, 49, fifty, forty nine,
casualty, casualties, injured,
injury, injuries, dead, died,
killed, deaths, victims
Specific references to those
involved in the attack
Concrete
Police, officer, officers, law
enforcement, club, scene
Description of the attack
scene
Concrete
Support, donate Helping behaviors Concrete
Shooter, attacker, gunman
Description of attack
perpetrator
Concrete
Attack Description of event Concrete or Abstract
Mass shooting, terror,
violence, tragedy
General description of event Abstract
Deadliest, worst, again,
another
Reference to other attacks Abstract
Mental, gun, guns, firearm
Societal issues related to
attack
Abstract
Stop, prevent, end
Actions related to future
attacks
Abstract

NOTE: Bolded terms were used in more than 1% of posts and were retained for factor analysis

Table 5
Actual versus expected fear term counts under Poisson and Negative Binomial distributions
(study 2)

# of Categories Used
0 1 2 3 4
Data 33922 2205 125 5 2
NB(0.052, 1.26) 33924 2199 128 7 0
Pois(0.068) 33864 2314 79 2 0

PSYCHOLOGICAL DISTANCE 54

Table 6
Actual versus expected concrete term counts under Poisson and Negative Binomial distributions
(study 2)

# of Categories Used
0 1 2 3 4 5
Data 31294 3481 841 465 154 24
NB(0.289, 0.361) 31287 3483 972 325 118 45
Pois(0.163) 29652 5965 600 40 2 0

PSYCHOLOGICAL DISTANCE 55

FIGURES
z

Figure 1. Density plots for all study 2 predictor variables. Note: Political orientation and income
estimates only shown for users with non-missing values (n=16,492).
PSYCHOLOGICAL DISTANCE 56

Figure 2. Reliable effects on fear-related term use. Note that, due to the high quantity of zero
values, y-axes are truncated to better visualize effects. Top left: Term use for top 10%
(dark gray) and bottom 10% (light grey) of time since attack, for bottom 10% of distance
variable. Proportion of zeroes: top 10%=.975, bottom 10%=.924. Top right: Term use
for top 10% (dark gray) and bottom 10% (light grey) of time since attack, for top 10% of
distance variable. Proportion of zeroes: top 10%=.932, bottom 10%=.913. Bottom left:
Term use for $150,000+ income bracket (dark gray) and <$50,000 income bracket (light
grey) Proportion of zeroes: $150,000+=0.939, <$50,000=0.935. Bottom right: Term use
for 10% most conservative (dark gray) and 10% most liberal (light grey) political
orientation scores. Proportion of zeroes: Conservative=0.949, Liberal=0.939.
PSYCHOLOGICAL DISTANCE 57

Figure 3. Reliable effects on concrete term use. Top left: Term use for top 10% (dark gray) and
bottom 10% (light grey) of time since attack, for bottom 10% of distance variable. Top
right: Term use for top 10% (dark gray) and bottom 10% (light grey) of time since
attack, for top 10% of distance variable. Bottom left: Term use for top 10% (dark gray)
and bottom 10% (light gray) distance values. Proportion of zeroes: top=0.881,
bottom=0.877.

PSYCHOLOGICAL DISTANCE 58

Figure 4. Comparison of effects for psychological distance variables on fear-related and concrete
term use, for one unit increases in each predictor (Time=1 day, Distance=1,000 miles;
Density=doubling of population density; Political=1 SD towards liberal;
Income=increase in one $50,000 bracket). NOTE: Political and income effects reported
for n=16,942 subsample, all others reported for whole sample.
0
5
10
15
20
25
30
35
40
45
Time Distance Time*Distance Density Political Income
% Change in Term Use
Concrete Terms Fear Terms
PSYCHOLOGICAL DISTANCE 59

Figure 5. Concrete term use for posts with at least two fear-related terms (dark gray) and posts
with no fear-related terms (light grey). Proportion of zeroes: At least 2 fear terms=.902,
no fear terms=.864.

Abstract (if available)

Abstract Individuals experience and react to disaster events at varying levels of psychological distance

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Preparing for natural disasters: investigating the effects of gain-loss framing on individual choices

PDF

Choice biases in making decisions for oneself vs. others

PDF

Outlier-robustness in adaptations to the lasso

PDF

Disaster near-miss appraisal: effects of attribution, individual differences, psychological distance, and cumulative sequences

PDF

A self-knowledge model of social inference

PDF

“What difficulty means for me”: predictors and consequences of difficulty mindsets

PDF

Sacrificing cost, performance and usability for privacy: understanding the value of privacy in a multi‐criteria decision problem

PDF

Sex differences in moral judgements across 67 countries

PDF

The moral foundations of needle exchange attitudes

PDF

The cost of missing objectives in multiattribute decision modeling

PDF

Bound in hatred: a multi-methodological investigation of morally motivated acts of hate

PDF

Theory of mind processing in expectant fathers: associations with prenatal oxytocin

PDF

Regular unipotent elements of finite maximal subgroups of classical groups

PDF

The spread of moral content in online social networks

PDF

Expressing values and group identity through behavior and language

PDF

The role of primary care in the use of specialty mental health services: an investigation of utilization patterns among African Americans and non-Hispanic Whites

PDF

People can change when you want them to: changes in identity-based motivation affect student and teacher Pathways experience

PDF

How perceived moral congruence shapes propensities to engage in pro-group behaviors

PDF

Prison of Peace in a jail setting: a randomized trial of a conflict resolution program for incarcerated women

PDF

Effects of childhood adversity on physiology and health in emerging adulthood

Asset Metadata

Creator Baucum, Matthew (author)

Core Title Psychological distance in the public’s response to terrorism: An investigation of the 2016 Orlando nightclub shooting

School College of Letters, Arts and Sciences

Degree Master of Arts

Degree Program Psychology

Publication Date 07/26/2018

Defense Date 04/17/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag construal level,data mining,OAI-PMH Harvest,social media,text analysis

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor John, Richard (committee chair), Bechara, Antoine (committee member), Dehghani, Morteza (committee member)

Creator Email baucum@usc.edu,matt.melissa.baucum@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-27417

Unique identifier UC11672126

Identifier etd-BaucumMatt-6494.pdf (filename),usctheses-c89-27417 (legacy record id)

Legacy Identifier etd-BaucumMatt-6494.pdf

Dmrecord 27417

Document Type Thesis

Format application/pdf (imt)

Rights Baucum, Matthew

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA