Abstract—Using the Electronically Activated Recorder (EAR), a new
methodology for sampling behavioral data in naturalistic settings, we
tracked the social lives of 11 people by recording 30-s snippets of ambient
sounds in their environment approximately every 12 min. Participants
wore the EAR continuously for 10 days from September 11, 2001.
Pre-September 11 baseline data were available for all participants.
Analyses of the coded sound information showed that although participants
did not change in their overall amount of interactions, they gradually
shifted from group conversations to dyadic interactions. Exploratory
analyses revealed that a relative increase in dyadic interactions over the
first 10 days after September 11 was marginally related to better psychological
adjustment at follow-up. The findings have relevance for the understanding
of stress and affiliation and normal psychological reactions
to emergencies.

The psychological study of real-life disasters faces a number of
challenges. Consequently, very little empirical information
on how cultural upheavals affect peoples’ social lives is available.
What is known is almost exclusively derived from after-the-fact
accounts retrospectively collected days, weeks, or months after the
actual events. Studying how people naturally change their social interactions
in the aftermath of a disaster is important from at least two perspectives .
First, such research can help clarify some vexing theoretical
issues surrounding the human stress response. Second, social disaster research
can provide critical information for health practitioners about normal
and healthy psychological reactions to emergencies.

In times of stress and uncertainty, humans seek out others. Different
theoretical accounts of this response have emphasized a desire for
self-evaluation and the calibration of one’s opinions and feelings (Festinger,
1954), alleviation of anxiety (Schachter, 1959), maximization of
social support (Cohen & Wills, 1985), affirmation of one’s own threatened
cultural worldview (Pyszczynski, Solomon, & Greenberg, 2002),
and facilitation of social sharing (Rimé, Finkenauer, Luminet, Zech, &
Philippot, 1998).

Although these accounts paint a multifaceted picture of why humans
affiliate in times of crisis, they remain silent about how this increased
social motivation manifests itself in peoples’ everyday lives.
Most of the research has looked at choice behavior in the laboratory.
In real-life settings, however, people actively select, shape, and interact
with their social world (Buss, 1987). For example, in times of crisis,
do people simply spend more time with others, or do they spend
more time with certain others? Do they prefer to be with one person or
with a group of people? Is there a temporal aspect to people’s need for
affiliation? That is, does it change in its expression as time goes by after
the onset of the stressor? Answers to these questions would help
develop a better understanding of the naturally occurring human stress

There exists a widespread cultural belief that talking about upsetting
life events is health promoting (Everly & Mitchell, 2001; Rimé et
al., 1998). Existing empirical evidence, however, lags behind practitioners’
recommendations and in fact does not support an unqualified
“the more the better” approach within the context of trauma (Rimé et
al., 1998). Indeed, early emotional disclosure may even disrupt rather
than support a natural coping process (e.g., Bisson, Jenkins, Alexander,
& Bannister, 1997; Brewin, 2001). It would be extremely valuable to
learn about the time course of how and with whom people naturally
share their traumatic experiences. Also, linking different social coping
styles in the aftermath of a disaster to psychological adjustment would
substantially add to the existing body of knowledge available to practitioners
(Lyons, Mickelson, Sullivan, & Coyne, 1998; Pennebaker &
Harber, 1993).


Unexpected events, such as natural or man-made disasters, are not
easily studied using traditional laboratory methods. Random assignment,
pretest-posttest assessment, and control groups are often impossible
to employ. Consequently, most research on cultural upheavals
relies on descriptive and correlational analyses of people’s retrospective
accounts as assessed via questionnaires or interviews, days, weeks,
or months after the events (e.g., Dougall, Craig, & Baum, 1999; Galea et
al., 2002; Mikulincer, Florian, & Weller, 1993; Schlenger et al., 2002;
Schuster et al., 2001; Silver, Holman, McIntosh, Poulin, & Gil-Rivas,
2002). Self-reports, however, are often subject to substantial retrospective
distortions resulting from inaccuracies in recalling details of distant
events. These biases can be particularly troublesome when it
comes to emotional memories (A.A. Stone et al., 2000). Coyne and his
colleagues have repeatedly pointed to the problems and pitfalls of overreliance
on questionnaires when studying coping responses to trauma
and have argued for the employment of a broader range of methods
(e.g., Coyne & Gottlieb, 1996; Coyne et al., 2001).

We have recently introduced the Electronically Activated Recorder
(EAR; Mehl, Pennebaker, Crow, Dabbs, & Price, 2001), an eventsampling
tool designed for the naturalistic assessment of acoustic behavioral
traces. In recording 30-s snippets of ambient sound in people’s
immediate environment every 12 min for several days, the EAR
makes it possible to track people’s everyday social lives from an unobtrusive
observer’s perspective (Mehl & Pennebaker, 2003). It combines
the relative efficiency of experience-sampling approaches (Reis &
Wheeler, 1991) with the unique potentials of behavioral observation
techniques (Craik, 2000). The EAR demonstrates good convergent validity
with traditional methods for studying naturalistic social life, but
offers unique potentials for assessing subtle aspects of people’s social
interactions that normally go unnoticed (Mehl & Pennebaker, 2003).

The attacks of September 11, 2001, were events that caught the
United States unprepared. Serendipitously, we had started an EAR
project on the previous morning. The study was originally designed to
shed light on aspects of everyday coping with personal traumas. After
the collapse of the World Trade Center towers in the morning of September
11, we reoriented our research plan to study what was then the
beginning of a potential cultural upheaval. Additional participants
were recruited from a database of people who had worn the EAR in
the months prior.


The purpose of the study was to track people’s natural social interactions
during a time of national crisis. Participants wore the EAR
continuously for the first 10 days after September 11. Pre-September
11 baseline data were available for all participants from either the monitoring
that started the morning of September 10 or a previous EAR
study. The ambient sounds captured by the EAR were coded and analyzed
for the extent to which participants’ social interactions changed in
the days following September 11. Also, the time course of talk related to
the events of September 11 was mapped.

Further, we sought to identify psychologically healthy social-interaction
patterns during times of crisis. To address this question, we
linked aspects of participants’ social interactions in the aftermath of September
11 to their subsequent psychological adjustment as measured by
the experience of event-related intrusions and avoidance symptoms. Because
of the small sample size, however, these analyses must be considered



Six people began an EAR study on September 10, 2001. Within 2
hr of the first plane crash on September 11, these 6 participants provided
their consent to wear the monitor continuously for the following
10 days. Between 10:00 a.m. and 4:00 p.m. (Central Daylight Time)
on September 11, 8 more participants were recruited from a database
of people who had previously worn the EAR for a different project in
the months before. Three participants had missing data for more than
1 entire day and were excluded from the analyses. The final sample
(N=11) consisted of 8 undergraduate students (2 freshmen) and 3 recent
college graduates. Seven participants were female; the mean age
was 20.9 (SD=2.4). No participant had close family or friends in the
areas targeted by the terrorist attacks. Seven participants were born in
the United States; the other 4 were permanent residents who had lived
an average of 10.1 (SD=4.3) years in the United States.

EAR System

The EAR system consists of a digital voice recorder (SONY Memory
Stick ICD-MS1), an external microphone (OPTIMUS Tie Clip
Microphone), and a controller microchip. The microchip was programmed
with a cycle of 30 s on and 12.5 min off. The recorder was carried around
in a small shock-protected case either attached to the participant’s belt or
around the participant’s shoulder. The microphone was clipped to the
lapel of a jacket or the collar of a shirt. For further details, see Mehl et
al. (2001).

As discussed elsewhere (see Mehl & Pennebaker, 2003; Mehl et
al., 2001), the privacy and confidentiality of the participants’ data were
ensured by a cascade of safeguards–most notably, the opportunity to
erase any objectionable recording. Note that the EAR project was approved
by an emergency meeting of the University of Texas Institutional
Review Board on September 11.


Pre-September 11 baseline

For 5 participants of the final sample (N=11), the study started in
the morning of September 10. They were familiarized with all aspects
of the EAR. It was emphasized that the recorder sampled less than 4%
of the time and they would have the chance to listen to their recordings
and erase objectionable parts. Participants were instructed to wear the
EAR the entire day, taking it off only at night and when the proper
function of the device would be jeopardized. The EAR recordings sampled
on September 10 served as a pre-September 11 baseline for this
set of participants.

The pre-September 11 baseline for the remaining 6 participants recruited
on September 11 was taken from a previous EAR study in the
months before. This study had required wearing the EAR for 2 consecutive

September 11 session

The 5 participants who started on September 10 were originally
scheduled to return to the lab on September 11 between 9:00 and 10:00
a.m. On their arrival to the lab, they were informed about the change in
procedure and provided their consent to wear the EAR continuously
for 10 days. Immediately afterward, the investigators recruited the
other 6 participants from a database of former EAR study participants.
After they were informed about the study and provided consent, they
were equipped with a recorder and started the monitoring between
10:00 a.m. and 4:00 p.m.

EAR checks and final session

Participants returned to the lab approximately every third day to
ensure the functioning of the device. At these visits, they had the opportunity
to review their recordings and censor unwanted parts. The
monitoring was stopped on the morning of September 21. Participants
returned the EAR and completed two questionnaires: the Big Five Inventory
(BFI; John & Srivastava, 1999), a 44-item measure of the Big
Five personality traits (Cronbach’s α for all scales >.79), and the
Trait Meta-Mood Scale (TMMS; Salovey, Mayer, Goldman, Turvey,
& Palfai, 1995), a 48-item emotional intelligence questionnaire (Cronbach’s α = .92).
They were then thoroughly debriefed. Participants received
compensation of $10 per day.


The participants were contacted again on September 26 for an assessment
of their psychological adjustment to the events of September
11. They completed the Impact of Event Scale (IES; Horowitz, Wilner,
& Alvarez, 1979), a widely used 15-item instrument measuring symp
toms of avoidance and intrusion after a traumatic event (Dougall et al.,
1999; Mikulincer et al., 1993). On a 4-point scale (1=not at all
, 4=often ), participants indicated how often they thought about the events
when they did not mean to, how often they had dreams about the events,
and how often they tried not to think or talk about them. An overall impact
score was calculated by averaging participants’ responses across all
15 items. Cronbach’s alpha for the scale was .81.

Data Preparation

Judges’ coding of social participation

Research assistants listened to the participants’ complete set of
sound files and at the end of each 30-s interval coded the following
features of social participation: Was the participant alone, with one
person, or with a group of people? Was the participant talking on the
telephone? Was the television, radio, or stereo on? The judges further
coded whether captured conversations as well as television and radio
broadcasts were related to the events of September 11. Interjudge reliabilities
were calculated from a training set of 392 sound files coded
independently by six research assistants. For all categories, Cronbach’s
alpha was .90 or higher.

Data aggregation

With the pre-September 11 baseline included, the 11 participants
provided a total of 9,139 daytime recordings until midnight of September
20, the official end of the study. Of this total, 8,471 (92.7%)
files contained useable sound information; 497 (5.4%) had insufficient
sounds to perform valid coding, and 171 files (1.9%) were coded as
“participant not wearing the EAR.” All of a person’s valid daytime
codings were aggregated into a pre-September 11 baseline and five
post-September 11 two-day monitoring periods (see Fig. 1).


The aggregated social-participation variables were subjected to
simple six-cell one-way analyses of variance with time as the independent
variable. Polynomial linear and quadratic contrasts were performed
to determine the form of the effects.

Social Interactions in the Aftermath of September 11

Natural conversations

Overall, the total amount of time the EAR captured participants in
conversations did not change from before to after September 11,
F < 1, n.s. As depicted in Figure 1 (solid lines), however, there was a clear
change in the conversation pattern. While the relative amount of dyadic
conversations increased over time (p = .005), the proportion of
conversations showed a negative linear trend (p = .02). On a descriptive
level, the effects were most pronounced over the weekend—
when there was a maximum peak in dyadic conversations as well as a
minimum dip in group conversations. Both linear trends held up, however,
when the analyses were performed on the five weekday periods
only, excluding the weekend—dyadic: FL(1, 10) = 8.41, p = .02;
group: FL(1, 10) = 4.99, p = .05. Changes in telephone calls after
September 11 followed a weak negative linear trend ( p = .12). There
was a spike in telephone calls on September 11 and 12.

The analyses for conversations about September 11 revealed highly
significant quadratic trends for the total amount of conversations,
F(1,10) = 18.37, p = .002, as well as the relative amount of dyadic, group,
and telephone talk (see Fig. 1, dashed lines). On September 11 and 12,
participants talked about the events in approximately a quarter of their
conversations sampled by the EAR. Eight days later, at the end of the
monitoring, the proportion of September 11-related talk was down to
less than 5%. Over the intervening weekend, September 11-related
group conversations continued to decrease, the drop in event-related
telephone calls stagnated, and dyadic conversations about the terrorist
attacks increased to a level comparable to that on September 11 and 12.

Time spent with others

The effects for the time participants spent with others mirrored the
effects seen in the natural conversations. Although the total amount of
time participants spent with other people did not change after September
11, F < 1, n.s., there was a significant linear increase in the relative
amount of time spent with one other person as compared with a
group, F(1, 10) = 9.55, p = .01.


Over the first 10 days after September 11, then, participants did not
change their overall amount of interactions. Instead, they shifted from
group and telephone conversations to in-person dyadic encounters. Although
the amount of talking about September 11 was high immediately
after the events and low 8 days later across all types of conversations, in
dyadic encounters the events continued to be discussed over the intervening

Fig. 1. Changes in participants’ social interactions in the aftermath of September 11, 2001
(N=11). From top to bottom, the panels show the proportions of conversations that took
place in dyads, in groups, and on the telephone, for the baseline period and each of five 2-day
periods beginning on September 11. Each panel also graphs the percentage of conversations
that were about the attacks. The F and p values in the graphs refer to linear (index L) and quadratic
(index Q) trends in repeated measurement analyses of variance.

Other Social-Participation Data: Media Exposure After
September 11

There was a significant quadratic trend in the amount of time the
EAR captured a television, radio, or stereo on (see Fig. 2, solid line).
Total media exposure increased from baseline to September 11 and 12
(p=.06) and then returned to baseline toward the end of the monitoring.
The percentage of television or radio coverage about September
11 also followed a quadratic function (see Fig. 2, dashed line). Although
roughly half of the television and radio broadcasts captured by
the EAR on September 11 and 12 were related to the attacks, the proportion
dropped to less than 10% by the end of the monitoring.

Exploratory Analyses: Social Interactions and
Psychological Adjustment

On September 26, participants completed the IES. Roughly 2 weeks
after the events, they still reported a moderate amount of posttraumatic
stress symptoms in the form of September 11-related intrusions and
avoidance behavior (M=1.92, SD=0.55; 1=no symptoms, 4=frequent
symptoms). We ran a set of correlational analyses to determine
how changes in participants’ social interactions after September 11 were
related to their psychological adjustment to the events. The slope of the
linear regression equation calculated for each participant from the five
post-September 11 data points was used as a proxy for change.

The analyses revealed a marginally significant relationship between
participants’ change in the relative amount of dyadic talk and their subsequent
IES scores (r=-.54, p=.09). The more participants increased
their proportion of dyadic talk over the first 10 days, the less psychological
impact they reported. In a complementary way, decreases in
participants’ proportion of group conversations were (nonsignificantly)
related to less frequent experiences of posttraumatic stress symptoms

Fig. 2. Changes in participants’ media exposure in the aftermath of
September 11, 2001 (N=11). The graph shows the percentage of
sound files in which the television, radio, or stereo was on, during the
baseline period and each of five 2-day periods beginning on September
11. The graph also shows the percentage of broadcasts captured
that were related to the attacks. The F and p values refer to quadratic
(index Q) trends in repeated measurement analyses of variance.
(r=.51, p=.11). Both correlations held, and even increased slightly
in magnitude (dyadic: r=-.57, p=.07; group: r=.54, p=.09),
when the IES scores were adjusted for differences in the two available
dispositional coping measures, emotional stability (from the BFI) and
emotional intelligence (TMMS), hinting that our findings tap more
into “healthy coping” than into “healthy people coping.”

Because of the small sample size, however, these analyses are clearly
exploratory. Although no other aspect of participants’ social interactions
showed a reliable association with their IES scores (i.e., total amount of
post-September 11 conversations: r=-.36, p=.28; amount of September
11-related talk: r=-.25, p=.45; change in September 11-related
talk over time: r=.07, p=.85), the low statistical power prevents an interpretation
of these findings.


In the first 10 days after the attacks, people gradually shifted their
interactions from group conversations and phone calls to in-person dyadic
encounters. In a set of exploratory analyses, this natural shift toward
dyadic encounters tended to predict better subsequent psychological
adjustment in the form of fewer September 11-related intrusions and
avoidance behaviors.

Considering the frequent finding that humans seek out others under
stress, it was surprising that there were no overall increases in social
interactions in the aftermath of the September 11 attacks. This raises
the question of the extent to which the stress-affiliation link established
in the laboratory holds up in naturalistic settings. Typically, in
the lab, participants are introduced to a stressful scenario while they
are alone and are then given the option to be with another person (e.g.,
Schachter, 1959). In the real world, however, people are embedded
into an existing social network and have established their own ways of
spending time with others (Hobfoll, 1998; Mehl & Pennebaker, 2003).
A disaster then disrupts this habitual social life and thus might lead to
a change rather than an increase in affiliative behavior—a potential
confound in a typical lab design.

What might have driven participants to gradually shift toward more
in-person dyadic encounters, reducing their group conversations and
phone calls as time passed? Whereas the initial group interactions
could have maximized informational input and served the need to most
efficiently learn about what happened, the dyadic interactions later on
may have facilitated psychological coping with the events (Fredrickson,
Tugade, Waugh, & Larkin, 2003; Hobfoll & deVries, 1995; Peterson
& Seligman, 2003). Clearly, September 11 had the potential to
undermine people’s cultural worldviews. One-on-one encounters can
provide the intimacy needed to reaffirm a shaken worldview (Pyszczynski
et al., 2002), reevaluate and calibrate one’s beliefs and opinions (Festinger,
1954), and help one come to terms with what happened by
gradually developing a personal narrative (L.D. Stone & Pennebaker,
2002; Taylor, 1983; Wortman & Silver, 1989).

From this perspective, it would have been informative to track with
whom people were actually interacting. Both ethical and methodological
considerations thwarted this kind of analysis. The findings of
Rimé et al. (1998), however, suggest that for their one-on-one talks,
participants most likely relied on their partners and close friends. Taylor
et al. (2000) pointed to an interesting potential sex difference related
to this issue. They reported evidence that men predominantly fall
back on their partners as a resource for social support, whereas women
rely on a much larger social network. It would be interesting to see
whether the number of conversation partners behind the post-September
11 increase in dyadic interactions was larger for females than for
males (Lyons et al., 1998). The inability to track with whom people
were speaking and the small sample size precluded such analyses.

Given prior research on the salutary effects of social sharing (Rimé
et al., 1998), the exploratory findings that a gradual shift to more inperson
dyadic interactions tended to predict subsequent event impact
has potentially important conceptual as well as practical implications.
Participants benefited most when they initially had a high amount of
group interactions and later sought more one-on-one encounters. Also,
the psychologically healthy pattern in this study was the same pattern
that best described people’s natural reactions. We hope that future research
will replicate these findings and further clarify the differential
psychological role of dyadic versus group interactions in coping with
trauma. In the aftermath of disasters, this would provide critical input
for therapists in deciding how to best allocate the limited resources for
psychological interventions in emergencies (Bisson et al., 1997;
Brewin, 2001).

By the same token, the interaction data hint at the ways people are
in immediate contact with one another in the hours and days after an
upheaval. Knowing how people naturally congregate in times of crisis
is critical given current concerns about contagious agents (e.g., smallpox)
associated with bioterrorism (Dittman, 2002; Kuhr & Hauer, 2001).
If the current findings generalize to a larger population, the fact that
large-group gatherings drop and dyadic interactions increase in the aftermath
of a publicly announced disaster can be an important consideration
for the development of large-scale models of disease (or even information)

This study was the first to compare everyday social interactions in
the aftermath of a massive cultural upheaval with social interactions
during a predisaster baseline. With its close-meshed sampling of acoustic
behavioral traces, the EAR method allowed for an objective assessment
of basic features of people’s conversations. The amount of time
spent in various types of conversations is a subtle aspect of social life
that easily evades people’s attention and thus is difficult to capture using
self-reports (Mehl & Pennebaker, 2003). Clearly, supplementing
the EAR data with information on how participants experienced their
social interactions after September 11 (e.g., pleasantness, degree of intimacy,
degree of disclosure) would have been very helpful in clarifying
the psychological meaning of the shift toward more in-person
dyadic encounters (Reis & Wheeler, 1991). Also, a series of methodological
shortcomings of this ad hoc emergency study, such as the
small sample size, the lack of a nondisaster control group, and the limited
number of available measures, raises potential concerns about the
internal validity and generalizability of the findings.

It is the serendipitously unique methodology of this project, combined
with the public, political, and scientific importance of the events
of September 11, that makes us confident that these preliminary findings
may be useful to a broad spectrum of researchers and practitioners.
Future research that includes larger real-world samples and uses
observation methods such as the EAR is clearly needed.

Acknowledgments—We are indebted to Monica Adame, Esther Chong,
Michael Cohn, Pilar Halgunseth, Emily Hester, Elizabeth Lyons, Mary
Clare Madeley, Kim Noles, Linsey Proctor, Kim Roaten, and Liz Wuehrman
for help in collecting the data and coding the sound files and to Andrea
Garcia for her help in all phases of the project. The research was funded by
a grant from the National Institutes of Health (MH52391).

Prev Next