As can be seen in , about 54% of study respondents were female. In addition, in terms of psychological health, indicates that 13% of residents had a history of predisaster depression and 12% had a history of predisaster panic attacks. Furthermore, over 10% met the DSM-IV criteria for a perievent panic attack. The means and standard deviations are also given for the other observed SEM study variables.
Bivariate Pearson’s correlation coefficients among the observed variables in the SEM model are presented in . As can be seen, experiencing a perievent panic attack is associated with all of the PTSD symptom subscales for both Year 1 and Year 2. In particular, perievent panic is associated with Year 1 reexperiencing, avoidance, and arousal symptoms (p < .001), as well as the Year 2 measures of these variables (p < .001). Higher exposure to World Trade Center Disaster events is also associated with these PTSD symptom clusters (p < .001). Finally, having a perievent panic attack is positively associated with greater exposure to World Trade Center Disaster events (p < .001).
SEM Model: Direct Effects
Although these correlations are suggestive, due to confounding, the longer-term direct impact of perievent panic on mental health status cannot be inferred from these data. Therefore, we assessed the direct effects of perievent panic on Year 1 and Year 2 PTSD measured as latent constructs, controlling for other factors. Our SEM contained 4 latent variables (Year 1 and Year 2 PTSD, Year 2 stressors, and Year 2 psychosocial resources) and 16 observed variables. The two PTSD latent variables had three indicator variables: re-experiencing, avoidance, and arousal symptoms. As described, we initially allowed the error terms for each Year 1 symptom group to correlate with its Year 2 counterpart. All of the exogenous variables (i.e., demographic and predisaster mental health measures) were allowed to correlate with each other. We also allowed all of these measures to have direct effects on all of the endogenous variables (e.g., income on exposure, perievent panic, stressor events, psychosocial resources, Year 1 PTSD, and Year 2 PTSD). The model specified direct effects between all Year 1 endogenous and Year 2 endogenous variables (e.g., perievent panic on Year 2 stressor events, Year 2 psychosocial resources, Year 2 PTSD) and contained 4 observed exogenous variables, 12 observed endogenous variables, 4 unobserved endogenous variables, and 16 unobserved endogenous variables, for a total of 36 variables. The model also estimated 9 covariances and 20 variances. With 136 distinct sample moments, 73 parameter estimates, the model had a χ2 = 187.89 (df = 63, p < .001). Although this χ2 indicated a poorly fitting model, other indices suggest an adequate fit, with a root mean square error of approximation (RMSEA) = .034, 90% CI [.029, .040], Bentler-Bonett normed fit index (NFI) = .966, and comparative fit index (CFI) = .977. To improve the model’s parsimony and reduce the possibility that we overcontrolled for the predisaster panic mental health measure, we eliminated non-significant direct pathways for this measure. After these changes, we recalculated all parameter estimates. The new model contained 136 distinct sample moments, 69 parameter estimates, and a χ2 = 189.77 (df = 67, p < .001). Based on the fit statistics, this second specified model fit the data well, with a RMSEA = .033, 90% CI [.028, .039], NFI = .966, and CFI = .977. For this model, we did not add any correlations or make other changes based on the modification indices.
presents a simplified depiction of the final structural model with standardized coefficients, indicating significant direct paths and omitting correlated error terms. (A complete final SEM model is available from the corresponding author.) As can be seen in and , World Trade Center Disaster exposure increases the likelihood of a perievent panic attack, Year 1 PTSD, and Year 2 stressor events (p < .001). Perievent panic is directly related to Year 1 PTSD (β = .22, p < .001), but not to Year 2 PTSD (β = .02, p = .461). Perievent panic also increases Year 2 stressor events and lowers Year 2 psychological resources (p = .025). Year 1 PTSD is positively related to greater Year 2 stressor events (β = .33, p< .001), negatively related to Year 2 psychological resources (β = −.20, p < .001), and positively related to Year 2 PTSD (β = .29, p < .001). As expected, both Year 2 stressor events and Year 2 psychosocial resources are associated with Year 2 PTSD (β = .34 and −.49, respectively, p < .001).
Figure 1 Simplified depiction of final structural equation model for posttraumatic stress disorder (N = 1681). WTCD = World Trade Center disaster; PTSD = posttraumatic stress disorder; Year 1 = baseline survey, Year 2 = follow-up survey. Control variables in the (more ...)
Structural Equation Model Showing β Coefficients for Direct Effects Linking Demographic, Pre-WTCD Mental Health, Exposure, Perievent Panic, Stressor Events, Psychosocial Resources, and PTSD
Further examination of variables in the model (), suggests that income and predisaster depression were associated with greater exposure to the World Trade Center Disaster (p < .001). For perievent panic, income lowered the likelihood of this outcome (p < .001), whereas being female (p = .002), and having predisaster depression (p = .006) and having predisaster panic (p = .004), increased the likelihood of a perievent panic attack. Gender (p = .003), income (p = .008), history of depression (p < .001), and panic predisaster (p < .001) were related to Year 1 PTSD, with income the only predictor to be negatively related to this endogenous variable. None of the demographic or predisaster variables were related to Year 2 stressor events. Being female (p =.007) and having a higher income (p <.001) increased Year 2 psychological resources, whereas predisaster depression decreased these resources (p = .004). Finally, none of the demographics or predisaster mental health measures was associated with Year 2 PTSD, except income (p = .046).
SEM Model: Indirect Effects (Mediation)
Mediation is suggested when an independent variable has an association with a dependent variable and the former also has an association with a mediation variable and, in addition, when the association between the independent and dependent variable is significantly reduced after the mediated variable is included in the model. As suggested, the total effect an independent variable has on a particular dependent variable is the sum of the direct and indirect effects. For brevity, we examine the direct, indirect, and total effects of perievent panic on Year 2 PTSD, as mediated by Year 1 PTSD, Year 2 stressor events, and Year 2 psychological resources.
As noted earlier, our SEM analyses show that perievent panic had no direct effect on Year 2 PTSD (β = .02, ns). The standardized total effect of perievent panic on Year 2 PTSD is β = .22 (p < .001), which means that the indirect or mediated effect of perievent panic on Year 2 PTSD is .20 (.22− .02 = .20). More specifically, individuals who meet criteria for perievent panic have about a .22 standard deviation increase in the probability of having PTSD2 years after the World Trade Center Disaster. However, that increase is almost entirely due to the fact that those individuals who have a perievent panic attack are also more likely to meet criteria for Year 1 PTSD, experience more stressor events between Year 1 and Year 2 postdisaster, and have fewer psychological resources 2 years postdisaster. We examined alternative models, which excluded Year 1 PTSD, Year 2 stressor events, or Year 2 psychological resources, respectively. The direct effect of perievent panic on Year 2 PTSD for these different specifications was .08 (p = .005), .07 (p = .008), and .08 (p = .002). Thus, it appears that perievent panic is about equally mediated by Year 1 PTSD, Year 2 stressor events, and Year 2 psychological resources in its effect on Year 2 PTSD.