Much scholarly literature deals with the clinical, legal, and ethical issues raised by the involuntary treatment of persons with mental illness who are predicted to be dangerous to others (1,2). A large fraction of this work focuses on the ethical, political, and legal debate about the kind and degree of danger that warrants involuntary treatment (3,4). Another considerable fraction describes and evaluates the procedures employed by mental health professionals to identify persons who meet the current standard of danger to others (5,6). However, relatively little theory or empirical work describes how persons other than mental health professionals assess the risk of persons with mental illness (7). This void strikes us as odd, because the community's perception of risk surely influences whether a person is presented to the mental health system for involuntary treatment, as well as the characteristics of these persons. Moreover, the assumptions of democratic societies imply that these perceptions also affect the political process that sets rules for whom the system ultimately admits for involuntary treatment (8). We attempted to fill part of the void by testing a theory that suggests that the events of September 11, 2001, made the residents of Florida, one of America's most populous and diverse states, more risk averse in their assessment of persons with mental illness.
We offer the general hypothesis that the threshold for what constitutes sufficient danger to warrant involuntary treatment declines in a community after widely reported or witnessed violent behavior. The hypothesis arises, in part, from the theory that humans estimate risk through two systems: an analytic system based on abstract reasoning and an experiential system based on feelings and past experiences. Slovic and colleagues (9) argued that in the latter system individuals respond differently to a stimulus depending on their "affect pool," which contains "all the positive and negative markers associated (consciously or unconsciously) with the images" to which the experiential system responds. Images of persons suffering from ambient hazards, such as criminal attacks and natural or manmade disasters, supposedly cause individuals to overestimate the risk to themselves from similar hazards.
A person who encounters disturbing behavior after witnessing others suffer harm may, according to Slovic and colleagues (9), react with the experiential system and perceive the behavior as more threatening than she or he would otherwise. Even if the incidence of disturbing behavior remains constant, the likelihood that someone will perceive it as threatening increases when an ambient hazard that was widely witnessed has recently inflicted harm in the community.
The literature includes a study by Catalano and colleagues (10) that is consistent with the theory that tolerance for disturbing behavior decreases when ambient hazards are inflicted on the population. The study examined data from San Francisco before and after the terrorist attacks of September 11, 2001, which were widely witnessed by way of television. The study found that more men and women with mental illness were involuntarily admitted to psychiatric emergency services for danger to others after September 11, 2001, than would have been expected from historical patterns. The study controlled for voluntary admissions to the same facilities to rule out the possibility that the increase in involuntary admissions reflected a general increase in disorder.
The study by Catalano and colleagues (10) did not invoke the theory by Slovic and colleagues (9). However, Catalano and colleagues speculated that if rising intolerance in the community, rather than greater police presence, accounted for the increase in involuntary admissions, there should have been increased requests from the community for police to deal with persons who were perceived to be threatening. Data supported the speculation in that calls that requested police help in dealing with threatening persons increased in a pattern congruent with the incidence of involuntary treatment.
The implications of the findings by Catalano and colleagues for the theory of Slovic and colleagues remain unclear for at least two reasons. First, the external validity of the findings cannot be assessed without replication elsewhere. Second, the use of involuntary admissions to psychiatric emergency services rather than the use of evaluations for admissions detracts from the test. Admission follows only after an assessment by one or more mental health professionals has determined that the person meets the technical standard for involuntary treatment. The number of admissions does not necessarily gauge changes in how police officers and the persons who summoned them judge dangerousness.
We contribute to the body of literature by testing the hypothesis that evaluations that were initiated by law enforcement officials in Florida for the involuntary examination of persons with mental illness who were thought to be dangerous to others increased in the weeks after September 11, 2001.
We tested the hypothesis separately for men and women because the literature concerned with environmental antecedents of psychiatric emergencies (11)—including studies concerned with the effects of September 11, 2001, on civil commitments (10)—has reported different responses by gender. These differences converge with the larger body of literature that has reported gender differences in violent behavior among persons with mental illness (12) and in biological and behavioral responses to stressors (13). The literature has also shown that representative samples of the population view men with mental illness as more threatening than women with mental illness (14).
We used data from Florida for several reasons. Florida has a large and diverse population and an excellent system for tracking involuntary examinations. Florida's nearly 17 million residents represent 5.8 percent of the U.S. population. The 1999 median household income of Floridians was $38,819, compared with $41,994 nationwide. Estimates from 1999 indicate that, as in the United States as a whole, about one-eighth of Floridians were living below the poverty line. Most Floridians are white (78.0 percent), 16.8 percent are Hispanic, 14.6 percent are black, and 1.7 percent are Asian (15).
Florida has centrally archived client-level data on involuntary psychiatric examinations since 1997. All facilities that conduct such examinations are required by Florida Statutes (section 394) to submit these data to the state archive (16).
Involuntary examinations may last up to 72 hours and require evidence of mental illness and of harm to others, harm to self, or self-neglect. Examinations take place in one of more than 100 receiving facilities approved and regulated by the Florida Department of Children and Families. Receiving facilities are publicly and privately funded and include psychiatric units of large hospitals, psychiatric hospitals, and inpatient units of community mental health centers. Examinations may be initiated in several ways: by law enforcement officials, by mental health professionals, or by order of judges. Psychiatrists, clinical psychologists, psychiatric nurses, and clinical social workers can conduct the examinations.
Involuntary examinations can lead to several outcomes. Individuals may be released after examination if they do not meet the criteria for mental illness, harm, or self-neglect. Others, if competent to do so, may agree to voluntary inpatient treatment. The remainder are retained for inpatient placement. Inpatient placement orders may be up to six months in length per order.
We used client-level data to create weekly counts of the number of men and women who were involuntarily examined on the basis of evidence of mental illness and harm to others. We created "areal" weeks beginning with Tuesdays, so that the week to which we attributed September 11, 2001, would start with the day of the attacks rather than with September 9, 2001. We began our series with the seven-day period that began with Tuesday, July 6, 1999, and ended it with the seven-day period that ended with Monday, December 31, 2001, to ensure that we could identify seasonality and that we would have sufficient power to detect effects of the size expected on the basis of the San Francisco study. Tuesday, September 11, 2001, began the 115th areal week of the 130-week test period.
The University of South Florida's social and behavioral institutional review board judged our research to be exempt from review, because the counts of weekly evaluations provided to us posed no risk to the privacy or health of any human participant.
We could never randomly assign persons with mental illness to communities in which we manipulate events such as the ones of September 11, 2001. Therefore, we turned to an interrupted time-series design for our study. This design would yield support for our hypothesis if the observed number of examinations initiated by law enforcement personnel on or soon after September 11, 2001, was greater than the number expected under the assumption of no effect. The statistical tests used in such designs typically expect a value within the 95 percent confidence interval of the mean of the values that were observed before September 11, 2001. However, time series often exhibit autocorrelation, including trends, cycles, and the tendency to remain elevated or depressed after high or low values in the series. Autocorrelation complicates hypothesis tests because the expected value of an autocorrelated series after a hypothesized interruption is not the mean of all values before the interruption.
Researchers typically deal with this problem by expressing autocorrelation as an effect of earlier values of the dependent variable (17). The residuals from a time-series equation with the correctly specified earlier values of the dependent variable exhibit no autocorrelation. The analyst can then add other independent variables to the equation to determine whether their coefficients are different from zero in the hypothesized direction.
Removing autocorrelation from the dependent variable before testing the effect of the independent variable yields the added benefit of avoiding spurious associations induced by shared trends and cycles. The estimated coefficients are net of shared autocorrelation.
Epidemiologists have offered an alternative method that measures the dependent variable in a comparison population and uses the series as a control variable in the test equation (18). This method, like that described above, removes patterns in the dependent variable that are induced by forces that are also at work in the comparison population. The approach also controls for unspecified variables that affect both populations but exhibit no patterns.
We combined the two approaches. To do so, we modeled the number of examinations initiated by law enforcement personnel as a function of those examinations initiated by mental health professionals. We inspected the residuals of the model for autocorrelation and assumed that forces unique to examinations that were initiated by law enforcement personnel induced any remaining autocorrelation. Then we removed remaining autocorrelation in the residuals by including, as in the purely historical approach, the appropriate earlier values of the dependent variable among the predictors. Finally, we added to the equation a variable scored 1 for the seven-day period starting with September 11, 2001, and 0 for all other seven-day periods. The coefficients of this variable are net of autocorrelation and of any confounding effects of phenomena that exhibit no autocorrelation but affect examinations that are initiated by law enforcement personnel and mental health professionals.
Our tests began with modeling weekly examinations that were requested by law enforcement officials for danger to others as a function of similar examinations requested by mental health professionals. We estimated separate models for men and women. We then inspected the residuals of these models for autocorrelation. We used the strategy attributed to Dickey and Fuller (19) and Box and colleagues (17) to identify and model autocorrelation. The strategy—autoregressive, integrated, moving average (ARIMA) modeling—draws from a very large family of models that describe autocorrelation in time series.
We then added a binary variable with a score of 1 for the seven-day period beginning September 11, 2001, and a score of 0 for all other areal weeks to equations that resulted from the above steps. We specified the variable with no delay as well as with a delay of 1, 2, and 3 areal weeks to capture any delayed associations and any associations that persisted longer than seven days. We estimated the equations that included the September 11, 2001, variable and then specified final equations by deleting any parameters that did not reach significance at p<.05 (single-tailed test). We also inspected the residuals of the final models to ensure that no autocorrelation appeared after reducing the number of predictors.
Over the 130 weeks of the study, law enforcement officials initiated a mean±SE of 25.96±.52 examinations for men and 13.47±.37 for women per areal week. Mental health professionals initiated an average of 26.45±.61 examinations for men and 13.63±.38 for women per areal week.
t1 shows the parameter estimates for the initial as well as the final models for examinations of persons who were perceived by law enforcement personnel to be mentally ill and potentially harmful to others. Results, for both men and women, support the hypothesis. Approximately 14 (mean of 13.7±3.8) more women than otherwise expected were involuntarily examined on the basis of evidence of harm to others in the areal week that began with September 11, 2001. Law enforcement officials presented more women for examination in the areal week that began with September 11, 2001, than in any other areal week in the entire 130-week study period. Law enforcement officials presented a mean of 10.7±5.1, 10.4±5.1, and 13.7±5.1 more men than expected for involuntarily examination over the three areal weeks that began with September 18, 2001. The 37, 38, and 39 examinations of men conducted over those weeks were the three highest values over the 130-week study period.
We conducted two additional tests to determine whether our findings resulted from analytic artifacts. We applied the routines developed by Chang and colleagues (20) to determine whether outliers other than those associated with the events of September 11, 2001, distorted our estimates. Identifying and controlling outliers did not change the results of the tests.
We also performed an additional test to determine whether the association between the September 11, 2001, attacks and the higher number of men who were presented for psychiatric evaluation persisted into the areal week that began with October 9, 2001. The observed values fell back within the 95 percent confidence interval (single-tailed test) during that week.
Our findings support the hypothesis that in the weeks immediately after September 11, 2001, law enforcement personnel in Florida presented more men and women for psychiatric examinations to avoid harm to others than would have been expected from historical patterns and the number presented by mental health professionals. Our findings cannot be attributed to trends, seasonality, other cycles, or the tendency of the examination time series to remain elevated or depressed after high or low values. We identified and controlled for all these forms of autocorrelation in our data. Nor can the findings be attributed to any "third variable" that affects examinations initiated by mental health professionals. We also ensured that outliers other than those associated with the events of September 11, 2001, did not distort our findings. Therefore, any argument that we made a type I error would have to identify an omitted variable that coincided with, but was not caused by, the events of September 11, 2001, that exhibited no autocorrelation and that did not affect examinations initiated by mental health professionals.
We interpret our results as consistent with the theory of Slovic and colleagues (9). That theory implies that witnessing harm done to others by ambient hazards lowers a population's tolerance for individuals who exhibit threatening behavior and leads to an increase in the number of persons involuntarily examined for mental illness and dangerousness to others.
It could be argued that the events of September 11, 2001, induced our results not by changing the perception of risk posed by persons with mental illness but by increasing the number of persons with mental illness whose behavior threatened others. However, this argument must make additional assumptions in order to explain our finding that mental health professionals in Florida did not present more clients for examination and to explain the findings of other studies that found that voluntary admissions to psychiatric emergency services did not increase on or immediately after September 11, 2001 (10). One assumption that could explain both of these findings is that mental health professionals, but not police, temporarily raised their standard for what constituted a sufficient threat to others to warrant presenting persons for examination. Another is that the persons who became threatening to others after the events of September 11, 2001, came to the attention of police but not mental health professionals. We know of no theory that would suggest either circumstance. Nor can we find empirical evidence for either theory in the literature.
Our findings about the duration of the "September 11th effect" can be compared to those of Gigerenzer (21), who found that Americans drove rather than flew to their destinations for at least three months after September 11, 2001, causing a net increase in transportation-related fatalities. In terms of the theory offered by Slovic and colleagues (9), images of planes crashing into buildings may have had a longer-lasting impact on the experiential system of air travelers than it had on the experiential system of law enforcement personnel responsible for preventing harm by persons with mental disorder in Florida.
Events other than, but close in time to, the attacks of September 11, 2001, could have stimulated the experiential risk assessment system of Floridians and influenced our findings. For example, the media gave much attention to an anthrax poisoning in Florida that was reported to be intentional on October 9, 2001. However, the timing of the effects we discovered detracts from this argument. The significantly elevated number of examinations of women and men fell back to expected levels well before October 9, 2001. Moreover, examinations requested by law enforcement personnel did not increase when the media reported the clearly intentional and unsolved anthrax poisonings in New York, Washington, D.C., and elsewhere that began October 15, 2001, and continued for two weeks.
Studies in states other than Florida that examined the association between the events of September 11, 2001, and assessments for danger would allow an estimation of any "dose response" associated with distance from the point of attack. As Catalano and colleagues (10) have argued, further tests should consider social as well as geographic distance from the attack sites. For example, social distance could be gauged by the number of telephone calls per capita to or from states and the attack sites on the day of the attacks.
Findings from both this and previous studies (10) have implications for theory and practice. The results lend empirical support for the theories offered by Slovic and colleagues (9) concerning risk assessment. Our findings imply that the events of September 11, 2001, caused many Americans, including law enforcement personnel, to have less tolerance than usual for persons with mental illness.
Our work contributes to the body of literature concerned with the stigma of mental illness (14). Considering our findings in light of the stigma literature highlights their implications for the emerging effort to plan psychiatric services in the event of further terrorist attacks (22). We believe that persons officially charged with guiding state and local planning for future terrorist attacks should consider recommendations intended to manage "experiential" risk assessment and its effects on the public health system (23). Those planning responses to future terrorist attacks should anticipate increased demand on organizations that are responsible for civil commitments. Our methods can also be used for sentinel surveillance of emergency psychiatric services; persons who monitor these services may wish to examine possible reasons for unexpectedly low or high levels of use. We also note that police and service providers might be made aware of, and counseled against, reduced tolerance for persons with mental illness during times of communal anxiety. The stigma attached to involuntary examinations and civil commitment may otherwise adversely affect the lives of persons with mental illness for extended intervals.
Dr. Catalano and Mr. Kessell are affiliated with the department of public health at the University of California, Berkeley, 320 Warren Hall, Berkeley, California 94720 (e-mail, firstname.lastname@example.org). Dr. Christy is with the Louis de la Parte Florida Mental Health Institute at the University of South Florida in Tampa. Dr. Monahan is with the School of Law at the University of Virginia in Charlottesville.
Coefficients from equations describing the association of the events of September 11, 2001, with law enforcement-initiated examinations for mental illness and danger to others in Florida for 130 "areal" weeks beginning July 6, 1999