On April 19, 1995, the Alfred P. Murrah Federal Building in Oklahoma City was destroyed by deliberate bombing. The bombing was a violent act perpetrated by U.S. citizens on U.S. citizens. This disaster was in many ways unique. At the time, it was the largest disaster of the century in this country. One hundred sixty-seven people were killed, including 19 small children. Hundreds more were injured in the building and in the surrounding neighborhood. More than 100 children lost a parent, and many more lost other family members. The bombing exposed our vulnerability to terrorism and aroused intense fear of domestic threat.
There is a growing literature addressing the effects of disaster on youths, which has been summarized in review articles (1–6) and books (7, 8). Reports have documented posttraumatic stress symptoms associated with exposure to community violence (9–11) and to war and warlike conditions (12–18). Studies generally find a positive correlation between the severity of trauma exposure and posttraumatic stress disorder (PTSD) symptoms (1).
Concern about level of exposure to the bombing and potential clinical problems in Oklahoma City youth prompted the authors to perform a clinical needs assessment of middle and high school students. The assessment involved a survey of more than 3,000 middle and high school students in the Oklahoma City Public School District before the schools closed for the summer recess, approximately 7 weeks after the incident. This report examines exposure, initial responses, and factors associated with posttraumatic stress symptoms in these youths.
The Oklahoma City Public School District is the second largest school district in Oklahoma. There are 10 middle schools, one alternative middle school, nine high schools, and one alternative high school in the district. The middle school (grades 6 through 8) and high school (grades 9 through 12) enrollment is approximately 15,200 students. Participants were volunteers recruited from classrooms of teachers agreeing to distribute the survey instrument. All students willing to participate were included in this convenient study group.
Clinical Needs Assessment Instrument
The clinical needs assessment instrument was designed specifically for the 1995 Oklahoma City bombing to include measures of exposure (physical, emotional, by means of television), initial response to the explosion, and posttraumatic stress and other symptoms present at the time of administration. The measure of initial response included 12 items assessing emotional and physiological reactions adapted from those developed by the University of California at Los Angeles Trauma Psychiatry Service for use after the 1994 Northridge earthquake (unpublished 1994 study by Wang et al.) and by Freedy and colleagues (19). Participants were asked to indicate their level of agreement with statements on a 5-point Likert scale: "strongly agree," "agree," "neither," "disagree," and "strongly disagree."
The measure of posttraumatic stress was adapted from the Impact of Event Scale—Revised (20). The Impact of Event Scale—Revised consists of 22 items representing three PTSD symptom clusters. It includes items measuring intrusion and avoidance from the Impact of Event Scale (21) and items assessing arousal added by Weiss and Marmar (20). The basic psychometric properties of the Impact of Event Scale—Revised have been established (20). Subjects were asked to rate the frequency occurrence of the 22 symptoms in "the past seven days" on a scale with four categories of response: "not at all," "rarely," "sometimes," and "often." Their posttraumatic stress symptom scores were a summation of all items on this scale.
The Oklahoma City Public School District administration arranged for the distribution of the clinical needs assessment instrument. Although we intended to have wide distribution of the instrument, it was clear that we would not achieve universal response because of the time constraints associated with academic activities and student absences. While the district’s central administration was supportive of the endeavor, the decision about participation ultimately rested with individual principals and teachers. The assessment was distributed to all middle and high schools 2 days before the summer recess, 7 weeks after the bombing, and was completed in classroom group settings. Consent was obtained in accordance with the Oklahoma City Public School District procedure for parental notification, which involved passive consent because of the clinical nature of the study, which was designed to provide information about the emotional status of youths who might need outreach services during the summer. The University of Oklahoma Health Sciences Center’s institutional review board approved the study.
The participants were described according to grade level. The middle school group consisted of students in the sixth through eighth grades. The high school group was composed of students in the ninth through 12th grades. Descriptive statistics were derived for each of the grade level groups and also across these groups for an overview of gender and racial/ethnic effects.
Although the posttraumatic stress symptom score was used as the primary dependent variable, several other dependent variables, identified because of their presumed clinical relevance, were also considered. These variables included exposure through physical proximity, knowing someone injured or killed in the blast, exposure by means of television viewing, initial emotional responses and physiological arousal, and other symptoms such as not feeling safe and worrying about self or family at 7 weeks. Descriptive statistics regarding these variables are presented here.
In addition to describing the total study group, we were interested in examining students with exposure and contrasting students with high and low exposure. To conduct these analyses, a comparison group was identified. The comparison group was defined as those students with little or no exposure, by using the following criteria: 1) did not hear the bomb as "very loud," 2) reported no buildings or cars near them damaged, 3) viewed little or no bomb-related television coverage, 4) experienced no injury to relatives or friends, 5) knew no one killed, and 6) attended no funerals. The comparison group included 122 youths meeting these criteria; the remainder constituted the exposed group.
The degree of relationship to a person injured or killed in the blast and the amount of bomb-related television viewing were analyzed by placing students meeting these criteria into subgroups within the exposed study group. For these analyses, one-way analyses of variance (ANOVAs) with six levels for the emotional exposure variables and five levels for the television exposure variable were used to identify overall significance. Differences among levels of the subgroups were examined with Duncan’s multiple range test.
For analysis of the exposed and comparison groups, a two-by-two (exposure versus nonexposure groups by gender) ANOVA was used to examine the main effects and the interaction of exposure and gender on posttraumatic stress symptom scores. Because of limitations on individual cell sizes, further factorial designs (i.e., exposed group by gender by race/ethnicity) were not explored.
An alternative analysis treated exposure as a continuous variable rather than a categorical variable. In this analysis, the scores for each of the exposure items were summed, correcting for direction (the greater the exposure, the larger rating it received) and subjected to regression analysis.
Regression analyses were conducted to clarify the relative importance of gender, race/ethnicity, grade level, and amount of bomb-related television viewing. The preliminary nature of these questions and the desire to develop an efficient model led us to conduct stepwise regressions. At the beginning of each step of the stepwise regression, the percent of variance accounted for by each independent variable (R2) was examined to guard against the influence of trivial differences on variable selection. Finally, a magnitude of effects analysis was conducted.
Characteristics of the Study Group
Demographic findings. A total of 19 schools distributed the assessment instruments, which were completed by 3,218 students. T1 identifies the distribution of student participants by gender, race/ethnicity, and grade level. Girls (N=1,760, 56.5%) outnumbered boys (N=1,354, 43.5%), while the actual enrollment of boys and girls in the school district was more even. The racial/ethnic distribution of the study group roughly paralleled school-wide distribution, with 43.7% self-reporting as African American, 34.9% as Anglo, 11.6% as Hispanic, 4.4% as American Indian, and 5.4% as other. While the racial/ethnic distribution for the specific schools in which participants were enrolled is unavailable, the distribution for the total Oklahoma City Public School District student body (elementary, middle, and high schools) during the 1994–1995 school year was 40.7% African American, 38.5% Anglo, 12.7% Hispanic, 5.2% American Indian, and 2.9% Asian American. Youths were categorized not by age but by grade level as either middle or high school students. A preponderance of the study group was at the middle school level (N=2,360, 75.8%). Discrepancies in study group sizes in various categories reflect missing data.
Physical proximity and exposure. The majority (N=2,729, 84.9%) of the total study group reported being in school on the day of the bombing. Over 60% of the students reported hearing (N=1,979, 61.5%) and/or feeling (N=1,988, 61.8%) the blast. For 746 students (23.2%), it was "very loud," and for 758 students (23.6%), it felt "very strong." Television exposure was also extensive; over two-thirds (N=2,137, 66.6%) of the participants reported that "most" or "all" of their television viewing was bomb related ("most": N=1,368, 42.6%; "all": N=769, 24.0%).
Personal loss and consequences. A surprisingly large number of youths experienced exposure by knowing someone injured or killed in the blast. T2, combining middle and high school students, identifies reported exposure to injury or death by relationship. Three hundred thirty students reported attending at least one bomb-related funeral; of these, 107 attended more than one funeral.
Fear and arousal. The results indicate strong reactions at the time of the blast for students throughout the district. A sizable number of students reported perceived life threat (N=431, 13.5%) and physiological arousal at the time of the bombing (T3). About two-thirds (N=2,135, 67.0%) reported worrying for the safety of a family member, over one-half (N=1,861, 58.4%) reported worrying for the safety of a friend, and over 40% (N=1,285, 40.6%) reported feeling helpless. At the time of the assessment, 7 weeks after the bombing, 464 (14.6%) of the students reported that they still did "not feel safe at all," and 1,084 (34.1%) reported worrying about themselves or their families.
Characteristics of the Exposed and Nonexposed Groups
The mean posttraumatic stress symptom scores were examined with respect to the amount of bomb-related television viewing in the exposed group (F=49.25, df=4, 2991, p=0.0001). The highest mean posttraumatic stress symptom score (mean=46.2, SD=13.6) was in the group reporting that "all" of their television viewing was bomb related. This group’s mean score was significantly higher than those of all other groups (Duncan’s multiple range test, p<0.05) (T4).
The mean posttraumatic stress symptom score in the exposed group was also examined in relation to personal loss and its consequences. With respect to reporting knowing someone injured (F=26.14, df=5, 2995, p=0.0001), mean posttraumatic stress symptom scores were highest for those reporting sibling injured (mean=51.0, SD=7.9), followed by those reporting parent injured (mean=47.4, SD=11.5). Those reporting a sibling injured had significantly higher mean posttraumatic stress symptom scores than all other groups (Duncan’s multiple range test, p<0.05) (T4).
The posttraumatic stress symptom analysis was examined by reported relationship to those killed in the blast for those in the exposed group (F=27.86, df=5, 2990, p=0.0001). Those reporting a sibling killed had the highest mean posttraumatic stress symptom scores (mean=49.0, SD=10.0), followed by those reporting a parent killed (mean=48.8, SD=11.4).
The two-by-two factorial (exposure group and gender) analysis revealed no significant interactions. The analysis did show a significant main effect of exposure and gender (T5). The exposed group had significantly higher scores than the nonexposed group. The girls had significantly higher scores than the boys.
When the total study group was used, stepwise regression analysis was performed with the posttraumatic stress symptom as the criterion variable and gender, race/ethnicity, grade level, and exposure as predictors. Before step one, the primary predictor of posttraumatic stress symptom was exposure. This variable accounted for 11% of the variance, with the other three variables each accounting for less than 3% of the variance. Therefore, step one included the exposure variable in the model. In exploring the remaining variance, we found that gender accounted for an additional 12% of the variance. The other two variables—grade level and race/ethnicity—accounted for only slightly less (11% each) before step two, the inclusion of gender in the model. Grade level, accounting for 13% of the remaining variance, entered the model at step three, unaffected by the removal of the variance accounted for by gender. The dummy coded race/ethnicity variable, while accounting for only slightly less variance than gender in step two and grade level in step three, did not enter the equation at any step. The overall model met statistical significance (F=143.56, df=3, 2891, p=0.0001) and accounted for 13% of the variance in the posttraumatic stress symptom scores. In the overall model, exposure accounted for the majority of the variance (11% of 13%). It should be noted that bomb-related television viewing contributed to exposure and therefore was not entered as a separate item.
In the final analyses, a stepwise linear regression analysis was used to build a predictive model for posttraumatic stress symptom scores from gender, race/ethnicity, grade level, and amount of bomb-related television viewing. The first variable that entered the model was bomb-related television viewing, accounting for 7% of the variance (R2=0.07). Other variables accounted for far less of the variance before the inclusion of the television exposure variable (less than 3%). After the inclusion of television exposure, gender was the most strongly related variable (R2=0.08). Although the other two variables accounted for only slightly less of the remaining variance, gender was chosen by the procedure. Finally, the last variable to enter the model at a statistically significant level was the grade level variable (R2=0.08). Inclusion of gender in step two had little effect on the amount of variance accounted for by grade level. As in the previous analyses, the race/ethnicity variable never entered the model. The final model, including the three contributing variables, was significant (F=88.45, df=3, 2942, p=0.0001), accounting for 8% of the total variance. When the entire model was reviewed, television exposure accounted for most of the variance (R2=0.07). The addition of the other two variables improved the predictive model less than 2%.
Most of the youths in this study were at school on the morning of the bombing. It is no surprise that their physical exposure was great given the magnitude of the blast and knowing that the bomb site is within the boundaries of the school district. In addition to physical exposure, we also addressed levels of emotional exposure and found wide-reaching effects of this disaster on the Oklahoma City community. More than 40% of the youths surveyed reported knowing someone injured, and more than one-third reported knowing someone killed. These data are consistent with a general population telephone survey of adults completed in the Oklahoma City community in the summer following the bombing (22). Exposure through television viewing was also predicted to be extensive, since virtually nothing except bomb-related coverage aired on the major local stations in Oklahoma City for days following the bombing.
Posttraumatic Stress Symptom Analysis
Three demographic variables were examined in the exposed study group: gender, race/ethnicity, and grade level. Gender produced significant differences, with girls reporting higher levels of symptoms than boys. These results are consistent with studies of large study groups, which generally find greater emotional distress in girls than boys following disasters (23–26). Of note, however, some studies fail to find gender differences, and some find qualitative gender differences (15, 27, 28). Few studies have reported racial/ethnic differences, and there is little consistency among the studies that do (23, 26, 28). The grade level variable, although it met entry criteria, accounted for little of the variance and therefore cannot be regarded as a significant predictor. Given the predominance of middle school students, this finding is not surprising.
Predictably, youths who reported that their parents or siblings were injured or killed had higher mean posttraumatic stress symptom scores than others. In the exposed group, the high mean posttraumatic stress symptom scores of those who reported a sibling injured or killed relative to those who reported a parent injured or killed are of interest. Although one might expect injury or death of a parent to have greater impact than that of a sibling, relatively little is known about reactions to sibling adversity (29, 30). The injury or death of an age mate may create a sense of personal vulnerability that is at least as great as the loss of protection youths must feel when a parent is injured or killed. This interpretation might also explain the high mean posttraumatic stress symptom scores in youths who reported injury to or death of a friend. Furthermore, when a sibling is injured or killed, grief may prevent parents from fulfilling their usual nurturing role with the surviving children. Finally, surviving children may model their grief behavior after the behavior of their parents.
One of the particularly interesting findings in the current study was the influence of television exposure on posttraumatic stress symptom scores. When exposure status, gender, race/ethnicity, grade level, and television exposure were subjected to a stepwise regression analysis, the primary predictor was television exposure. These data suggest that when close personal consequences may be relatively low (i.e., relatively few experienced the injury or death of a parent or sibling), the media appear to play a role in sustaining posttraumatic stress symptoms. The role of this variable is being more thoroughly examined in an article currently in preparation.
Like most disaster response studies, this one was difficult to conduct because the bombing was unpredicted and the investigators had not prepared to initiate the study. In addition, the bombing created a high level of chaos and grief in the community, which meant that professionals were busy attending to pressing clinical needs.
Access to the school-based population was facilitated by the U.S. Department of Education, which sent a team of experts to assist in the response of the school district, but that meeting occurred only weeks before the close of school for summer recess. Therefore, the assessment instrument had to be prepared with short notice, and it was administered very late in the academic year, at a time when school personnel and students were focused on other issues.
While a large number of students were examined, it is unclear how representative these students were of the entire middle and high school student body. The Oklahoma City Public School District student body for these grades had more evenly split gender differences than this study group; the racial/ethnic distribution of the study group was similar to the school-wide distribution, and the study group’s grade level was predominately of middle school age. This was a convenience study group, in which teachers agreed for their classrooms to participate, and then students within a classroom agreed to participate. In addition, the data were gathered in a group setting, which may have biased responses, and the self-report method is subject to intentional and unintentional distortions by the participants.
It should be recognized that the large number of students in the study increased the power and probability of obtaining statistical significance. A magnitude of effects analysis revealed an effect size of 0.04. This small effect size could reflect the fact that the observed differences had little, if any, clinical significance. On the other hand, several factors may have influenced the results and reduced the opportunity to observe larger effects. For example, measurement error in both predictor and criterion variables produced by a number of factors (including the method of administration) may have increased variability in such a way as to conceal group effects. We are attempting to address this issue with laboratory assessments of subgroups of exposed children. It is important to recognize, however, that small effect sizes do not necessarily detract from the clinical significance of these findings. That is, the clinical impact and scientific relevance of the data may be considerable because of the uniqueness of the event, despite the small effect size.
This study documents the intensity of community exposure, immediate fear and arousal, and lingering symptoms of stress in youths following the bombing. Results from surveys such as this are useful in a variety of ways, both for the immediate situation and for future incidents. These Oklahoma City data were used to assist in the planning of service delivery and to identify students with the greatest clinical need. The data have been used for training educators and counselors who work in the school district and to support funding requests. Furthermore, these data will provide comparison information to be used by other communities when exposed to similar future incidents.
Presented in part at the 149th annual meeting of the American Psychiatric Association, New York, May 4–9, 1996. Received April 20, 1997; revisions received Feb. 23 and Nov. 20, 1998; accepted Jan. 4, 1999. From the Department of Psychiatry and Behavioral Sciences, University of Oklahoma Health Sciences Center. Address reprint requests to Dr. Pfefferbaum, Department of Psychiatry and Behaviorial Sciences, University of Oklahoma Health Sciences Center, P.O. Box 26901, WP-3470, Oklahoma City, OK 73190-3048; firstname.lastname@example.org (e-mail)