Expressing Thoughts and Feelings Following a Collective Trauma: Immediate Responses to 9/11 Predict Negative Outcomes in a National Sample

Collective traumas can negatively affecl large numbers of people who ostensibly did not experience events direc1ly, making it particularly important 10 identify which people are most vulnerable to developing mental and physical health problems as a result of such events. It is commonly believed 1ha1 successful coping with a traumatic event requires expressing one's thoughts and feelings ahout the experience, suggesting that people who choose not 10 do so would be a1 high risk for poor adjustment. To test this idea in the context of collective rraumn, 2.138 members of n na1ionaUy representative Web-enabled survey panel were given the opportunity 10 express 1heir reactions 10 the terrorist altacks of September 11, 200 I, on that day and those following. Follow-up surveys assessing menial and physical health outcomes were completed over the next 2 years. Contrary 10 common helief, participants who chose nol 10 express any initial reaction reported belier outcomes over time than did those who expressed an initial reac1ion. Among those who chose 10 express 1heir immediate reactions, longer responses predicted wor.;e outcomes over time. Implications for myths of coping, posttrauma interventions, and psychology in the media arc discussed.

It is commonly believed in clinical practict! and public consciousness that expressi ng one's thoughts and feelings about a traumatic event is necessary for successful coping (e.g., Everly & Mitchell, 1999; see Wortman & Boerner, 2007;Wortm an & Silver, 1989, for additional discussion). TI1e mental and physical health benefits of experi mental discl osure are supported by re· search fi nd ings (Frattaroli, 2006), but in the domain of early psychological intervention after trauma, widespread clinical application has outpaced rigorous research that has successfully demonstrated the efficacy of encouraging expression (McNally, Bryant. & Ehlers, 2003). The existing research into the benefits of expression has 1101 addressed a closely related aspect of the assumption: choosing not to express in the early aftermath of a trauma is actively harmful if not pathological. If true, people who make a choice not to express when given an opportunity to do so should be at particularly high risk for poor outcomes over time. In conlrast. if people w ho are more distressed arc more likdy lo express (e.g., see Pennebaker, Zech, & Rime, 2001 ), choosing not to express in the wake of a trauma may renect resilience rather lhan vulnerability. This question has important implications for the relationsh ip between expression and coping with trauma as well as for determining efficient allocation of clinical rcsourc~s following large-scale community disasters with many affected survivors.
Such la rge-scale collective traumas are the focus of lhc present investigation. The terrorist attacks of September 11 , 2001. represent a collective trauma in that people across lhe country suffered adverse effects, even though the vast majori ty of them did nol suffer direct and tangible losses (Marshall et al., 2007;Schlenger ct al., 2002;Schuster el al., 2001 ;Silver, Holman, Mcintosh, Poulin, & Gil-Rivas, 2002). In contrast to other traumas that may be experienced more intensely by a given individual than by the surrounding community (e.g., violent assault, death of a loved one), a collective trauma is more likely to create a similar situation across the community. Expression after an individual trauma is likely to be directed toward an audience of unaffected observers, whereas expression after a collective trauma is likely to be directed towa rd an audience that is also affected by the event (cf. Pennebaker & Harber, 1993).
Beliefs about the importance of expression hold a prominent place in the intersection between psycholog. ical science and public understanding. Keith Ablow, a psychiatrist featured on the nationally broadcast Today television show. exemplified this in his expres.~ion of the comments that opened this article, afh.:r a student gunman shot and killed 32 people on the campus of Virginia Polytechnic Institute and Stale University (Virgi nia Tech). Most members of the Virginia Tech community were not physically injured and did not directly witness the shootings, but they were nonetheless affected. Statements like this, which ostensibly represent the state of scientific knowledge, an: likely to garner media attention, especiall y in the wake of a collective trauma.
Clear answers about expression and collective trauma are thus doubly important. First, it is not obvious which individuals among the large number who have not experienced ··direct" loss in the immediate afterm ath of a collective trauma should be most likely to experience lasting negative effects. Understanding the relationship between willingness to express and long-term outcomes can help to remedy thjs problem. Second, collective traumas are instances in which psychological science is not onl y relevam but also brought to the cente r of public attention. Faulty conclusions dispersed in the media may harm laypeople who act on them of their own act·ord as well as harm the field of psychology just as its practitioners strive to build a case for research support in times of limited governmental funding.
To this end, the cu rrent study tested the relationship between immediate posttrauma expression choices and 2-year longitudinal mental and physical health outcomes in a nationally representative sample ro11owing a collective trauma: the terrorist attacks of Septembe r 11, 2001. The critical issue is whether lack of expression in the face of a collective trauma reflects vulnerability or resilience.

Myths of Copi11g
In their reviews of coping with loss, Silver (1989, 2001) concluded that .. myths of coping .. exist, based on the observation that several common assumptions about the process or coping wi th bereavement are unsupported by empirical data. Bereavemen t and trauma typically overlap (Strocbc, Schut, & Stroebe, 1998); when extending these argumen ts to the context of coping with traumatic events, two clements arc particularly relevant: (a) failing to exhibit distress is problematic: and (b) it is important to "work through .. or come to terms with the negative experience. Expressing one ·s thoughts and feelings about a trauma shm1ld play a key role in both of these clement<>, which suggests that failing to express should predict poor adjustment. Despite assertions that lack of observable grief is pathological (e.g., Horowitz, 1990), and frequent endorsement among clinicians of the existence of ·'delayed grief reactions" in which initial denial or inhibition of distress results in later maladaptive resurgence (Middleton, Moylan, Raphael, Burnett. & Martinek, 1993), such beliefs an.: not supported by empirical evidence (Wortman & Boerner, 2007;Wortman & Silver, 200 1). Indeed, there is little evidence that expression of emotions has any beneficial effect following bereavement (Stroebe, Stroebe, Schul Zech, & van den Bout, 2002), and there is some evidence that it may even impede successful coping (Bonanno & Ke ltner, 1997).

Benefits of Experiment(I/ Disclos11re
Nonetheless, a growing body of research has dcmonstr:lled benefits of experimental disclosure, in which participants typically are randomly assjgned either to express their thoughts and feelings about a personally meaningful topic (e.g., a traumatic event) or to perform a control task. Pennebaker and Beall ( 1986) conducted the seminal study in this tradition, showing that participants in a trauma-expression condition reported fewer illness-related doctor visits than others did. In a recent meta-analysis, Frattaroli (2006) reviewed 146 randomized disclosure studies. Results revealed an overall benefit of expression. Breaking down the existing research into categories by dependent variable further revealed the following: (a) an overall benefit for psychological health, including specific benefiL~ for distress, depression, positive function ing, anger, and anxiety; and (b) an overall benefit ror reported physical health, including benefits ror specific disease outcomes and illness behavior:..
Two poims important for the domain of early coping with trauma arc noteworthy. First, no significant effects emerged in Frartaroli' s (2006) meta-analysis for the general psychological "i.1rcss" category or dependent variables or its componcnL posttraumatic stress (P'TS) symptoms. which is a key trauma-related outcome. Second, the mean amount of time between the target event and disclosure was 15 months, well beyond the timeframe of early intervention. These points indicate that this meta-analysi!> and the research on which it is based do not provide a definitive statement applicable to the immediate posttrauma context.

Early Posurauma fnterve111io11
Empirical evidence regarding the effectiveness of early posttrauma intervention designed to facilitate expression may help shed light on expression's value. Psychological debriefing is the most common type or intervention, and within this category of techniques, critical-incident stress debriefing (CISD; Mitchell, 1983) is the most widespread (McNall y et al., 2003). An essential objective of psychological debriefing in general and CISD in particular is to encourage expression of one's thoughts and foel· in gs about a traumatic event soon after it happens. Indeed, expression is thought to be a necessary component of successful coping. Engaging in it should thus reduce the ri~k of subsequent mental health problems resulting from the trauma, including posuraumatic stress disorder (PTSD; Everly & Mitchell, 1999;Mitchell, 1983; for additional discussion, see McNally et al., 2003).
Despite the frequent application of single-session CISD in clinical practice, reviews of methodologically rigorous studies have failed to support its utility. In a meta-analysis, van Emmerik, Kamphuis. Hulsbosch, and Emmelkamp (2002) found lhal CISD did not signilicanlly improve PTS or other trauma-related symptoms (e.g., general anxiety and depression) and did not differ from no intervention at all. In a narrative review, McNally et al. (2003) similarly concluded that there exists a lack of convincing evidence to support the use of psychological debrieling.

Choosing Not to Express
The evidence reviewed thus far supports general benelits of exprei.sing one's thoughts and feelings about a trauma. llo"ever. these benefits are evident with relatively distant, rather than recent. traum:is and have not supported mitigation of P'fS symptoms. Rime and colleagues (for reviews, see Pennebaker et al., 2001. andRime, Finkenauer, Luminct, Zl.•ch, &Philippot, 1998) have furLher investigated social sharing about a negative event. They have found that most people do share their emotions with others. Although such sharing leads individuals to report perceived benefits, it dO\!s not predict actual emotional recovery from the event (e.g .. Zech & Rime. 2005). Four broad types of research have been conducted that support these conclusions: (a) creating an emotional experience in the laboratory (e.g., watching a disturbing video), (b) asking participants about an event from their past that they find most troubling, (c) daily diary methodology in which participants report on the everyday evenLc; they have just experienced, and (d) contacting participants after they experience an emotional event (e.g., childbirth). These approaches provide important insight but do not directly address expression in the immediate aftermath ol a collective trauma. In addition, when longitudinal data were collected in these studies, it was typically for a matter of weeks with small samples, which may not bl! long enough or powerful enough 10 find possible delayed reactions.
In sum, existing research has left unanswered important questions regarding the re lationship between expressing thoughL<; and feeling:. and coping success. Consistent with Wortman and Silver's ( 1989, 200 I) myths of coping, McNally et al. (2003) raised the key issue thul "pmfossionals working with trauma survivors may have too quickly concluded that the initial disinclination or survivors to disi.:u~s their trauma constitutes a form of dysfu nctional avoidani.:c likely to hinder recovery" (p. 66). Importantly, although experimental manipulations i.:crtainly have their place, they do not address the matter of self-~election in postlrauma expression: Whal does it mea n when people choose to express versus not lo express, and wha t implications does this have for subsequent mental and physii.:al health outcomes?
If expression is beneficial, individuals who choose not to ex-prc~ immediately after a collective trauma should exhibit greater mental and physical health symptoms over time compared with those who do express their feelings. However-consistent with Rim~'s findings (e.g., Pennebaker et al .• 200 I;Rim~ et al., 1998)-if belief in the value of expression is nothing more than an unsupported assumption about coping, choosing not to express may represent a true lack of trauma-related distress rather than pathological denial. In other words, when compared with less dislre!>sed individuals, those who experience more intense distre.c;s after a trauma should be more likely lo express their feelings. This greater distress, in turn, should predict greater long-term symptoms.

Challenges of Trauma Research
Investigating the role of immediate posttrauma expression choices in coping outcomes poses several challenges. Silver ct al. (2006) identified a number of problems typical of trauma research. First, small and nonreprescntative samples (e.g., natural disa:.ter survivors in a particular area) are the norm, potentially limiting gencrali£ability and dinical applicability. Second, da ta coll~ct~on often begins too late, without pretrauma measures of func110111ng or immediate posttrauma responses. This c."Omplicates interpretation of subsequent outcomes and precludes access 10 rich sources of data. Third, studies are often nol longitudinal, and when lhC)' are, they often do not follow participants for longer than 12 months, potentially missing long-tenn effects.

Terrorist Attacks of September 11, 2001
Oy investigating responses to this collective trauma in particular, the present study was able to avoid several limitations. First the national scope of the event made il possible lo draw a large and dive rse national sample. Second. using an existing survey panel (sec Method for details) allowed assessments of prctrauma mental and physical health as well as immediate posttrauma reactions. Third, re:.pondenlS could be followed for 2 years po~llrauma.

Overview
Beginning on September 11, 2001, respondents had the opportunity to provide their react. ions to Lhe terrorist attacks by using an open-ended prompt. With these responses. two questions ~ere tested: (a) Did choosing to express thoughL~ and feeling:. by responding to the prompt versus choosing not to express (i.e., not responding) predict mental and/or physical health outcomes over the follow ing 2 years? (b) Among individuals who expressed, did length of response predict longitudinal outcomes?
It was possible 10 generate competing predictions. If choosing not to express at all or expressing only minimally in the immediate afte rmath or a collective trauma is harmful and renects vulnerability 10 poor adjustment. then individuals who did not respond or submiued shorter rc:.ponscs should exhibit higher symptoms over limt:. relative to those who responded and submitted longer re-:.ponscs. In contrast, if choosing not 10 express is not harmful and renects n:siliencc and a true lack of posltrauma distress, nol responding or submiuing shorter responses should be associated with lowe r symptoms over time.

Datn Collection With a Web-Ennbled Panel
The study sample, provided by Knowledge Networks lni.:. (KN), an online survey research company, was drawn from a national!)• representative Web-enabled panel that was created through traditional probability methods (i.e., using random-digit dialing I RDDj; for detail'>, see Silver cl al., 2002Silver cl al., , 2006. To ensure representation of population segments that would not otherwise have Internet access, KN provides panel houS\!holds with an Internet connection and Web TV to serve as a computer monitor. In exchange, panel members agree to complete 3-4 short surveys a month sent through their password-protected e-mail addresses. Unl ike typical Internet panels, in which people who already have Internet access choose to opt in. no one L 'an volunteer for the KN panel: all participants are selected with RDD. Thus, the KN probabilitybased, Web-enabled panel is demographically comparable with samples that are obtained by RDD survey methodology. Krosnick and Chang (2001) reported an empirical comparison of the KN panel to a traditional ROD sample and found it to be comparable in terms of both demographics and ·'psychographics" (e.g., selfperceptions, civic attitudes, political attitudes, and behavior). The recruitment response rate for the current study was approximately 53%---comparablc with traditional RDD sam ples (Krosnick & Chang, 2001). Once participants have been selected for the panel, responding 10 any given survey is voluntary, and the provision of Internet service is not dependent on completion of any specific survey. Eve n though panel members complete surveys regularly. there arc no significant differences over time in responses given by "seasoned" participants from "'naive" ones (Dennis. 200 I).
De mographic variables are assessed for all individuals whe n they enroll in the KN panel. On entry into the KN panel and prior to September 11, 2001 , respondents also completed a survey of their mental and physical he:ilth history that assessed whether a physician had ever diagnosed them with any of 35 physical and mental health problems, im:luding depression and anxiety disorder.
Respondents were informed about the study and its risks and benefits prior to completing each survey; subsequent completion of the surveys was considered informed consent to participate. The research was co nducted in compliance with the university's Internal Review Board.

Current Sample
On September 11, 2001, KN e-mailed its panelists the following open-ended prompt: " If you would like, please share your thoughts on the shocking events of today." Panel members were allowed to provide written responses to the open-ended prompt until September 21, 2001. Approximately 36,000 KN panel members were available to receive surveys at that 1ime; of these, 19.593 opened the e-mail containing the prompt and 13,958 responded. Independently, in the years following the 9/11 attacks, our research team collected longitudinal data from a nationally representative sample or the adu lt United States population randomly selected from the KN panel (see Silver et al., 2002Silver et al., , 2006: 3, 170 respondents completed subsequent waves of data over the fo llowing 2 years. Within this group, a subsample of 2,138 respondents also read the openended prompt immedia1ely after 9/'11. Of this sample, 1,559 chose to write a response to the open-ended prompt; the remaining 579 saw the prompt but chose not to respond (for more details on the subsample who responded, see Chu, Seery, Ence, Holman, & Silver, 2006).

Measures
Response to ope11-e11ded prompt. Two va riables were created from responses lO the open-ended prompt descri bed above: (a) a dichotomous measure of whether or not participants responded to the prompt and (b) a measure of the length of response (in characters) provided by those who did respond.

2001.
Exposure to the C)fll attacks.
hems modified from prior research on disaster exposure (Holman & Silver. 1998;Koopman, Classen, & Spiegel, 1994) assessed respondents· 9/1 1-relatcd ~x posure. Individuals were categorized into one of three levels of exposure: direct exposure-in the World Trnde Center (WTCJ or Pentagon, seeing or he<1ring the attacks in person, or having a close relationship wi th someone in the targeted buildings or airplanes: live media exposure-watching the attacks on television live a$ t.hey occurred; and 110 live exposure-seeing video replay or learning of the attacks only after they had occurred. United States Postal Service residential zip codes were used to compute distance from the WTC, categorized into groups representing individuals who lived within 25 miles; 25-100 miles; 100-500 miles: 500-1.000 miles; and over 1,000 miles from the WTC.

Other pore111ia/ explallatory covariates.
An index of ph ysician-diagnosed mental health problems with values of 0 (110 diagnoses) or I (depression, a1ixiety, or both) was created from the pre-9/11 hea lth survey. A count of pre-9/11 physician-diagnl>sed physical health ai lments was also created from the health diagnosis checklist and used as a covariate in all analyses.
Acute stress responses were assessed 2 weeks post-attacks by using the Stanford Acute Stress Reaction Questionnai re (Cardena, Koopman, Classen, Waelde, & Spiegel, 2000). Items were revised to a 6.5 grade Kincaid reading level. and respondents reported whether they ··experienced .. or ·'did not experience" 9/ 11 stressrelated symptoms (a = .88). At the same time, respondents completed the Brief COPE (Carver, 1997), a measure of 14 different coping s trategies (e.g., active coping, denial, emotional support seeking, self-blame). Partici pants indicated on a 4-point scale 1he frequency with which they used each strategy to cope with the 9/ 11 terrorist attacks.
Participants also completed a mod ified version of the World Assumptions Scale (Janoff-Bulman, I 989), a measure that assesses beliefs about the benevolence and meaningfulness of the world, in each of the surveys administered between 2 and 24 months post-9/11. This measure had good reliability across all waves (as = .79-.87).
Lifetime exposure to stressful events was assessed by aski ng participants whether they ever experie nced each of 37 negative events (e.g., child abuse, divorce) and the age(s) at which they occurred. This measure was modified from the Diagnostic Inter-view Schedule trauma seclion (Robins, Helzer, Croughan, Williams, & Spi17.er, 1981 ), was expanded lo include a wider variety of stressful events by using primary care patients' reports of lifetime stress (Holman. Silver, & Wai tzkin, 2000), and has provided rates of specific events comparable with those in other community samples (Brc:slau ct al., 1998;Kessler, Sonnega, Bromet, & Nelson, 1995). A continuous variahle was computed representing the total numher of prc-9/11 stressors. Finally, 18 months post-September 11, 2001, participants completed the Ten-Item Personality lnvcntllry (Gosling, Rentfrow. & Swann, 2003), a brief measure of the Big Five personality domains .

Analytic Strategy
A11alyses. Analyses were conducted with generalized estimating equations (GEE), a population-ave raged analysis appropriate for longitudinal s urvey data that accommodates missing. data and provides necessary adjustments of standard errors. The analysis combines assessment points for a given dependent variable, yielding a single significance test fo r each predictor across all assessments. Because two differen t scales were used to assess PTS symptoms (the lES-R at 2 and 6 months pos ttrauma; and lhc PCL at 12, 18, and 24 months posttrauma), the two spans of time corresponding to each scale were analyzed separately. All analyses were conducted with STATA for Macintosh (Version 9.2), specifyi ng the robust option.
Covariates. All analyses controlled for the following (hereafter referred ltl as the standard covariates): demographics (gender, ethnicity, age. income, marital status, and education); degree of exposure to and distance from lhe attacks; and prc-9/11 mental and physical health history. To establish that results were not driven primarily by respondents who were distant fro m or relatively unexposed to th e trauma, interactions between the primary predictors and distance and exposure were also tested. There were l,TI9 respondents with complete data on all covar iates of interest, constituting the sample used for these analyses.
As previous research has demonstrated that acute stress is a risk factor for subsequently developing PTSD (e.g., Brewin, Andrews, & Rose, 2003;Ozer. Best, Lipsey, & Weiss, 2003), separate analyses tested whether the predictive effect-; of responding L o the prompt were independent of acute stress symptoms assessed 2 weeks post-attacks. We also conducted secondary analyses to rule out plausible alternative explanations for our findings (see below).
Tra11sformatio11s. The following variables were highly positively skewed, so inverse or natural logarithmic transformations were performed: length of written response to the open-ended prompt (number of characters). acute stress response, generalized dis tress, PTS sympto ms, and pre-9/ 11 physical hea lth diagnoses.
Coef[iciem reporting. To make coefficients repo rted in tables and the text more interpretable, all continuous predictors (e.g., response length) were s tandardized and all continuous outcome variables except for diagnosed physical disorders (see below) were divided by the standard deviation of all observations across all waves of data collecti on. Coefficients thus reflect effect si1.cs in units of standard deviations. Because GEE util izes maximum li kelihood estimation, traditional measures of effect s ize such as variance accounted for cannot be calculated. For categorical predictors (e.g., choosi ng to respond to the prompt vs. choosing not to respond), B coefficient~ represent the difference between the com-pared groups in standard deviations of the outcome variable. For con tinuous predictors (e.g., length of response to the prompt). f3 coefficients represent the number of standard deviations of change in the outcome variable predicted fo r each standard deviation change in the predictor. For dichotomous outcomes, odds ratios represent the relative likelihood of the outcnmes as a function or category membership (categorical predictor) or, for continuous predictors, each standard deviation change in the predictor (e.g., the degree of increase in the likelihood of choosing lo respond to the prompt vs. choosing not to respond fo r each standard deviation increase in age). Because physical diagnoses are count data, incidence-rate ratios ( IRRs}-analogous to odds ratios-arc reported for that outcome variable. Power analysis. According to Twisk (2003), with three waves of data collection and within-subject correlations of 0.5 between waves, 1, 178 total respondents are required to achieve power of 0.8 when alpha is 0.05 and the expected effect is 0.1 standard deviations in magnitude. This estimate suggests that the current design and sample provide ample power to detect small-tomedium effects.

Sample Characteristics
At the start of the study, the sample ranged in age from 18 to 9·1 years old. with a med ian of 48 years, and was 50.6% women and 49.4% men. Almost 73% of the sample self-identified as White (non-Hispanic). 10.6% as Hispanic, 9.4% as Africa n American (non-Hispanic), and 7.2% as Other, which included Asian. Median household income was $40.000-$49,999. Approximately 6 1% of the sample was married, 15% was divorced or separated, 16.2% was single, and 7.7% was widowed. Just over 9% of the sample attained less than a high school degree, 35.7% held a high schoo l degree, 29.7% attended some college, and 25.3% held a college or adva nced degree. The number of pre-9/1 I physician-diagnosed physical health ai lments ranged from 0 to 26, with a median of 3. Prior to September 11 , 2001, 14.7% of the sample reported that a physician had diagnosed depression, anxi ety, or bolh. Just over 6% of the sample lived within 25 miles of the WTC, 5.2% lived between 25 and 100 miles, 20.4% between 100 and 500 miles, 23.9% between 500 and 1,000 miles. and 44.3% lived over l.000 miles. More than 4% of the sample reported having been directly exposed to the attacks, 63.4% reported having been exposed hy watching the attacks li ve on TV, and 32.4% repmted no live exposure.

Predictors of Reading the Open-E11ded Prumpt
A suhsample of 2, 138 respondents reported longitudinal data in the 2 years post-September 11, 2001, and read the open-ended prompt immediate! y after the al tacks. An additional 1,041 reported longitud inal data but did not read the open-ended prompt because they did not open the survey e-mail within the time limit. Differences between these two groups were assessed with a logistic regression analysis, using the variables included in the standard group of covariates (gender. ethn icity, age, income, marital status. education, degree of exposure lo and distance from the attacks, and pre-9/ 11 mental and physical health history) lo predict rending the open-ended prompt versus not reading it. Results revealed that the following we re significantly (ps < .05) more likely to read the prompt: older respondents (odds ratio (OR] = 1.317; 95% confidence interval (Cl! • 1.187, 1.462). widowed respondents compared with married respondents (OR -1.466; 95% Cl = 1.012, 2.125). respondents.,.. ho lived I 00-500 miles away from the WTC (OR -1.881 ; 95% Cl -1.255, 2.8 19) or 500-1,000 miles away from the WTC (OR = 1.481 ; 95% Cl = 1.004, 2.184), compared with those who lived within 25 miles, and respondents with a higher number of pre-9/ 11 physical health difficulties (OR = 1.122; 95% Cl = 1.0 19, 1.237). The following were less likely lo read the prompt: women (OR = 0.827; 95% Cl = 0.693, 0.988). and higher-income responde nts (OR = 0.868; 95% Cl = 0.786, 0.958.). No other predictors reached significance. Because lifetime trauma history was assessed after September 11, 2001, it was added to the standard covariate model in a separate analysis: it did not significantly predict reading the prompt.

Prediccors of Attrition
In a GEE analysis, using the standard covariates 10 predict failures 10 participate in longitud inal assessments over the combined five waves of posl-9/11 surveys (each wave coded dichotomously as participated (I) vs. did not participate [OJ) revealed that U1e following people were significantly (ps < .05) le51> likely to miss al>scssmcnts (i.e., more likely lo have provided data): older respondents (OR 0.830; 95% Cl ~ 0.782, 0.881). and respondents with a high school degree (OR = 0.734; 95% Cl = 0.611, 0.881), some college (OR ~ 0.815; 95% Cl = 0.673, 0.988), and a college degree or higher (OR -0.762; 95% Cl = 0.625, 0.930). relative to those with less than a high scho<ll degree. No other standard covariates reached significance. When respo nding to the prompt and length of response were added individually lo the standard covariate model, neither variable significantly predicted missing assessments. When lifetime trauma history was added to the standard covariates in a separate analysis, it also failed to predict attrition significantly. A l>ubstantial portion of the eligible adu lt sample continued to participate in follow-up assessme nts (ranging from a 74o/..-9 1% part icipation rate at each wave), an d overall the sample rema ined representative of the United States adult population over time (see Silver et al., 2006).

Predictors of Expression
Responding ro the prompt. Usi ng the standard covariates to predict choosing to respond to the prompt. a logistic regression analysis revealed that older respondents were significantly (ps < .05} more likely lo choose to respond than were younger respondents (OR • 1.226; 95% Cl = 1.081, 1.391). and respondents who reported more pre-9/ I 1 physical health diagnoses were more li kely to respond lO tl1e prompt (OR = 1.196; 95% Cl = 1.061, 1.348).
No other standard covariates reached significance. In a separate analysis in which lifetime trauma history was added to the model containing the standard covariates, people who reported more lifetime traumatic events were more likely to choose to respond (OR = 1.199; 95% Cl = 1.047, 1.374).
Lengtlr of respo11se to prompt. Using the standard covariates lO predict response length, a linear regression analysis revealed that women wrote significantly (ps < .05) longer res1 >0nses than did men (B -0.258), and respondents who reported more pre-9/ I I physical health diagnoses wrote longer responses(~ = 0.069). No other standard covariates reached significance. In a separate analysis in which lifetime trauma history was added to the model containing the standard covariates, people who reported more lifetime traumatic events wrote longer responses(~ = 0.165). The primary longitudinal analyses reported below included all standard covariates. regardless of s ignificance in these preliminary models.

Expressio11 and Adjustment Over Time
Compared with participants who elected not to respond to the open-ended prompt, participants who did respond exhibited worse mental heal th o utcomes (see Table l), and in particul ar those who responded reported higher PTS symptoms from 2 to 6 months and 12 to 24 months post-9/1 1, even after contro lling fo r exposure to and distance from the auacks. Importantly. choosing to respond remained a significant predictor of PTS symptoms 12 to 24 months pust-9/1 l, even after adjusting for 9/11-related acute stress response.
Among participants who responded to the open-ended prompt, longer responses were associated with worse mental and physical health (see Table 2), and in particular those who wrote more reported both higher generalized distress from 6 to 24 months posttrauma and more physician-diagnoi.ed physical ailments 12 to 24 months posurauma. independent of degree of exposure to and distance from the allacks. The effect for physical diagnoses remained significant after adjusting for 9/11-related acute stress response.
These findings were generally not moderated by exposure to or distance from the 9/ 11 attacks, as most interaction terms were not significant. Significant or nearly significant interactions did who lived in close proximity to th e WTC, those who responded to the prompt exhibited poorer mental health over ti me than did t ho~c who chose not to respond (from 2 to 6 months: B = 0.671; 95% Cl = Cl.164, l.177;p < .01; and from 12to24 months: 13 = 0.727; 95% Cl = 0.336, 1.118: p < .001); this difference was also significant amllng more distant respondents, but of smaller magnitude (from 2 to 6 months: 8 = 0.17 1; 95% Cl = 0.044, 0.298: p < .01; and from 12 10 24 months: B = 0.158; 95% Cl = 0.059, 0.256; p < .0 I). Thus, the interactions showed Iha! the effect for responding versus not responding was in fact stronger among participants who hved closest to the WTC, relative to those who lived farther away.

Seco11dary Analyse!>
We conducted several additional analyses to test alternative explanations that might account for the observed relationship between expression and subsequent ou tcomes, including how people who chose 10 express might differ from those who did not do so. These analyses included the standard covariates described previously.

Relationship between initial and s11bseq11e111 expression.
We ~lieve the prompt thal respondcnls were exposed to represents a reasonable proxy for being approached (i.e., by a clinician) in the immediate aftermath of a collective trauma and given an opporlunity lo express. However, a potential criticism that would limit the appl icability of our findings is lhal people who did not respond to the prompt may have si mply expressed elsewhere, perhaps because they preferred a face-to-face interaction. If people who chose not Table 2 to express initially also expre~d less according to different or subsequent measures, it would support the notion that our prompt represents a proxy for expressing in general, across situations. Al 2 weeks posttrauma, respondents reported their use of coping strategies (Brief COPE), including items assessi ng their seeking of emotional support and venting (i.e., expressing negative feelings NOie. Standard covaria1es included gender. cihnicity, age, income. marital status, education, degree of exposure to and distance from the a1tacks, and pre-9/l 1 mental and physical health history. ~s represent coefficients calculated by (a) standardizing the transformed values of response length and (h) standardiz.ing each outcome variable by using rite overall SD of all observations across all waves of data collection. Response length incidence-rare ra1ios (IRRs) were calculated by using standardized vnlues of the predictor. reported both seeking more emotional support (8 = 0.169; 95% Cl = 0.062, 0.276: p < .01: sr = .0053) and venting more (D = 0.188; 95% CJ = 0.079, 0.298; p < .001 ; s1;l. = .0066). This patiern is consistent with these participants being more likely to express posttrauma in ge neral, not just to our prompt. When these coping strategics were included as covariates in the aforementioned longitudinal analyses, resull~ paralleled the pattern described when controlling for acute stress response.
Expression as a reflection of distress. If reluctance to express immediately postttauma reflects resilience and true Jack of distress, then choosing to express and expressing more versus less should be associated with relatively high levels of distress in the short term. Consistent with this logic, regression analyses revealed that responding to the open-ended prompt (B = 0.154; 95% CI = 0.042, 0 .265; p < .01 ; s? = .0044) and writing a longer response (13 = 0.094; 95% Cl = 0.034, 0.154; p < .01; sr 2 = .0085) were associated with significantly higher acute stress response at 2 weeks posttrauma. High levels of initial distress should also predict attempts lo cope with that distress. Accordingly, responding to the prompt was a!'sociated with higher reported use of the following coping strategies at 2 weeks posttrauma (assessed via Brief Pretrauma social 11etworks. Members of the KN panel provided additional information before September 11 , 2001, including their memberships in a number of organizations. Specifically, they reported memberships from a list of 18 categories of social, political, and religious groups. Using a count of total memberships as a proxy for breadth of pretrauma social network allowed us to test if smaller and potentially impoverished networks predicted greater expression. If participants with impoverished networks, or those who were lonely, had few other outlets for expression, it could have motivated greater expression in response to our Web-based prompt and accounted for poorer mental and physical health over time. However, counter to this explanation, pretrauma organization me mbership predicted only response length, such that having more memberships was associated with longer responses (13 = 0.073; 95% Cl = 0.0 15, 0.130; p < .05: sr 2 = .0043). Including organization membership as a covariate in longitudinal analyses did not affect the longitudinal long-term adjustment results.

Lifetime rrcmma.
Exposu re to lifetime trauma and negative events is a risk factor for subsequent vulnerability (e.g., Turner & Lloyd, 2004). If a higher number of lifetime traumas arc assoc iated with greater distress post-9/11 , and greater immediate distress is associated with higher likelihood of expression, lifetime trauma history could account for the observed relationship between expression and differences in mental and physical heal th ov.:: r time. When controlling for lifetime trauma in longitudinal analyses, the PTS symptom and physical healU1 results remained significan1 and only the effect of response length predicting generalized distress from 6 to 24 months posttrauma dropped from significance (p -.34).
World assumptions. Differences in beliefs about the benevolence and meaningfulness of the world may be associated with the likelihood of expre. ssion after a trauma and subsequent health outcomes-especially to the extent that a collective traum a con-nicLS with these beliefs. However, in GEE analyses, respondi ng. to the prompt and length of response did not predict respondent~' beliefs as reported over 2 to 24 months post-9/1 l. Simil arly, controlling for World Assu mptions Scale subscales did not change the pattern of resu lts. .0043), and openness to experience (13 = O.l 00; 95% Cl = 0.034. 0.167;p < .01; sr = .0097). Importan tly, however, conlrnlling for the five di mensions did not affect the longitudinal long-term adjustment results reported above.

Discussion
Our resul ts did not support the common assumption that choosing 11ot to express one' s thoughts and feelings in the immediate aftermath or a collective trauma-or expressi ng them only minimally-is harmful and indicative of vulnerability to future negative consequences. Instead, the opposite pattern emerged. Respondents who elected not to express when given the opportunity to do so exhibited lower PTS symptoms over the following 2 years. compared with respondents who chose to express. In addition, less expression in the form of shorter responses predicted lower generali zed distress and belier physical health over time.
Rather than indicating pathology (see McNally ct al., 2003), reluctance Lo express appeared to reflect resilience (i.e., better long-term adjustment). This resilience persisted over 2 years, showing no signs of the delayed onset of symptoms that would be expected if a lack of immediate post-9/11 expression reflected underlying pathology. This is consistent with the ex planation that individuals who experienced more trauma-related distress were more likely both to choose to express and express more once the;:y did so. Expression as a proxy for initial distress thus predicted subsequent mental and physical health symptoms over time. Moreover, it does not appear that this is simply true among only those distant from or relatively unexposed to the tra uma, as these results were either (a) not moderated by exposure to or distance from the 9/11 attacks or (b) in some cases ac tua ll y strongest among people geographically closest to tJie attacks.
The predictive value of expression choices was further established by analyses Lhat controlled for acute stress response symptoms assessed al 2 weeks posttrauma. Results for both mental and physical health remained significant when accounting for this indicator of early distress, which itself is a risk factor for subsequent symptoms. Secondary analyses provided additional supporting evidence. First, participants who responded to the prompt had larger pre-9/l l social networks and vented more about the attacks to their social networks shortly afterwards, consistent wilh our argument that our methodology represents a proxy of expression across situations. This suggests that the observed findings are applicable beyond the context of responding to a Web-based prompt. Second, choosing to respond to the prompt and longer responses predicted greater use of a variety of coping strategies, consistent with expression renecting greater distress, which should have motivated attempts to cope. Third, further establishing the value of assessing expression in the immediate aftermath of a collective trauma, other individual differences measured both prcand posttTauma-including demographic characteristics, mental and physical health history, soda! network breadth, lifetime trauma history. world views, and Big Five personality domainsfailed to account for the rcsullS we observed.
Offering individuals the opportunity to express therefore presents an immediate measure of vulnerability to trauma that may predict variance in future outcomes not accounted for by other methods. Three additional aspects of our study bolster this conclusion. First, a large and diverse national sample that provided pretrauma assessments of mental and physical health history, as well as 2 years of posttrauma assessments, addressed limitations typical in trauma research (set ! Silver et al., 2006). This suggests that the current findings should be applicable to real-world contexts. Second. the results did not depend on the content of respondents' expression. Chu et al. (2006) analyzed this content and found only a few significant predictors of distress and PTS symptoms. Third, expression was not confounded with negative social feedback. When people express after a trauma, they often do not receive the support they desire and expect from their social network. which can have negative consequences for adjustment over time (Tait & Silver, 1989). Here, respondents expressed in a '·social vacuum,'' without any expectation of personal contact or response. thus creating a conservative test of expression uncontaminated by negative concomitants.

The Role of Expressio11 in Coping Witlr Trauma
The current findings support the ''myths of coping'' described by Worlman andSilver (1989, 2001) and the work of Rime and colleagues (Pennebaker ct al., 200 I;Rime ct al., 1998) in the domain of coping with a collective trauma. The notion that expressing one's thoughts and feelings in the immediate aftermath of a collective trauma is a necessary step in successful long-term adjustment was not supported. Instead, people who choose not to express appear able to cope very effectively.
Reconciling these results with previous research may further elucidate the role of expression in coping. Frauaroli (2006) concluded that experimentally induced expression can be beneficial, but this appears to be limited to expression that occurs long after the trau ma. Similarly, reviews of psychological debr iefing (e.g., CISD) as early posttrauma intervention have not supported the technique's etlicacy (McNally et al., 2003;van Emmerik ct al.. 2002). If the process of coping begins immediately after trauma, assessments made at that time rather than months later should provide an optimal test of the role of .:xpression in that process. However, it seems plausible that different mechanisms are at work when initially warding off effects of trauma than when rcnecting on a distant event. For example, expression that occurs simultaneousl y with a biological stress response (e.g., in the immediate aftermath of an event) may have a more powerful impact on long-term adjustment by solidifying or strengthening the !llrcssrelated physiological responses. However, expression at a time when one is mentally renecting (long after initial physiole>gical changes have subsided) is likely to have very different consequences over time. If true, a:.suming that expression functions identically across situat. ions obscures important differences. Even rigorous research can create a "'myth of coping" if its conclusions are applied too broadly.

Clinical Application
The current findings have implications for posttrauma interventions. The prompt utilized here represenL5 a useful analog to being approached by a clinician. The resulLc; suggest the importance of allowing individuals to choose for themselves whether to express their thoughts and feeli ngs after a collective trauma and, more broadly, to choose to participate in interventions rather than being compelled to do so. If individuals who experience the greatesl distress are more likely to express when given an opportunity, they may also be more willing to seek help actively. Assessing people's willingness to express thoughts and feelings may identify who is at risk for later problems and could thus benefit from effectiw intervention. Such time-efficient screening should be particularly valuable in the context of collective trauma, in which large numbers of people receive ostensibly equi valent exposure, yet only some will go on to experience lasting negative effects. While support for the efficacy of CISD is lacking (McNally ct al., 2003;van Emmerik et al., 2002), other interventions that begin later in the process may be more successful, such as those based on cognitive behavior theory (e.g., Foa & Rothbaum, 1998).

Limitations
While the present design is an improvement over a great deal of prior trauma research, we acknowledge several limitations of the current investigation. First, our pre-September 11, 2001. mental health measure was dichotomous (physician diagnosis vs. no diagnosis) and was thus not optimally sensitive. Among the outcome measures, only the physical health measure was administered both before and after September 11, 2001; ideally, all of the outcome measures would have been. This would provide an even stronger basis for inference. It is important to note, however, that assessing any pre-measure at all is highly unusual in trauma research, given the inherently unpredictable nature of such events, particularly those on the scale of collective traumas.
Second, both measures of physical and mental health asked respondents for self-reports of physician diagnoses. Ideally. thei;e reports would be corroborated by medical records. We note, howevl!r, that we used a health measure that has been benchmarked against the Centers for Disease Control and Prevention·s National Center for He<illh Statistics annual National Health Interview Survey (National Center for Health Statistics, U.S. Department of Health and Human Services, 2000), which itself has been validated against medical records.
Third, the current investigation does not resolve if the respondenrs who chose to express actually could have benefited from that or subsequent expression. For such respondent~. it is possible that having the opportunity to express-perhaps over time. to their social networks-protected them from suffering even worse outcomes. Because our research question focused on respondents' choice of expression and was hence correlational in nature, our data do not speak to what effect a manipulation of expression (e.g., Pennebaker & Beall, 1986) might have had.
Fourth, our findings may not generalize to other social contexts of expression. As described above, we believe the prompt from this study serves as a useful analog to importan t real-world contexts. However, it may be the case that expressing in a socially rewarding environment (e.g .. to a supportive spouse) may engender additional benefits not well captured in our data.

Conclusion
Contrary to common ass umption, this stud y demonstrates that individuals wht) choose not to express their thoughts and feelings in the immed iate aftermath or collective trauma are capable of coping successfully and in fact are more likely to do so tha n individuals who do express. This has important implica tions for understanding the role of expression in the coping process and for early posttrauma intervention. On a broader level, this also highlights the dangers of relying on hunches, common sense, and other ·'myths of coping" when attempting to provide intervention after a collective trauma. Doing so can result in was ted time, money, and effort, as well as misappropriation of resou rces away from those truly at risk and active interference with sorne individuals' natural coping processes. Despite the best of intentions, uninformed efforts to help may do more harm than good.
Finall y, by virtue of their scale alone, collective traumas become the center of media attention. This not only contributes to the collective nature of the event but also has the effect of putting the field of psychology itself in the spotlight. Psychologists are called on to contribute to the public's understanding of the traum a thro ugh the media, dispersi ng the apparent word of science to a wide aud ience. The damage caused by misstatements and faulty conclusions drawn from intuition rather than empirical data can thus multiply beyond an individual client. Such statements may be the only contact many people have with psychology, especially if they experienced the trauma through media coverage. Even brief sound bites could guide the public's expectations for how they and those around them should be responding, including expressing thoughts and feelings. Collective traumas thus represent an important opportunity for psychology not only to make a positive impact on society but also lo make a case for its own relevance in the eyes of the public and funding agencies. This raises the slakes for understanding collective trauma and ensuring that psychologists' conclusions accurately renect the data.