Social Science & Medicine 68 (2009) 2190–2198

Contents lists available at ScienceDirect

Social Science & Medicine

journal homepage: www.elsevier.com/locate/socscimed

Looking for resilience: Understanding the longitudinal trajectories

of responses to stressq
Fran H. Norris a, *, Melissa Tracy b, Sandro Galea b
a
Dartmouth Medical School, Psychiatry/NCPTSD, VA Medical Center, 215 North Main Street, White River Junction, VT 05009, USA
b
University of Michigan, MI, USA

a r t i c l e i n f o a b s t r a c t

Article history: Taking advantage of two large, population-based, and longitudinal datasets collected after the 1999
Available online 4 May 2009 floods in Mexico (n ¼ 561) and the September 11, 2001 terrorist attacks in New York (n ¼ 1267), we
examined the notion that resilience may be best understood and measured as one member of a set of
Keywords: trajectories that may follow exposure to trauma or severe stress. We hypothesized that resistance,
Resilience resilience, recovery, relapsing/remitting, delayed dysfunction, and chronic dysfunction trajectories were
Disaster
all possible in the aftermath of major disasters. Semi-parametric group-based modeling yielded the
Terrorism
Posttraumatic stress disorder (PTSD)
strongest evidence for resistance (no or mild and stable symptoms), resilience (initially moderate or
Mexico severe symptoms followed by a sharp decrease), recovery (initially moderate or severe symptoms fol-
USA lowed by a gradual decrease), and chronic dysfunction (moderate or severe and stable symptoms), as
these trajectories were prevalent in both samples. Neither Mexico nor New York showed a relapsing/
remitting trajectory, and only New York showed a delayed dysfunction trajectory. Understanding
patterns of psychological distress over time may present opportunities for interventions that aim to
increase resilience, and decrease more adverse trajectories, after mass traumatic events.
Ó 2009 Elsevier Ltd. All rights reserved.

It has long been observed that people often function remarkably hampered by two common problems in the research literature. One
better than objective circumstances suggest they should. For problem is the frequent assumption that resilience can be inferred
example, many youth experience academic success despite signif- on the basis of an absence of psychopathology at a single point in
icant childhood adversity (Cowen, Wyman, & Work, 1996; Gar- time. Resilience may be manifest in the absence of psychopa-
mezy, 1974). In the context of traumatic event experiences, most thology (or more completely in the presence of wellness; see Norris
first responders stay well despite the horrors of their work (Norris et al., 2008) but the concepts are not synonymous. There are
et al., 2002; van der Velden et al., 2006) and rates of posttraumatic multiple routes to good mental health, of which resilience is just
stress disorder (PTSD) in the general population are far lower than one. Moreover, the absence of psychopathology at one point in time
rates of exposure to potentially traumatic events (Breslau et al., does not assure that it was absent previously or will not occur at
1998). These observations have generated substantial interest in some later point in time. The second problem, related to the first, is
resilience, defined variously as the process of, capacity for, or investigators’ loose and post hoc usage of the term resilience to
outcome of successful adaptation after trauma or severe stress capture all unexamined reasons for failing to find psychopathology
(Butler, Morland, & Leskin, 2007; Egeland, Carlson, & Sroufe, 1993; following substantial traumatic stress.
Layne, Warren, Watson, & Shalev, 2007; Masten, Best, & Garmezy, However, in recent years there has also been considerable
1990; Norris, Stevens, Pfefferbaum, Wyche, & Pfefferbaum, 2008; progress on several fronts. These contributions come from various
Werner & Smith, 1982). disciplines but share in common a central concern with the timing,
Despite this longstanding interest in resilience, considerable duration, and course of responses to stress. First, there is growing
confusion remains with regard to how the concept is best consensus that resilience is better characterized as adaptability
researched. Advancements in understanding resilience have been than as stability (Adger, 2000; Klein, Nicholls, & Thomalla, 2003). In
other words, resilience is a process of ‘‘bouncing back’’ from harm
q This research was supported by Grants R01 MH51278, R01 MH66391, and P60
rather than immunity from harm (Garmezy, 1993; Layne et al.,
MH082598 from the National Institute of Mental Health.
2007). This image can be traced to its origins in mathematics and
* Corresponding author. Tel.: þ1 802 296 5132. physics, where resistance was defined as the force (stress) required
E-mail address: [email protected] (F.H. Norris). to displace a system from equilibrium, whereas resilience was

0277-9536/$ – see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.socscimed.2009.03.043
 F.H. Norris et al. / Social Science & Medicine 68 (2009) 2190–2198 2191

defined as the time required for the system to return to equilibrium serious injury (American Psychiatric Association, 1994), severe
once displaced (Bodin & Wiman, 2004). In physics, resilience has stress is a broader construct referring to various losses, uncer-
little to do with how large the initial displacement is or even how tainties, challenges and demands. In addition to resistance, resil-
severe the oscillations are but is more precisely the speed with ience, and recovery trajectories, there are at least three other
which homeostasis is achieved. Applying this analogy to the human trajectories of potential interest in research about the consequences
stress response, we should use the concept of resistance (not of stressful events: relapsing/remitting, in which symptoms display
resilience) to describe situations where dysfunction is minimal a cyclical course; delayed dysfunction, in which PTSD or some other
because coping resources have effectively blocked the stressor. trauma-related disorder emerges after considerable time has
Resistance is an unlikely course in the aftermath of extreme passed; and chronic dysfunction, where an initial stress reaction
stressors, such as disasters, where distress is nearly universal in the persists.
first weeks or months (Norris et al., 2002). Thus, resilience is Longitudinal research on trauma and bereavement substanti-
a different trajectory than is resistance. ates the presence of distinct symptom trajectories over time.
Second, it has become increasingly recognized that resilience Orcutt, Erickson, and Wolfe (2004) studied a predominantly male
should be differentiated from recovery. In his influential paper on sample of Gulf War veterans within five days of their return to the
psychological resilience, Bonanno (2004) characterized recovery as United States and approximately 2 and 6 years later. Two distinct
involving a period of dysfunction lasting several months or more, growth curves characterized the data. The largest group of veterans
followed by a gradual return to pre-event functioning. Resilience, showed low levels of PTSD symptoms initially and little change. The
he argued, may involve transient perturbations, lasting as long as other group showed slightly higher symptoms initially, followed by
several weeks, but generally involves a stable trajectory of healthy significant increases over time. These trends would be most
functioning. It is now commonly accepted in the disaster field that consistent with our proposed resistance and delayed dysfunction
some distress is a normal reaction to an abnormal event (Flynn, trajectories, but the long intervals may have made it difficult to
1994). Most of the time, however, transient dysfunction is followed capture other potential patterns and the nature of the sample
by a reasonably rapid return to pre-disaster levels of functioning precludes broad generalization. O’Donnell, Elliott, Lau, and Creamer
(Norris et al., 2002). Thus, resilience is a different trajectory than is (2007) studied a predominantly male sample of injury survivors
recovery. assessed prior to hospital discharge and 3 and 12 months post-
Putting these perspectives together, we suggest, as have some event. Patterns in their data also pointed to a larger resistant
others (Layne et al., 2007), that resilience may be best understood subgroup (these persons were low in PTSD symptoms at all time
and measured as a trajectory and, more to the point, as one member points) and a smaller subgroup with chronic dysfunction (these
of a set of possible trajectories that may follow exposure to trauma persons had higher levels of symptoms initially and grew more
or severe stress (see Fig. 1). Whereas trauma refers to a sudden symptomatic over time). Ott, Lueger, Kelber, and Prigerson (2007)
experience or confrontation with actual or threatened death or cluster analyzed longitudinal data (collected on average at 4, 9 and

Fig. 1. Hypothesized trajectories of the course of stress responses.

2192 F.H. Norris et al. / Social Science & Medicine 68 (2009) 2190–2198

18 months postevent), from a predominantly female sample of In this paper, we focus exclusively on trajectories of PTSD
older bereaved spouses. The largest group of participants showed symptoms. PTSD encompasses a range of re-experiencing, avoid-
‘‘common grief’’ by being high in symptoms at Wave 1 followed by ance, and arousal symptoms tied to a particular event (American
slow, steady improvements (recovery). The second ‘‘resilient’’ Psychiatric Association, 1994). It is among the most prevalent
group was least distressed initially but nonetheless showed modest psychological problems following major disasters (Galea, Nandi, &
additional improvement. The third and smallest group, ‘‘chronic Vlahov, 2005; Norris et al., 2002). For retrospective research on
grief,’’ was highly distressed at each time point, thereby showing resilience, the number of disaster-related PTSD symptoms affords
a chronic dysfunction trajectory. Taken together, these studies well the advantage of a hypothetical zero mean before the event.
illustrate Peleg and Shalev’s (2006) assertion that the study of
change in symptoms as an outcome of interest in its own right is Method
among the major shifts in PTSD research in recent years.
Despite recent advances, cross-sectional studies still make up Study 1: Mexico
the vast majority of research on the consequences of traumatic
events and generally cannot distinguish these trajectories from one Sampling and interviewing procedures
another. Nor can they distinguish delayed dysfunction from chronic Visits to the two selected communities revealed that identical
dysfunction, and they may miss relapsing/remitting trajectories sampling procedures would not be possible. In Villahermosa, the
altogether. At minimum, such distinction requires three postevent flood damage was extensive, and victims were dispersed across
waves to draw inferences about these trajectories; ideally, studies a large sector of the city. The context necessitated a probability
aiming to make this distinction would also provide pre-event sampling design to draw a sample of adults representative of the
measures. afflicted population. From affected census tracts in Villahermosa,
In this paper, we sought to examine the presence of these 653 households were sampled randomly in proportion to the tracts’
various trajectories in the aftermath of disaster. Disasters are population sizes. Of the 601 eligible households (non-eligible units
particularly interesting social phenomena where we can assess were vacant lots or businesses), 530 were successfully contacted
these trajectories in the general population. This context offers and the adult who answered the door was asked to provide
advantages relative to past research that has identified patterns in a sociodemographic interview about the household. Of these
highly selected samples such as male combat veterans and older households, 470 agreed to complete this initial interview. On the
widows. Since disasters affect large numbers of persons, we might basis of the most recent birthday, one adult resident was then
expect a range of trajectories to be observed that conform to all, or randomly selected from each participating household and asked to
most of, the trajectories we hypothesize here. Therefore, we participate in an in-depth psychological interview. Of these, 461
hypothesized that all patterns might occur with measurable completed the psychological interview, for a final Wave 1 response
frequency, but that resilience would be the most common pattern rate of 77% of those assessed as eligible and 87% of those actually
evidenced by populations affected by disasters. Moreover, we contacted.
posited that if resilience is the expected human response to stress, In Teziutlán, the stricken hillside communities were con-
it should occur reliably across disaster types and in different demned, and all families were relocated to a new community
cultural and resource contexts. To test these ideas, we examined the outside of the original city. The size of the community did not
trajectories of symptoms across two sharply different events in two necessitate sampling, and all households were included in the
different settings. sampling frame. Of the 235 households provided with plots in the
The first of these two events occurred in Mexico in 1999. In new community, 209 were successfully contacted. Only 1 house-
October of that year, a stationary tropical depression in the Gulf of hold refused the demographic interview. Of the 208 households
Campeche generated torrential rains, widespread flooding, and that completed the demographic interview, 205 participants
devastating mudslides in nine Mexican states. To capture the completed the psychological interview, for a final response rate of
variability in the way this event was experienced, we studied two 87% of those eligible and 98% of those actually contacted.
different communities: Villahermosa, the capital of the coastal state When the sample was compared to Mexican population data on
of Tobasco, population 500,000, and Teziutlán, a mountain city in key characteristics, the proportion of women in the sample was
the state of Puebla, population 180,000. These communities anchor higher than it should have been (55%). Analyses of the socio-
the geographic range of the disaster. The extent and duration of the demographic data indicated that the bias occurred at the point of
flooding were actually worse in Villahermosa, but the sudden and selection for the psychological interview, although the reason for
unexpected mudslides in Teziutlán caused dramatic losses, this was not clear. This selection was made at the end of the
bereavement, and trauma. Officials in Mexico characterized this demographic interview, well after the informant had provided the
event as the worst flooding disaster of the decade, if not of the birthdays, birth years, and present residence status of each
century; more than 400 people died, and at least 200,000 people household member. Fieldwork supervisors reviewed audiotapes of
lost their homes (Red Cross, 1999). each interview and verified that the interviewer selected the
The second of the two events was the terrorist attacks on New appropriate adult (the one with the most recent birthday) for the
York City in September 2001. On the morning of September 11, psychological interview regardless of who gave the sociodemo-
2001, four commercial airliners were hijacked from US airports. graphic interview or who was home at the time of that initial
Two of the planes crashed into the twin towers of the World Trade interview. Analyses of the household demographic data indicated
Center (WTC) in lower Manhattan, New York City. The WTC towers that female participants were quite representative of the larger
collapsed in the hours after they were hit. Although most people population of women, but male participants underrepresented
working inside the WTC evacuated the buildings soon after the younger, lower-income, less-educated men (who are perhaps less
planes hit, the collapse of the towers resulted in the death of residentially stable). With effect sizes (d) in the range of 0.09–0.12,
approximately 2800 people. Among the dead were office workers the magnitude of the bias appeared to be quite small. To derive an
who were in the buildings at the time of their collapse and rescue unbiased population estimate, weights were applied to correct the
personnel who had been climbing the towers in an attempt to bring sex distribution to a 55:45 ratio of women to men in each city.
survivors down to ground level. The attacks of September 11 were The initial interviews were conducted 6 months postdisaster, in
the largest single terrorist attack ever on US soil. April 2000. Attempts were made to re-interview all participants
 F.H. Norris et al. / Social Science & Medicine 68 (2009) 2190–2198 2193

12, 18, and 24 months postdisaster. In Villahermosa, 385 or 84% of (4, 2752) ¼ 0.31, ns, sex, severity of exposure, and PTSD status, c2s
the participants completed all four psychological interviews, as did (1, 2752) < 2.36, ns.
176 or 86% of the participants in Teziutlán. Attrition was unrelated Sampling weights were developed and applied to the data to
to Time 1 measures of city, sex, severity of exposure, and PTSD correct for potential selection bias relating to the number of
status, c2s (1, 666) < 1.50, ns. household telephones, persons in the household, and over-
All interviews were completed by trained, local interviewers in sampling. We also applied post-stratification weights to make the
respondents’ homes in private. The demographic interviews lasted follow-up survey samples demographically representative of the
about 1 h, and psychological interviews lasted an average of 2 h. NYC metropolitan area population according to the 2000 US
Demographic and psychological interviews were typically Census.
completed on separate days. Fieldwork managers later revisited
each participating household to deliver a letter of thanks and to ask Measures
the respondent for his or her impressions of the interview and We used the National Women’s Study (NWS) posttraumatic
interviewer. Study procedures were approved by institutional stress module questions to assess PTSD. This module assesses the
review boards in the United States (Georgia State University) and presence of Criteria B (re-experiencing, e.g. intrusive memories,
Mexico (University of Guadalajara and the Institute for Oaxacan distressing dreams), C (avoidance, e.g. efforts to avoid thoughts
Studies) and were reviewed for adherence to federal (U.S.) guide- associated with the trauma, loss of interest in significant activities),
lines for conducting research in international settings. and D (arousal, e.g. difficulty falling asleep or concentrating)
symptoms and determines content for content-specific symptoms
Measures (e.g., content of dreams or nightmares) if symptom presence is
Current (past 6-month) PTSD was measured by using a modified endorsed. Except for the time frame for reporting symptoms, the
version of Module K of Version 2.1 of the Composite International same measure was used at each wave. At the first interview,
Diagnostic Interview (CIDI), developed and translated into Spanish participants were asked about symptoms since the September 11,
by the World Health Organization (WHO, 1997). The CIDI has been 2001 terrorist attacks; at all follow-up interviews they were asked
used widely in prior epidemiologic studies, including a four-city about symptoms since the last interview. Symptom endorsement
study in Mexico (Norris et al., 2003). The CIDI assesses all DSM-IV was dichotomous (yes/no); the number of symptoms endorsed and
Criteria for PTSD (American Psychiatric Association, 1994) as they reported to be related to the September 11th attacks (range 0–17)
emerge after a specified event. To measure current disaster-related was calculated for each survey wave. The National Women’s Study
PTSD at each wave, the questions referred to symptoms attributed PTSD module was validated in a field trial against the PTSD module
to the flood and experienced within the past 6 months. A count of of the Structured Clinical Interview for DSM-III-R administered by
affirmative responses to CIDI symptom questions (range 0–17) mental health professionals. In the field trial, instrument sensitivity
provided a continuous measure of PTSD symptoms, a ¼ 0.89. was 99% and specificity was 79% (Kilpatrick et al., 1998).
For the purposes of this paper, we should note that the NWS
Study 2: New York City PTSD module is very similar to the CIDI PTSD module. Both
measures ask one question for each criterion symptom for a total of
Sampling and interviewing procedures 17 symptom questions, use a dichotomous (yes/no) response
We conducted a random-digit-dial household survey to recruit format, and capture symptoms experienced at any time within
baseline respondents approximately 6 months after the September a specified interval.
11 attacks (March 25–June 25, 2002). The sampling frame consisted
of all adults (18 years of age and older) in the NYC metropolitan Data analysis
area, including the following contiguous geographic areas: New
York City and Nassau, Westchester, Suffolk, and Rockland counties We used semi-parametric group-based modeling to identify
in New York State; Hudson, Essex, Bergen, Passaic, Union, Mid- trajectories of posttraumatic stress symptoms across survey waves
dlesex, Monmouth, Morris, and Somerset counties in New Jersey; in each of the two studies separately. All analyses were restricted to
Lower Fairfield county in Connecticut. Interviews were conducted participants who completed all four survey waves. For the Mexico
in English, Spanish, Mandarin, and Cantonese by trained inter- study, we fit censored normal models with a minimum of 0 and
viewers using translated and back-translated questionnaires and maximum of 17, since posttraumatic stress symptoms were
a computer-assisted telephone interview system. Households were approximately normally distributed in the sample. For the New
screened for eligibility by location. If eligible, an adult in each York City study, we fit zero-inflated Poisson regression models to
household was randomly selected by choosing the adult whose account for the greater number of zeros in the count of post-
birthday was closest to the interview date. Up to ten attempts were traumatic stress symptoms than would be expected under the
made to conduct the interview. The cooperation rate for the Poisson distribution. The Bayesian Information Criterion (BIC) was
baseline survey was 56% [cooperation rate ¼ (completed inter- used to select the best-fitting model, with changes in the approx-
views þ quota-outs þ screen-outs)/(complete interviews þ quota- imation to the Bayes factor greater than 10 indicating a better fit of
outs þ screen-outs þ refusals þ premature terminations)] and the the model with an increased number of groups (Jones, Nagin, &
overall response rate was 34% [response rate ¼ (completed inter- Roeder, 2001); the substantive importance of the trajectory groups
views þ partial interviews)/(all eligible residential telephone was also considered. After selecting the model with the optimal
numbers þ telephone numbers of unknown eligibility)]. Further number of trajectory groups, we determined the appropriate shape
details on sample selection are provided elsewhere (Galea et al., of each trajectory group (i.e., linear, quadratic, cubic) based on
2003). statistical significance (p < 0.05).
Three follow-up interviews were conducted approximately 6, In interpreting intercepts, we characterize averages of 0–3
18, and 30 months after the first interview (September 25, 2002– symptoms as no or mild, 4–8 symptoms as moderate, and 9þ
January 31, 2003, September 25, 2003–February 29, 2004, symptoms as severe. Although these cut-points are more-or-less
December 15, 2004–November 30, 2005). Of the 2752 original arbitrary, we chose them to impose consistency in description
participants, 1267 or 46% completed all four survey waves. Attrition across the two studies. This categorization was strongly related to
was unrelated to Time 1 measures of area of residence, c2 PTSD status in the Mexico data, c2 (4, 561) ¼ 536.71, p < 0.001. Of
2194 F.H. Norris et al. / Social Science & Medicine 68 (2009) 2190–2198

persons classified as having severe distress (9þ symptoms), 82% a group exhibiting 7 posttraumatic stress symptoms throughout
met all symptom criteria (B, C, and D), and 18% met one or two the study period and comprising 12% of the sample (Group 2,
symptom criteria. Of persons classified as moderately distressed moderate and stable), and a group exhibiting 12 posttraumatic
(4–8 symptoms), only 10% met all criteria, and 90% met one or two stress symptoms throughout the study and comprising 10% of the
criteria. Of persons classified as having no or mild distress (0–3 sample (Group 5, severe and stable). The remaining two groups
symptoms), none met all criteria, 53% met one or two symptom exhibited declines in symptoms. Group 3, comprising 32% of the
criteria (almost always re-experiencing, which requires only sample, showed a severe level of symptoms (10 symptoms) during
a single B symptom), and 47% met no criteria. With regard to the first 6 months after the disaster, but their symptoms declined to
percentages meeting all criteria for PTSD, including functioning and moderate levels (6 symptoms) by 12 months and to mild levels
duration as well as symptoms, these values were 51%, 3%, and 0% (3 symptoms) by 18 months, where they stabilized, thereby
for severe, moderate, and no-mild distress categories, c2 creating a quadratic trend. Group 4, comprising 11% of the sample,
(2, 561) ¼ 214.35, p < 0.001. showed symptoms as high as the stable severe group at 6 months
(13 symptoms), but their symptoms declined gradually in a linear
fashion throughout the study period, reaching a moderate level
Results (7 symptoms) by 24 months post. The mean posterior probabilities
of group assignment ranged from 0.702 to 0.926 for the five
Study 1: Mexico trajectory groups.

The 561 residents of Villahermosa and Teziutlán who completed Study 2: New York City
all four interviews averaged 37 years of age (SD ¼ 13) and 8 years of
education (SD ¼ 5) and were 55% female. Disaster-related The 1267 residents of the New York City metropolitan area who
bereavement was experienced by 28%, life threat by 70%, and much completed all four survey waves averaged 44 years of age (SD ¼ 16)
or enormous property damage by 48%. Of persons who reported at and were 55% female, 63% non-Hispanic White, 15% African
least one PTSD symptom, 68% reported that the problems began American, 12% Hispanic, and 9% Asian or other race/ethnicity. Only
within a week of the event, 81% within a month. Of these same 7% of the sample reported having less than a high school education,
persons, 48% reported that the problems stopped within 1 month of and 69% had at least some college education. Nearly a third of the
their onset. Across the four intervals, 17%, 10%, 8%, and 9% reported sample (29%) was directly affected by the September 11, 2001
having consulted a medical doctor about their disaster-related attacks, including being present in the World Trade Center at the
symptoms, and 9%, 6%, 3%, and 1% reported having consulted some time of the attacks, being physically injured, having a relative or
other professional. This profession was a natural healer or religious friend killed, having possessions lost or damaged, losing a job as
leader more often than a psychiatrist or psychologist (n ¼ 8). a result of the attacks, or being involved in the rescue or recovery
Model comparisons using the BIC indicated that five trajectory efforts. However, it is important to note that all participants
groups provided the best fit to the patterns of posttraumatic stress potentially experienced fear and uncertainty related to the attacks.
symptoms in the sample (Fig. 2). As shown in Table 1, three of the Across the four intervals, 1.3%, 1.5%, 1.4%, and 1.7% of the sample
five groups were defined by the intercept parameter only (Groups 1, reported having consulted a physician about their September 11th
2, and 5), indicating a stable number of symptoms throughout the attack-related symptoms, and 1.9%, 2.2%, 1.6%, and 1.2% reported
study period. These included a group characterized by approxi- having consulted a psychiatrist or psychologist. Additionally, 2.8%,
mately 2 posttraumatic stress symptoms during the study period 2.7%, 1.4%, and 1.9% reported having consulted some other profes-
and comprising 35% of the sample (Group 1, mild and stable), sional, most commonly a counselor, social worker, or minister.
Seven trajectory groups emerged as optimal to describe the
patterns of September 11, 2001 related posttraumatic stress
symptoms in the sample (Fig. 3 and Table 2). Group 1 comprised
40% of the sample and was characterized by virtually no post-
traumatic stress symptoms during the study period, defined by
only an intercept term. Two groups (Groups 3 and 6) exhibited
declines in posttraumatic stress symptoms; both were defined by
a quadratic trajectory and started from an initial level of approxi-
mately 4 symptoms at 6 months post. Group 3 (10% of the sample)
decreased to <1 symptom by 12 months and remained there for the
rest of the study period, whereas Group 6 (9% of the sample)
exhibited a more gradual recovery, with about 2 symptoms still
reported by group members by the end of the study period.
Three of the trajectory groups were characterized by increases in
posttraumatic stress symptoms during the study period, with cubic
trends providing the best fit for each of these three groups. Group 2
(13% of the sample) increased from 0 symptoms at Time 1 to an
average of 2 symptoms approximately 12 months after the
September 11th attacks and remained at this mild level of symp-
toms throughout the remainder of the study period. Group 4
(14% of the sample) exhibited a mild level of 2 symptoms during the
first 2 years but increased to a moderate level of 4 symptoms by the
end of the study period. Group 5 (10% of the sample) started at
Fig. 2. Trajectories of PTSD symptoms among residents of Villahermosa and Teziutlán
a moderate level of 5 symptoms and continued to increase,
in Mexico (n ¼ 561) after the 1999 flood. Numbers in parentheses refer to the wave of reporting a still moderate but higher level of 8 symptoms by the
assessment. end of the study period.
 F.H. Norris et al. / Social Science & Medicine 68 (2009) 2190–2198 2195

Table 1
Parameter estimates, prevalence, and mean posterior probability of assignment for each PTSD symptoms trajectory group among residents of Villahermosa and Teziutlán in
Mexico (n ¼ 561) after the 1999 flood.

Group Symptom trajectorya Parameter Estimate (SE)b p-Value Prevalence Mean posterior
probability (SD)c
1 Stable, mild Intercept 2.308 (0.161) <0.001 34.5% 0.926 (0.133)

2 Stable, moderate Intercept 6.881 (0.699) <0.001 12.0% 0.702 (0.163)

3 Decreasing, severe (1) Intercept 17.686 (0.991) <0.001 32.0% 0.821 (0.180)
to moderate (2) Linear 1.453 (0.130) <0.001 – –
Quadratic 0.036 (0.004) <0.001 – –

4 Decreasing, severe (1) Intercept 15.377 (1.355) <0.001 11.4% 0.818 (0.146)
to moderate (4) Linear 0.343 (0.079) <0.001 – –

5 Stable, severe Intercept 12.343 (0.393) <0.001 10.0% 0.827 (0.161)

a
Mild: 0–3 symptoms; moderate: 4–8 symptoms; severe:  9 symptoms; numbers in parentheses indicate survey wave.
b
Standard error.
c
Standard deviation.

Finally, as in the Mexico study, one group was characterized by remitting trajectory, and only New York showed a delayed
a consistently severe level of posttraumatic stress, approximately dysfunction trajectory (Group 4).
12 symptoms, throughout the study period (Group 7; 3% of the
sample); a quadratic trajectory provided the best fit for the pattern
exhibited by this group. The mean posterior probabilities of group Discussion
assignment ranged from 0.806 to 0.937 for the seven trajectory
groups. Taking advantage of two longitudinal datasets, we sought in this
study to examine the notion that resilience may be best understood
and measured as one member of a set of trajectories that may
Observed and hypothesized trajectories
follow exposure to trauma or severe stress. We had hypothesized
that resistance, resilience, recovery, relapsing/remitting, delayed
Table 3 relates the trajectories observed in Mexico and New York
dysfunction, and chronic dysfunction trajectories were all possible
back to the hypothesized trajectories shown in idealized form in
in the aftermath of major disasters. With one exception (relapsing/
Fig. 1. The strongest evidence emerged for resistance (mild and
remitting), all of the hypothesized trajectories (Fig. 1) occurred in
stable symptoms: Mexico Group 1, New York Groups 1 and 2),
the data with measurable frequency.
resilience (moderate or severe symptoms at the first postevent
Even though both of the studied events were quite serious, from
interview followed by a sharp decrease: Mexico Group 3, New York
one third (Mexico) to one half (New York) of these participants
Group 3), recovery (moderate or severe with gradual decrease:
exhibited resistance by never showing more than mild distress,
Mexico Group 4, New York Group 6), and chronic dysfunction
operationally defined as 3 PTSD symptoms. On the basis of past
(moderate or severe and stable symptoms: Mexico Groups 2 and 5,
disaster research that has focused on sample averages, we expected
New York Groups 5 and 7) trajectories, which were prevalent in
to see relatively little resistance, as many longitudinal studies have
both samples. Neither Mexico nor New York showed a relapsing/
shown a pattern of higher initial symptoms that dissipate over time
(see Norris et al., 2002). The present analyses, however, have shown
that a substantial subset of the stricken population may exhibit
trends that do not conform to this generalized result. As defined
here, resistance does not preclude the existence of some mild
distress or even the presence of minor fluctuations within the mild
range. Although we have characterized this trajectory as resistance,
it is similar to the definition of resilience provided by Bonanno
(2004) as an individual’s capacity to maintain healthy symptom-free
functioning following stressful life events. However, in keeping with
the important distinction between stability and adaptability for
understanding divergent responses to stress, we argued that
stability would be better characterized as resistance and adapt-
ability as resilience, and therefore the latter concept would be more
appropriately used to describe the capacity to quickly rebound from
an initial experience of distress created by an environmental change.
Perhaps because of the unexpectedly high prevalence of resis-
tance, the prevalence of resilience was lower than anticipated, at
least in New York, where the prevalence was only 10%. This group
never showed severe distress but rather began with moderate
distress, rapidly improved to mild distress, and subsequently
showed no distress. In Mexico, the prevalence of resilience was
substantial (32%); this group initially showed severe distress,
Fig. 3. Trajectories of PTSD symptoms among residents of the New York City metro-
operationally defined as 9 PTSD symptoms, but improved rapidly,
politan area (n ¼ 1,267) after the September 11, 2001 attacks. Numbers in parentheses showing only moderate distress (4–8 PTSD symptoms) at Wave 2.
refer to the wave of assessment. However, it took another few months for this group to resemble the
2196 F.H. Norris et al. / Social Science & Medicine 68 (2009) 2190–2198

Table 2
Parameter estimates, prevalence, and mean posterior probability of assignment for each PTSD symptoms trajectory group among residents of the New York City metropolitan
area (n ¼ 1,267) after the September 11, 2001 attacks.

Group Symptom trajectorya Parameter Estimate (SE)b p-Value Prevalence Mean posterior
probability (SD)c
1 Stable, mild Intercept 1.847 (0.174) <0.001 40.1% 0.921 (0.135)

2 Increasing, mild (1) Intercept 7.617 (3.239) 0.019 13.3% 0.806 (0.179)
to mild (2) Linear 1.108 (0.433) 0.011 – –
Quadratic 0.044 (0.016) 0.006 – –
Cubic 0.001 (0.0002) 0.003 – –

3 Decreasing, moderate (1) Intercept 3.210 (0.573) <0.001 10.1% 0.834 (0.175)
to mild (2) Linear 0.291 (0.066) <0.001 – –
Quadratic 0.004 (0.001) <0.001 – –

4 Increasing, mild (1) Intercept 2.360 (0.812) 0.004 14.3% 0.829 (0.175)
to moderate (4) Linear 0.335 (0.133) 0.011 – –
Quadratic 0.015 (0.006) 0.015 – –
Cubic 0.0002 (0.00008) 0.029 – –

5 Increasing, moderate (1) Intercept 0.320 (0.504) 0.525 9.9% 0.878 (0.139)
to moderate (4) Linear 0.195 (0.056) 0.001 – –
Quadratic 0.007 (0.002) 0.001 – –
Cubic 0.00009 (0.00003) 0.001 – –

6 Decreasing, moderate (1) Intercept 0.885 (0.305) 0.004 9.3% 0.862 (0.152)
to mild (4) Linear 0.079 (0.025) 0.001 – –
Quadratic 0.002 (0.001) <0.001 – –

7 Stable, severe Intercept 2.049 (0.215) <0.001 3.1% 0.937 (0.104)

Linear 0.039 (0.015) 0.008 – –
Quadratic 0.001 (0.0002) 0.002 – –
a
Mild: 0–3 symptoms; moderate: 4–8 symptoms; severe:  9 symptoms; numbers in parentheses indicate survey wave.
b
Standard error.
c
Standard deviation.

resistant group by showing only mild distress. This interpretation because the two groups’ initial level of symptoms was virtually
of the data from Mexico reflects a somewhat relativistic concep- identical. Although resistance, resilience, and recovery trajectories
tualization of resilience, as even in the resilient group, the have different implications for illness burden, they may be
‘‘bouncing back’’ took place over many months. It could be argued considered together as good outcomes. Their combined prevalence
that Mexico had two recovery subgroups and no incontrovertible was high in both Mexico (78%) and New York (72%).
trajectory of resilience. Neither the Mexico nor the New York analysis yielded a group
Theoretically, recovery differs from resilience primarily in the that could be characterized as relapsing/remitting. Our studies may
speed of improvement, i.e., recovery takes place more gradually, have lacked the ability to detect this trajectory, as the measures
over a longer interval of time. The prevalence of recovery in Mexico were unlikely to identify individuals who worsened, but then
was 11%; this group improved from severe to moderate distress improved, within an interval. And, possibly, the duration of the
during the course of the study, which concluded at 2 years post- studies (2 years in Mexico, 3 in New York) was not long enough to
disaster; extrapolation of the recovery trajectory indicates that it detect later relapses. Examples of relapsing/remitting trajectories
might take another year or two (i.e., until 3–4 years postdisaster) are few in disaster research, but Phifer and Norris (1989) alluded to
for this group to reach the mild levels of distress experienced by the such patterns in their longitudinal study of older adults after a
resistant and resilient groups at 2 years post. The prevalence of late-spring flood in Eastern Kentucky. In that study, mean symp-
recovery in New York (9%) was about the same as in Mexico, and toms peaked at the second postdisaster wave (in the spring after
most of this group’s improvement took place between 2 and 3.5 the flood) and showed a secondary peak two waves later during the
years post. In New York, the difference between the resilience following spring, raising the question that the heavy rains typical of
(Group 4) and recovery (Group 6) trajectories was especially clear spring in the mountains served as reminders of the earlier, more
severe flood. Anecdotally, service providers often express concerns
about ‘‘anniversary effects’’ that could be evidenced as relapsing/
Table 3 remitting patterns.
Observed trajectories in Mexico and New York classified according to the hypoth-
Perhaps the most surprising finding of the study was the
esized trajectories.
substantial prevalence of delayed dysfunction in New York (14%).
Hypothesized Mexico New York However, the New York results are actually quite consistent with
trajectory (Fig. 1)
Observed trajectory Total % Observed trajectory Total % the conclusions drawn by Andrews, Brewin, Philpott, and Stewart
(Groups, Fig. 2) (Groups, Fig. 3) (2007) on the basis of their review of the literature on delayed-
Resistance 1 34.5 1, 2 53.4 onset PTSD. They noted that delayed onset of PTSD is rare in the
Resilience 3 32.0 3 10.1 absence of any prior symptoms, but accounts for approximately 15%
Recovery 4 11.4 6 9.3
of civilian trauma cases when delayed onsets are defined to include
Relapsing 0.0 0.0
/remitting exacerbations of prior symptoms. They also noted that little is
Delayed 0.0 4 14.3 known about what distinguishes the delayed form of the disorder
dysfunction from the more typical immediate-onset form.
Chronic 2, 5 22.0 5, 7 13.0 Finally, chronic dysfunction trajectories were evident in both
dysfunction
Mexico (22%) and New York (13%). In Mexico, this included one
 F.H. Norris et al. / Social Science & Medicine 68 (2009) 2190–2198 2197

group with stable and severe symptoms (10%) and a second group and CIDI PTSD modules are not identical, and the methods of data
with stable and moderate symptoms (12%). In New York this collection differed (phone in New York, face-to-face in Mexico).
included one group with stable and severe symptoms (3%) and Another shortcoming is that the measures captured only the
a second group with moderate symptoms that worsened but stayed presence rather than severity of each symptom. Additionally,
within the moderate range (10%). Although they constitute only group-based trajectory modeling may find only local minima,
a minority of the population, individuals who fail to recover are leaving the possibility of a different true pattern of trajectory
extremely important for planning health and mental health care. groups than observed in these analyses; however, the optimal
The identification of predictors of group membership is beyond the number of trajectory groups for each study was robust across
purpose of this particular paper, but past research suggests that different starting values for each trajectory, which provides some
a trajectory of chronic dysfunction is likely the result of severe evidence that the procedure did not find only local minima (Jones
initial exposure to the disaster in combination with ongoing et al., 2001).
adversities and secondary stressors (Galea, Tracy, Norris, & Coffey, Our paper also has three limitations in terms of its scope. First,
2008; Norris et al., 2002). we focused solely on individual-level differences in recovery
Our studies had several key strengths, especially compared to trajectories, but we do not mean to imply that resilience is funda-
the norms in disaster research (see Norris, 2006). First, we used mentally a characteristic of the person. A host of adaptive capacities
data from two studies with four-wave longitudinal designs. Only (e.g., economic resources, social capital) functions at the commu-
about one in four disaster studies are longitudinal and only one in nity level to promote or impede individual and population resil-
ten have at least three postdisaster waves, the minimum number of ience and recovery (Norris et al., 2008). Second, we focused solely
time points required to distinguish the hypothesized trajectories on the total number of PTSD symptoms as the outcome of interest.
from one another. Second, both the Mexico and New York studies Different trajectories might have emerged for different PTSD clus-
were population-based. Randomly selected population samples ters, such as re-experiencing or avoidance, or for different
account for only about one in five disaster studies, and previous outcomes, such as depression or somatic complaints. Nor did we
studies of the trajectories of trauma recovery have focused exclu- consider the related question of posttraumatic growth. Third, we
sively on selected groups, such as male combat veterans, hospi- focused solely on determining the prevalence of the hypothesized
talized injury survivors, or older bereaved spouses. This attribute of trajectories, rather than on their determinants. Identifying the
our studies allowed us to consider the prevalence of various environmental, social, psychological, and biological determinants
trajectories in the general population. Third, although we lacked of these trajectories is an important direction for future research.
pre-disaster measures, our measures of PTSD were event-specific In conclusion, our research helps to define ‘‘resilience’’ not as
and thus overcame the problem to the extent possible. It is worth a somewhat circular explanation for the absence of psychopa-
noting that our assessment of trajectories rests on the assumption thology, but as one specific trajectory that can be identified,
that we are capturing only new postevent symptoms (there is measured, distinguished from other trajectories, and studied across
a hypothetical 0 symptoms point in each group at Month 0). This stressors and settings. Distinctions between various postevent
would not be the case if we had assessed depression or anxiety symptom trajectories are important because it is quite likely that
symptoms that occur with some frequency in populations and are the three trajectories that yield good long-term outcomes (resis-
not explicitly linked to disaster or trauma. However, we cannot rule tance, resilience, recovery) have different determinants and are
out the potential for false attributions of the cause or onset of best promoted by different intervention strategies introduced at
symptoms. Fourth, our measures of PTSD assessed intervals (e.g., different times. Postdisaster interventions must occur at multiple
past 6 months) rather than points (e.g., past week). If we had levels (individual, family, community) and evolve over time, as
assessed only those symptoms present at the times of the inter- needs change. Whereas resistance is the hypothetical ideal, the best
views, we would have missed PTSD symptoms that were present possible outcome of mass trauma intervention is not always
earlier in the intervals and possibly failed to detect resilience. This resistance, nor is it always resilience. Nevertheless, appropriate
feature was especially important at Wave 1, as we captured interventions should increase the likelihood of resilience among
symptoms experienced within days of the event. The retrospective people who are not resistant and the likelihood of recovery among
nature of the assessment was an unavoidable shortcoming of these people who are not resilient (Norris & Stevens, 2008). Both
studies. More assessments of shorter intervals would provide secondary and tertiary prevention strategies, if efficacious and
superior data, but the cost and participant burden are prohibitive. effective, have a place in the continuum of postdisaster health care.
Finally, the focus on mass disaster as the stressor largely eliminates
concerns that the stressor, or the timing of its occurrence, is References
somehow confounded with person characteristics or pre-existing
developmental trajectories. By considering two different types of Adger, W. (2000). Social and ecological resilience: are they related? Progress in
Human Geography, 24, 347–364.
disasters in different contexts (Mexico, natural disaster; New York, American Psychiatric Association. (1994). Diagnostic and statistical manual of mental
terrorist attack), we gain greater confidence in the generalizability disorders (4th ed.). Washington, DC: Author.
of the results. However, these results do not necessarily generalize Andrews, B., Brewin, C., Philpott, R., & Stewart, L. (2007). Delayed-onset post-
traumatic stress disorder: a systematic review of the evidence. American Journal
to other types of events, such as sexual assault or other forms of of Psychiatry, 164, 1319–1326.
interpersonal violence. Bodin, P., & Wiman, B. (2004). Resilience and other stability concepts in ecology:
Other than the shortcomings previously acknowledged, the notes on their origin, validity, and usefulness. ESS Bulletin, 2, 33–43.
Bonanno, G. (2004). Loss, trauma, and human resilience: have we underestimated
primary weakness of the approach used here was its subjectivity. the human capacity to thrive after extremely aversive events? American
The labels applied in Figs. 2 and 3 were value-neutral, but the act of Psychologist, 59, 20–28.
relating the observed trajectories back to the hypothesized trajec- Breslau, N., Kessler, R., Chilcoat, H., Schultz, L., Davis, G., & Andreski, P. (1998).
Trauma and posttraumatic stress disorder in the community. The 1996 Detroit
tories involved interpretation. As we noted in the case of Mexico,
Area Survey of Trauma. Archives of General Psychiatry, 55, 627–632.
the resilience and recovery trajectories were not as dramatically Butler, L., Morland, L., & Leskin, G. (2007). Psychological resilience in the face of
different as would be ideal. In real life, the timing of recovery is terrorism. In B. Bongar, L. Brown, L. Beutler, J. Breckenridge, & P. Zimbardo
more likely to be a continuum than a categorical variable. Another (Eds.), Psychology of terrorism (pp. 400–417). NY: Oxford University Press.
Cowen, E., Wyman, P., & Work, W. (1996). Resilience in highly stressed urban
potential issue is that between-setting differences could be children: concepts and findings. Bulletin of the New York Academy of Medicine,
attributable to methods. Although they are quite similar, the NWS 73, 267–284.
2198 F.H. Norris et al. / Social Science & Medicine 68 (2009) 2190–2198

Egeland, B., Carlson, E., & Sroufe, L. (1993). Resilience as process. Development & Norris, F., Friedman, M., Watson, P., Byrne, C., Diaz, E., & Kaniasty, K. (2002). 60,000
Psychopathology, 5, 517–528. Disaster victims speak: part I, an empirical review of the empirical literature,
Flynn, B. (1994). Mental health services in large scale disasters: an overview of the 1981–2001. Psychiatry: Interpersonal and Biological Processes, 65, 207–239.
Crisis Counseling Program. NCP Clinical Quarterly, 4, 1–4. Norris, F., Murphy, A., Baker, C., Perilla, J., Gutierrez-Rodriguez, F., & Gutierrez-
Galea, S., Nandi, A., & Vlahov, D. (2005). The epidemiology of post-traumatic stress Rodriguez, J. (2003). Epidemiology of trauma and posttraumatic stress disorder
disorder after disaster. Epidemiologic Reviews, 27, 1–14. in Mexico. Journal of Abnormal Psychology, 112, 646–656.
Galea, S., Tracy, M., Norris, F., & Coffey, S. (2008). Financial and social circumstances, Norris, F., & Stevens, S. (2008). Community resilience and the principles of mass trauma
the incidence and course of PTSD in Mississippi during the first two years after intervention. Psychiatry: Interpersonal and Biological Processes, 70, 320–328.
Hurricane Katrina. Journal of Traumatic Stress, 21, 357–368. Norris, F., Stevens, S., Pfefferbaum, B., Wyche, K., & Pfefferbaum, R. (2008).
Galea, S., Vlahov, D., Resnick, H., Ahern, J., Susser, E., Gold, J., et al. (2003). Trends of Community resilience as a metaphor, theory, set of capacities, and strategy for
probable post-traumatic stress disorder in New York City after the September disaster readiness. American Journal of Community Psychology, 41, 127–150.
11 terrorist attacks. American Journal of Epidemiology, 158, 514–524. O’Donnell, M. L., Elliott, P., Lau, W., & Creamer, M. C. (2007). PTSD symptom
Garmezy, N. (1974). The study of competence in children at risk for severe trajectories: from early to chronic response. Behaviour Research and Therapy,
psychopathology. In E. Anthony, & C. Koupernik (Eds.), Children at psychiatric 45(3), 601–606.
risk. The child in his family, Vol. 3 (pp. 77–97). New York: Wiley. Orcutt, H. K., Erickson, D. J., & Wolfe, J. (2004). The course of PTSD symptoms among
Garmezy, N. (1993). Children in poverty: resilience despite risk. Psychiatry: Inter- gulf war veterans: a growth mixture modeling approach. Journal of Traumatic
personal and Biological Processes, 56, 127–136. Stress, 17, 195–202.
Jones, B. L., Nagin, D. S., & Roeder, K. (2001). A SAS procedure based on mixture Ott, C. H., Lueger, R. J., Kelber, S. T., & Prigerson, H. G. (2007). Spousal bereavement in
models for estimating developmental trajectories. Sociological Methods & older adults: common, resilient, and chronic grief with defining characteristics.
Research, 29, 374–393. Journal of Nervous and Mental Disease, 195(4), 332–341.
Kilpatrick, D. G., Resnick, H. S., Freedy, J. R., Pelcovitz, D., Resick, P., Roth, S., et al. Peleg, T., & Shalev, A. Y. (2006). Longitudinal studies of PTSD: overview of findings
(1998). The posttraumatic stress disorder field trial: evaluation of the PTSD and methods. CNS Spectrums, 11(8), 589–602.
construct criteria A through E. In T. Widiger, A. Frances, & H. A. Pincus (Eds.), Phifer, J., & Norris, F. (1989). Psychological symptoms in older adults following
DSM-IV sourcebook, Vol. 4 (pp. 803–844). Washington, DC: American Psychiatric disaster: nature, timing, duration, and course. Journal of Gerontology: Social
Association Press. Science, 44, 207–217.
Klein, R., Nicholls, R., & Thomalla, F. (2003). Resilience to natural hazards: how Red Cross. (1999). Mexico flood victims mourn dead, plead for help. Retrieved
useful is this concept? Environmental Hazards, 5, 35–45. October 11, 1999 from. www.DisasterRelief.org/Disasters/991011Mexico.
Layne, C. M., Warren, J. S., Watson, P. J., & Shalev, A. Y. (2007). Risk, vulnerability, van der Velden, P., Christiaanse, B. B. A., Kleber, R. J., Marcelissen, F. G. H.,
resistance, and resilience: toward an integrative conceptualization of post- Dorresteijn, S. A. M., Drogendijk, A. N., et al. (2006). The effects of disaster
traumatic adaptation. In M. Friedman, T. Keane, & P. Resick (Eds.), Handbook of exposure and post-disaster critical incidents on intrusions, avoidance
PTSD: Science and practice (pp. 497–520). New York: Guilford Press. reactions and health problems among firefighters: a comparative study.
Masten, A., Best, K., & Garmezy, N. (1990). Resilience and development: contribu- Stress, 9(2).
tions from the study of children who overcome adversity. Development & Werner, E. E., & Smith, R. S. (1982). Vulnerable but invincible: A longitudinal study of
Psychopathology, 2, 425–444. resilient children and youth. New York: McGraw-Hill.
Norris, F. (2006). Disaster research methods: past progress and future directions. World Health Organization. (1997). Composite International Diagnostic Interview
Journal of Traumatic Stress, 19, 173–184. (CIDI). Version 2.1. Geneva: World Health Organization.