Defining Disaster

There is no single, agreed-upon definition of disaster either within or across disciplines. Definitions used in practice and research vary widely, reflecting different objectives and interests in regard to the causes, consequences, and processes involved in disasters. The following discussions touch on the broad spectrum of processes involved in disasters, including, but not limited to, the impact on the healthcare system; the short- and long-term effects on people's health and livelihood; and the behaviors of individuals, groups, and organizations in relation to disasters.

Accordingly, a disaster is “any community emergency that seriously affects people's lives and property and exceeds the capacity of the community to respond effectively to the emergency.”Reference Noji¹ As an extreme example, the 2011 Tōhoku Earthquake and Tsunami (M9.0) that occurred on March 11, 2011, in the Pacific Ocean off the coast of Japan's northeastern region was the biggest earthquake ever recorded in Japan. It caused powerful tsunami waves 10 to 40 meters high, which reached up to 6 km inland, devastating the coastal areas and leaving over 18,000 people dead or missing. The disaster was further compounded by the loss of power and subsequent meltdown of reactors at a nuclear power station affected by the earthquake and tsunami. Large quantities of radioactive contaminants were released, which led to evacuations of surrounding areas. More than 2 years later, many of the victims of this disaster who lost their homes, neighborhoods, and livelihoods still lived in temporary housing settlements, depending on their savings, disaster compensation, and donations. National and local governments are struggling over reconstruction and redevelopment. This event clearly overwhelmed the response and recovery capacity of the community at the individual, household, and organizational levels. Studies of this disaster legitimately go beyond its impact on people's health and the healthcare system.

The term disaster is often used interchangeably with the terms “emergency” and “hazard,” although there are formal distinctions. An emergency is a threatening situation that requires immediate action but may not necessarily result in loss or destruction. If an emergency is managed successfully, a disaster may be averted. A hazard is a possible source of danger that upon interacting with human settlements may create an emergency situation and may lead to a disaster. For the purposes of this chapter, all three terms will be used, and the distinctions in meaning will be maintained.

Historical Overview of Disaster Research

Historically, sociological disaster research has been dominated by exploratory research designs, whereas epidemiological research emphasizes the importance of explanatory designs.Reference Cisin, Clark, Baker and Chapman^2–Reference Guetzkow, Baker and Chapman⁸ Exploratory studies usually focus on examining new areas of research or the feasibility of conducting more structured research with an emphasis on developing hypotheses. Descriptive and explanatory studies, in contrast, start with hypotheses and emphasize minimizing bias and maximizing external validity, with explanatory studies also attempting to infer causality. The next section of this chapter (Current State of the Art) provides greater detail on study design.

The perceived need to enter the field immediately after a disaster encouraged disaster researchers to utilize exploratory study designs rather than more structured descriptive designs. Researchers thought they were dealing with perishable data that had a limited time frame for collection. Information was thought to be unavoidably fleeting, vanishing quickly after a disaster because of memory decay, removal of debris, and other activities. Furthermore, it was assumed that disaster-associated in- and out-migrations were rapidly changing the target population and their communities in ways that could not be captured by the research. Consequently, early research on disasters relied on data obtained through semi-structured interviews with selected informants after quick entry into a community immediately post-impact. Over time, this perceived need to enter the disaster area immediately has been referred to as the “window of opportunity” and has been adopted by practitioners and policymakers as well as other research disciplines including engineering, seismology, medicine, and public health.

Disaster researchers trained in the social sciences have been concerned with the applicability of social theory to the study of disasters and, in reverse, the contributions that disaster research can make to the development of theory. References to theory in the early disaster epidemiology literature are oblique, with the exception of concerns about biological plausibility. Contemporary social epidemiological research more frequently incorporates theory, a subject that is discussed more fully later in this chapter, under the heading of Disaster Vulnerability.

Epidemiology Publications, January 2007–April 2013

In the first edition of this chapter, the authors conducted systematic, although not exhaustive, searches for disaster-related research articles in the epidemiological literature published between 1987 and 2007. The review that follows covers the period of January 2007 through April 2013. We examine articles published in the Morbidity and Mortality Weekly Report, Prehospital and Disaster Medicine, and four epidemiologic journals (American Journal of Epidemiology, Annals of Epidemiology, Epidemiology, and Epidemiologic Reviews). We identify the location of the disaster, the research team, and the extent to which bibliographies include references to the broad social science literature, in addition to medical and epidemiologic journals. As a means of comparison, we provide a similar review of articles on disasters in two social science journals known for publishing disaster research (Environment and Behavior and International Journal of Mass Emergencies and Disasters) and determine the extent of cross-reference to the medical and epidemiologic literature (see Table 1.2).

A total of seventy-seven articles were identified with the following distribution: twenty-nine in the Morbidity and Mortality Weekly Report (MMWR), twenty in the American Journal of Epidemiology, fourteen in the Annals of Epidemiology, twelve in Epidemiology, and two in Epidemiologic Reviews. Although our review focuses on journals published in English, thirty-four of the articles report on disasters that occurred outside the United States, and most of those articles were written by non-U.S. researchers. These articles examine the full range of disasters and disaster-associated morbidity, mortality, service delivery, and needs assessments. Topics of study included the 2009 H1N1 pandemic influenza (n = 23, of which 19 were in MMWR); other influenza outbreaks including historical events (n = 4); combat and war in both contemporary and historical settings (n = 15); weather events involving extremes of heat and cold (n = 5); the terrorist attacks of September 11, 2001 (n = 3); wildfires (n = 3); floods (n = 3); earthquakes (n = 2); preparedness for disasters (n = 2); a pub fire; a dioxin spill; dust storms; a power outage in the northeast United States; a typhoon; a hurricane; a coal mine disaster; a tornado; school mass homicides; two review articles on global surveillance and humanitarian relief workers; and two historical vignettes on the Halifax explosion and the Johnstown flood.

In contrast to the earlier 20-year period when the American Journal of Epidemiology published an average of one disaster article each year, an average of more than three articles was published each year between 2007 and 2013. Most of the studies are atheoretical (not designed to test a hypothesis or theory), and many combine existent cohort studies with a natural experiment. As before, the emphasis has been on mortality, morbidity, injuries, and psychological distress. Like the early field research conducted by social scientists and psychologists, many studies lack denominator data or information about the population they represent.

The journal Epidemiology, sponsored by the International Society for Environmental Epidemiology, publishes mostly conference abstracts, but also published twelve disaster-related articles between 2007 and 2013. Of these, five were conducted by U.S.-based researchers, five by research groups outside the U.S., and two by groups comprised of both U.S. and non-U.S. researchers. There were a total of eleven references to social science research. During this same time period, the Annals of Epidemiology published fourteen articles on disasters. Data were collected using surveys, registries, and other existent secondary sources of information. Eleven references were made to social science literature in the fourteen studies.

Prior to 2007, in addition to the literature noted previously, Epidemiologic Reviews published review articles on psychiatric distress from disasters, pandemic influenza, toxic oil syndrome, and heat-related mortality. Since 2007, two review articles, one on global public health surveillance and the other on trauma-related mental illness, have at least tangential relevance to the study of disaster epidemiology.Reference Castillo-Salgado^29, Reference Connorton, Perry and Hemenway³⁰

Since January 2007, the CDC periodical MMWR has published twenty-nine articles about disasters throughout the world, with fifteen published in 2009. Most articles combined surveillance with a case series, but articles on school-associated homicides and coal mining included historical reviews of similar events with contemporary surveillance reports. A case-control study in Sudan evaluated an intervention designed to reduce the spread of cholera, and a population-based needs assessment following Hurricane Ike reported on injuries and other health-related needs. Two studies of household preparedness for emergencies and disasters were surveys. There are no references to social science research in any of the twenty-nine articles in MMWR. The lack of such references is particularly surprising in the two articles about household preparedness, given that household preparedness and evacuation behavior have been the focus of a substantial amount of social science research dating back to 1950.

Social Science Publications, January 2007–April 2013

For a brief comparison, we also examined articles on disasters that were published in two social science journals to see whether authors cited research conducted in medicine or epidemiology. Because traditional disaster research primarily originated in sociology, the number of articles published in traditional sociology journals between 2007 and 2013 was examined first. Over that period, only a total of five articles were published in the American Sociological Review, the American Journal of Sociology, and Social Problems, and a single special issue with seven articles was published in Social Forces. The absence of articles in these journals emphasize the extent to which the traditional disaster research paradigm in sociology has become restrictive and unproductive as suggested both by Tierney and in the special issue of Social Forces.Reference Tierney^34, Reference Brunsma and Picou³⁵ Tierney notes that:

As a consequence, disaster researchers have become isolated from mainstream sociology and tend to publish in extreme event and multidisciplinary journals such as Environment and Behavior, the International Journal of Mass Emergencies and Disasters, Disasters, the Natural Hazards Review, the Journal of Contingencies and Crisis Management, Risk Analysis, and Natural Hazards. Environment and Behavior and the International Journal of Mass Emergencies and Disasters were selected as two examples of this broader social science literature.

Environment and Behavior published seventeen articles on disasters between January 2007 and April 2013. Nine articles focused on disasters in the United States and eight on non-U.S. disasters. All articles cited at least one theoretical context for the research being conducted, and data were collected through environmental observation, self-administered and internet questionnaires, in-person interviews, telephone surveys, panel studies, and the combination of multiple sources of data. Across the seventeen articles, there were twenty-one references to medical or epidemiology journals.

The International Journal of Mass Emergencies and Disasters (IJMED) – established in 1983 by the International Sociological Association's Research Committee on Disasters – focuses on theory, research, planning, and policy related to the social and behavioral aspects of disasters or mass emergencies. Papers concerned with medical, biological, physical engineering, or other technical matters are accepted if social and behavioral features of disasters are also discussed. Between January 2007 and April 2013, seventy-three articles were published in IJMED with forty-five focused on disasters in the United States and twenty-eight on disasters outside the United States. Articles in IJMED were less likely than those in Environment and Behavior to provide a theoretical context for the study, but all cite previous relevant research. Two special issues were published during this period on Gender and Disasters (August 2010), and the Theory of Disaster Recovery (August 2012). Across the 73 articles there were 119 references to medical journals and 11 to epidemiology journals. Interestingly, the articles focused on the identification of bodies after disasters had only a few references to relevant medical and epidemiologic literature.

Summary

The previous review demonstrates that roughly half of the disaster-related articles published between 2007 and early 2013 in some of the key English-language epidemiology journals, and a few social science journals, are about disasters occurring outside the United States. Articles published in epidemiology and emergency medical journals rarely cite a theoretical context for their analyses or provide cross-citations to the social science research on disasters. In contrast, articles published in the social science journals reviewed here are often placed within a theoretical structure but with limited references to relevant literature in medicine and epidemiology. These findings suggest that the many disciplines engaged in hazard and disaster research remain largely self-contained, with restricted knowledge of research conducted in other areas and disciplines, constraining the diversity of perspectives that could be brought to bear on critical issues.

Disaster Research Methods

There are multiple scientific perspectives involved in disaster research, and the methods used to study disasters are equally varied. The appropriateness of one methodological approach over another is determined by the specific question the researcher is trying to answer and the discipline in which the researcher was trained. A number of books provide expert guidance on disaster research methods.Reference Norris, Galea, Friedman and Watson^36–Reference Stallings³⁸

Disaster Research Objectives

The objective of disaster research can be exploratory, descriptive, or explanatory. Exploratory studies are the least structured type of research endeavor, often examining new areas of research or the feasibility of conducting more structured research. The emphasis is on developing hypotheses, frequently involving in-depth data collection from a relatively small group of purposively selected research subjects. It should not be assumed that exploratory studies are easier to conduct or less time consuming simply because they tend to be performed on a smaller scale or without the use of large sets of quantitative data.

Descriptive studies, in contrast, start with formal hypotheses or research questions and seek to accurately describe a situation by deriving estimates of important outcome distributions (e.g., disease occurrence by person, place, and time) or associations between variables and theoretical constructs in a population. Like descriptive studies, explanatory studies are driven by hypotheses. The aim, however, is to explain causal relationships. Explanatory research is also referred to as analytic research in epidemiology.Reference Kelsey, Whittemore, Evans and Thompson³⁹ In both descriptive and explanatory studies, emphasis is placed on selecting samples that are representative of the population being studied and minimizing bias in data collection.

Disaster Research Settings

The study of disasters can occur in many different physical and temporal contexts. Among disaster health researchers and epidemiologists, data collection activities have been focused largely in high yield areas where disaster victims are likely to congregate, such as emergency departments. Research conducted in these settings captures the numerator, that is, the number of people with different health afflictions who present themselves in these settings. This approach provides no information on the larger community from which these individuals emerged (i.e., the denominator) or the extent to which they represent the range and severity of disaster-related morbidity in a population. It can even lead to misattribution of the cause for morbidity in the absence of a rigorous protocol. As a case in point, Peek-Asa and colleagues examined coroner and hospital records following the 1994 Northridge earthquake in California.Reference Peek-Asa, Kraus, Bourque, Vimalachandra, Yu and Abrams⁴⁰ They found that, when compared with their systematic, individual medical record review, initial reports overestimated earthquake-related deaths and hospital admissions by misattributing deaths and injuries that presented for care shortly after the earthquake.

Population-based studies, in contrast, enable researchers to estimate the number of individuals in a community who were afflicted in some manner because they focus on the denominator, or the entire community at risk. A study conducted in Iceland after a volcanic eruption in 2010 utilized an existing population registry to identify and survey all adult residents in the municipalities closest to the volcano and an additional sample of demographically matched residents from a non-exposed area in the northern part of the country.Reference Carlsen, Hauksdottir and Valdimarsdottir⁴¹ This population-based cross-sectional survey was able to estimate the proportion of the population afflicted by symptoms likely related to the volcano eruption and determine that residents living in the exposed area had markedly increased prevalence of respiratory and other physical symptoms compared to non-exposed residents. The dose–response pattern that emerged, with the highest symptom prevalence found in those living closest to the volcano, strengthened the evidence that the symptoms found in the study were caused by exposure to the eruption.

Disaster research may also occur in different temporal contexts. An organizational structure for disaster planning, response, and research conceptualizes disaster events as occurring in a cycle. There are slight variations in the way different researchers divide and label the critical periods, but three phases are common to all schemas.Reference Tierney, Lindell and Perry⁴² These are the “pre-impact,” “trans-impact,” and “post-impact” periods, also described as the “disaster mitigation and preparedness,” “emergency response,” and “disaster recovery” periods. The U.S. National Research Council recommends that cycles typical of hazards on one hand, and disasters on the other, be integrated in recognition of the importance of collaborative cross-disciplinary research.⁴³

The pre-impact period is the time frame leading up to a disaster event. This period involves two major activities, hazard mitigation and disaster preparedness, which help reduce vulnerability to disaster impact. Emergency preparedness planning and research may be conducted during this phase. Baseline information about disaster readiness and emergency planning may be collected as well. The trans-impact period focuses on warning, evacuation, immediate response, and disaster relief activities. The post-impact period revolves around disaster recovery. It is important to note that these divisions serve as an organizational scheme and are neither fixed nor absolute. In fact, they may blend together depending on the outcome of interest.

More recently, studies have been conducted during all phases of the disaster cycle, extending the window of post-impact data collection and using longitudinal designs (comparing data before and after a disaster) when appropriate baseline data are available. The notion that disaster-related memory is stable over time is supported by research conducted in three successive time periods following the 1994 Northridge earthquake in California.Reference Bourque, Shoaf, Nguyen and Stallings⁴⁴

The stages of the “disaster cycle” can be related to the different levels of morbidity and mortality prevention. Within the field of epidemiology, the term “prevention” is broadly used to understand the spectrum of efforts to eliminate or reduce the negative consequences of disease and disabilityReference Last⁴⁵. Traditionally the term has been defined in levels of primary, secondary, and tertiary prevention to help delineate different healthcare foci. Primary prevention involves individual and group efforts to protect health through activities such as improving nutrition and reducing environmental risks. These efforts are made before disease or disability occurs, and they are the main focus of public health. In terms of the health threats posed by disasters, primary prevention efforts represent individual and group disaster mitigation and preparedness activities.

Secondary prevention consists of measures that facilitate early detection and treatment, such as health screening, to control disease or disability and reduce the potential for harm. In terms of disasters and their health consequences, secondary prevention can be likened to early warning systems, evacuation efforts, and immediate disaster response and relief because these efforts are designed to reduce later harm in the face of a newly introduced disaster health threat.

Tertiary prevention strives to reduce the long-term impact of disease and disability by eliminating or reducing impairment and improving quality of life. These efforts are generally the focus of rehabilitation. Tertiary prevention of disaster-related health effects might be understood as disaster recovery efforts, in which the goal is to eliminate impairment caused by a disaster and to rebuild communities and infrastructures. Figure 1.1 integrates the temporal stages of a disaster, levels of prevention, and disaster-related activities.

Figure 1.1. The Disaster Cycle.

Disaster Research Variables

Regardless of the phase of the disaster cycle that is being studied, the choice of research variables requires careful consideration. This selection is guided by the researcher's disciplinary or theoretical background as well as by the unit of analysis (i.e., individuals, groups, organizations, or communities). Variables that are expected to have an effect on the outcome of interest are the independent variables. A key independent variable in epidemiologic disaster research is the level or dose of exposure to a disaster. This exposure can be measured in various ways, such as the intensity of shaking experienced in an earthquake or the extent of personal loss due to a disaster. Alternatively, dose can be measured in terms of pre-disaster exposure to public information campaigns or other preparedness messages.Reference Wood, Mileti, Kano, Kelley, Regan and Bourque⁴⁶ Demographic characteristics of the population at risk or those exposed to the disaster are also considered as independent variables or as effect modifiers that influence people's experiences in an event.

The range of possible outcomes or dependent variables in disaster research is extremely wide due to the multidimensionality of the disaster phenomenon and the corresponding multidisciplinary nature of disaster research. The major disciplines involved in disaster research today include geography, geology, engineering, economics, sociology, psychology, public policy, urban planning, anthropology, public health, and medicine.

Geographers and geologists study the relationship between human settlements and hazards (e.g., earthquake faults, hillsides, and floodplains), or the “hazardscape,” and engineers examine the extent of structural damage that can be caused by a disaster. Economists assess the economic and financial impact of disasters, sociologists and psychologists study the behavioral responses to disasters and disaster risk, and health professionals are primarily interested in the effect of disasters on people's health and the healthcare infrastructure. Depending on when (i.e., during which part of the disaster cycle) the dependent variables are measured and how the study is designed, researchers can forecast the amount of loss and damage that might be done or prevented, measure the actual impact of a disaster, assess the effectiveness of interventions in reducing disaster impact, and predict the course of long-term recovery, each in terms of the dependent variables of interest to the researcher.

As the number of disasters increases worldwide, the field of disaster research grows, with new disciplines being added or previously minor disciplines becoming more prominent. These changes affect the dependent variables that are studied in disaster research. For example, subsequent to September 11, 2001, the study of terrorism has grown dramatically within this field. Studies have assessed different outcomes of terrorism, including the public's response to terrorism and the health impact of terrorism events.Reference Kano, Wood, Bourque and Mileti⁴⁷ Similarly, the occurrence of SARS and influenza pandemics and their repercussions on a global scale have given further impetus to public health emergency research in recent years.Reference Bensimon, Smith and Pisartchik^48, Reference Fang, Wang and de Vlas⁴⁹

Disaster Research Study Designs

The appropriate study design depends on the research objective; whether it is exploratory, descriptive, or explanatory/analytic (as described earlier); and the feasibility of the study given available resources. The designs described here are frequently used in the social sciences and in epidemiology to study a wide range of phenomena, including those related to disasters.

Experimental studies involve comparing outcomes between those who receive a certain treatment and those who do not, holding all other known factors constant. A treatment can be any independent variable that is expected to have an effect on the dependent variable. In experiments, the researcher controls the level of the independent variable, or exposure, in an attempt to isolate its effect. Experiments involve random assignment of subjects to treatment groups (i.e., randomization) to increase the likelihood that the groups will be comparable in regard to characteristics other than the main independent variable that may affect the outcomes. Truly experimental designs can offer evidence with the highest internal validity (i.e., evidence of causality) and thus are suitable for explanatory research. As an example, researchers tested the effectiveness of a behavioral treatment for earthquake-related posttraumatic stress disorder by randomizing a group of survivors of the 1999 Turkey earthquake with a clinical diagnosis of posttraumatic stress disorder into treatment and non-treatment groups. This study identified significant effects of the behavioral intervention at weeks 6, 12, and 24, and 1–2 years post-treatment. Experiments might also be conducted in which human subjects are not involved, for example, to test whether certain structural designs mitigate damage in an earthquake. They are not used, however, to investigate how people are affected by or respond to disasters because it is unethical and, in most cases, impossible to manipulate exposure to a disaster.

There are many natural social settings in which the researcher can approximate an experimental design without fully controlling the stimuli (determining when and to whom exposure should be applied and randomizing the exposure) as in a true experiment. Collectively, such situations can be regarded as quasi-experimental.Reference Cook and Campbell^50–Reference Rothman and Greenland⁵² Quasi-experiments are frequently used in the social sciences for explanatory research. This includes studies in which a group of individuals who were naturally exposed to a disaster is compared to a group of non-exposed individuals, or to those with varying degrees of exposure, to identify possible differences in the occurrence of key outcomes. In the absence of an actual disaster, level of exposure to disaster “risk” (e.g., distance from a hazard) instead of exposure to the disaster itself can be the exposure of interest in studying certain behavioral responses (e.g., emergency preparation). As will be discussed in a later section, people may also be indirectly exposed to a destructive event, for example, via media reports.

In epidemiology, study designs that are not experimental, including quasi-experimental designs, are called observational studies.Reference Kelsey, Whittemore, Evans and Thompson³⁹ Here, subjects are studied under natural conditions without any intervention by the researcher. Only naturally occurring exposures and outcomes are examined in these types of studies. A cohort study is one of the typical designs used in epidemiology in which the researcher identifies a group of exposed individuals and a group of non-exposed individuals, or individuals with varying degrees of exposure, and follows the groups to compare the occurrence of specific outcomes. In disaster research, for example, long-term health outcomes could be compared between groups of residents in the same disaster-affected community based on their level of exposure to the index disaster or between residents of a disaster-affected community and residents of a similar community not affected by a disaster.

Following the 1995 Kobe, Japan earthquake, a cohort of school children were assessed for posttraumatic stress reactions at four points in time over the 2 years after the disaster. Children who lived in areas directly affected by the earthquake were compared with children of the same age group who lived in distant areas that were not directly affected.Reference Uemoto, Asakawa and Takamiya⁵³ It was found that greater exposure to the earthquake was associated with more fear, anxiety, and depression or physical symptoms, with younger children exhibiting greater vulnerability. Exposure was defined as the extent of survivors’ experiences related to home damage, injuries to oneself, fatalities or injuries among family members, and having to be rescued or to stay in shelters after the earthquake.

Another common study design in epidemiology applied to disaster research is the case-control study. As with cohort studies, this design is appropriate for explanatory research aimed at understanding the association between exposure and outcomes. In contrast with cohort studies, however, instead of determining exposure status first and then observing outcomes, a case-control study begins by identifying groups of people who naturally have or do not have the outcome of interest (i.e., cases and controls, respectively) and then retrospectively determining their exposure status. For example, a matched case-control study was conducted in a village of Southern China where a powerful typhoon struck in August 2006.Reference Shen, Feng and Zeng⁵⁴ A census was conducted to determine residents who had died or been injured in the typhoon (i.e., the outcome of interest). A comparison with those residents who had survived without injury led to the identification of risk factors for typhoon-related injury and death. These included proximity of the house to the sea and behavioral factors such as failure to reinforce doors or windows and staying near a door or window during the typhoon.

Quasi-experimental, cohort, and case-control studies can all offer relatively high internal validity. They can also maximize external validity, or generalizability to a larger population, if population-based sampling is used. One of the major challenges to using these designs is defining disaster exposure. For example, one might posit that everyone in the United States was exposed to the September 11, 2001, attacks on the World Trade Center and Pentagon, even though most people were not proximal to the disaster sites. Nonetheless, they may have experienced it vicariously through the media, their friends, or family. A quasi-experimental study conducted in the United Kingdom (UK) compared the responses collected before and after the September 11 terrorist attacks in a longitudinal household panel survey. Investigators demonstrated that a terrorist attack in one country negatively impacted the wellbeing of residents in another country through vicarious exposure.Reference Metcalfe, Powdthavee and Dolan⁵⁵ However, the amount of this exposure was not measured or validated in this study. As an example, researchers did not assess whether or how much the respondents had actually viewed or heard any media coverage of the event. Rather, given the extensive and prolonged worldwide media coverage of the disaster, it was assumed that all of the surveyed population had been exposed by the time of the post-9/11 survey. Epidemiologists are often interested in identifying dose–response relationships, that is, the relationship of observed outcomes to varying levels of exposure. A dose–response relationship strengthens the internal validity (i.e., causal claims) of the research findings.

Observational study designs are also appropriate for descriptive studies, in which the objective is to accurately describe the distribution of variables or associations between variables in a population. Non-experimental designs have low internal validity but can have a high degree of external validity if they are conducted with a probability sample of the population. The greatest challenge in conducting population-based studies in disaster research is identifying the population to which the study results can be generalized, or, in other words, establishing the denominator for population estimates. Catastrophic disasters further complicate this issue because the population from which data is collected may be unstable or different from what it had been before the event due to disaster-associated in- and out-migrations, deaths, and alterations in procedures used to compile administrative records.Reference Smith and McCarty⁵⁶ A study of migration patterns in the wake of Hurricanes Katrina and Rita in the United States found that places characterized by greater proportions of disadvantaged populations, housing damage, and more densely built environments were significantly more likely to experience out-migration following the hurricanes.Reference Myers, Slack and Singelmann⁵⁷ In this case, surveying only the people who remained in the area will likely underestimate the full extent of physical and socioeconomic damages caused by the storms. Any population-based study conducted post-disaster will need to account for such nonrandom patterns of change in the base population.

A non-experimental observational study design that is suitable for in-depth, exploratory research is the case study (or a case series). In this type of study, cases are deliberately selected for examination without insuring they are statistically representative of a population, thus compromising external validity. Internal validity is also low because systematic comparisons between cases and non-cases are not performed. The main benefit of case studies is that they lead to a better understanding of rare or new phenomena and the development of hypotheses. Much of the early disaster research in the social sciences used case studies (see earlier section on Historical Overview of Disaster Research). Case studies are also used in disaster medicine and epidemiology to describe the unique characteristics of deaths, injuries, illnesses, and other health outcomes associated with disasters.⁵⁸

In addition to the distinction between experimental and observational designs, there is also a difference between methodologies in terms of how frequently data are collected over a study period. When data are collected at only one point in time, it is called a cross-sectional or prevalence study. It is best used to describe the state of a population at a given time. Cross-sectional designs can also be used to identify causal associations between variables, where the evidence for causation is based on the application of theory and inferential logic rather than time sequence.Reference Aneshensel⁵⁹ In other words, theoretical models are used to determine whether the hypothesized independent variable logically precedes the dependent variable. Therefore, cross-sectional designs can also be used in explanatory research, although they are most naturally used for descriptive research.

Cross-sectional studies conducted before a disaster occurs can provide valuable baseline data on health status, knowledge of risks, attitudes toward preparedness, and actual preparedness behavior at the individual, organizational, or community level. In reality, most disaster studies using a cross-sectional design are conducted after the event has occurred to assess its impact. Examples of these kinds of studies include the post-disaster, rapid health surveys routinely conducted by the CDC as well as by local public health officials. Results of post-disaster, cross-sectional studies must be interpreted with care, especially when baseline data are not available. Although it is tempting to associate post-disaster observations with the index event, it must be recognized that findings from a post-disaster, cross-sectional study reflect conditions that existed before the event as well as conditions that arose during or afterward. Therefore, not all cases or conditions identified in a post-disaster, cross-sectional study are new (i.e., incident cases). The cases identified in a cross-sectional study, including both old and new, are referred to as prevalent cases.

Even new cases that occur after a disaster may have little or no causal association with the incident itself. Among the prevalent cases identified after an event, errors are frequently made in distinguishing between incident cases (or conditions) caused by the disaster, incident cases unrelated to the disaster, preexisting cases that were exacerbated by the disaster, and preexisting cases that were unaffected by the disaster. Chronic conditions are especially prone to such classification errors, although a carefully designed study can allow researchers to make causal attributions to the index event. As an example, following the magnitude 8.8 earthquake that struck Chile in 2010, the Chilean government re-interviewed a subsample of respondents to a national socioeconomic survey for a longitudinal study of posttraumatic stress symptoms, which had just been completed before the earthquake.Reference Zubizaretta, Cerdá and Rosenbaum⁶⁰ This enabled a clear distinction of pre- and post-exposure conditions. The study also employed new statistical methods to match the exposed and unexposed individuals on forty-six covariates to further strengthen internal validity. As a result, this study was able to produce strong evidence of elevated posttraumatic stress symptoms associated with earthquake exposure.

Longitudinal studies collect data more than once over a long period of time. This methodology is used less frequently than cross-sectional designs because it typically requires more resources and a longer-term commitment to the study. It has the advantage, however, of allowing researchers to examine trends and changes over time. It can also provide stronger evidence for causality because temporal ambiguity is reduced or eliminated. In disaster research, longitudinal designs are often used for documenting a community's course of recovery from a disaster, or for observing changes between periods interrupted by a disaster (i.e., pre- and post-disaster).

Examples of longitudinal designs include repeated cross-sectional studies (in which new samples of the population are studied each time) and cohort studies, also referred to as panel studies or repeated-measures studies (in which data are collected at multiple times from the same group of subjects). Repeated cross-sectional designs are especially useful when pre-disaster data are available for a population that was later affected by an incident. To illustrate, a study was conducted to estimate the impact of Hurricane Katrina on mental illness by comparing results of a post-hurricane survey with those of an earlier survey.Reference Kessler⁶¹ The populations from which the probability samples were drawn were comparable (although the post-hurricane population frame was limited to survivors) and the measures used to assess outcomes were identical. Results showed that the estimated prevalence of mental illness doubled after the hurricane.

Although repeated cross-sectional studies have the advantage of studying samples that are representative of the population at each time of data collection, cohort studies allow for the examination of change over time within a group. Cohort studies, however, often suffer from loss of follow-up (i.e., respondents intentionally or unintentionally stop participating in the study). For instance, respondents to a survey conducted after the 1994 Northridge, California earthquake were re-interviewed 4 years later to determine if their prior experience affected their response to another anticipated disaster, a slow-onset El Niño weather pattern.Reference Siegel, Shoaf, Afifi and Bourque⁶² Of the 1,849 households originally interviewed after the 1994 earthquake, 1,353 (73%) agreed to a follow-up interview, but less than half of them could be contacted at the time of the follow-up study. Ultimately, 414 were interviewed, yielding a 22.4% response rate of those interviewed at baseline. Loss of follow-up is expected to be high in areas where the population is very mobile, such as in large urban areas.

A further aspect of study designs is the timing of data collection in relation to the outcome of interest associated with the index disaster. In a concurrent design, both exposure and outcome data might be collected at the time the event occurs, or shortly afterward. In a prospective study, which is only possible using a longitudinal design, exposure data are collected from the target population before the event (in this case, the disaster) has occurred, and outcome data are collected subsequently. In these instances, the study may be initiated for other purposes but can be adapted to the disaster researchers’ needs. Lastly, in a retrospective design, data are collected on events or conditions that occurred in the past by using archival or recalled information. An example here is a study based on a review of hospital records after a disaster. Case-control studies are retrospective by design because prior exposure data are collected after cases are identified. Although most observational studies can use any one of these designs, or a combination of them, experiments by definition can only be concurrent or prospective because it is impossible to go back in time to manipulate study variables.

Some study designs have been underutilized in disaster research. Case studies using laboratory simulations were used in early disaster research, but have not been used in recent times, perhaps because of the difficulty of simulating the complexities of a disaster.Reference Drabek and Haas⁶³ Moreover, the external validity, or generalizability, of results from laboratory simulation studies might be compromised due to the highly artificial and decontextualized nature of a laboratory setting. It has been noted, however, that disaster simulation exercises in the field, which are routinely conducted to train emergency management personnel, are underutilized opportunities for disaster research.Reference Drabek and Stallings⁷ A recent study has also noted that disaster exercises are an opportunity to test and evaluate research protocols.Reference Legemaate, Burkle and Bierens⁶⁴

Retrospective designs are relatively underutilized compared to concurrent or prospective designs because they are not as suitable for research during the immediate post-disaster period, where researchers frequently focus. These include retrospective case studies (which involve the historical analysis and reconstruction of events that occurred in the past), historical cohort studies (which involve the analysis of data on cohorts that were followed up in the past), and case-control studies.Reference Scanlon and Stallings¹⁰ Case-control studies are appropriate for studying rare outcomes and thus would be suitable for studying disaster-associated phenomena.

Disaster Research Data Collection

As with most other types of investigations, disaster research utilizes both qualitative and quantitative data. Qualitative data are often collected through field observations, in-depth interviews, focus group discussions, and archival research. They offer very detailed information about a specific individual or group, place, time, and/or phenomenon that is of interest to the researcher. Qualitative data collection methods are frequently used in exploratory or descriptive studies in which the objective is to investigate an issue or describe a phenomenon about which there is little existing information. A historical example of qualitative disaster research is Form and Nosow's 1958 study of community responses to a tornado in Michigan.Reference Form and Nosow⁶⁵ A more recent study examined the dynamics of existing and emerging social networks among Latino survivors of Hurricane Katrina, which struck the Gulf Coast of the United States in 2005.Reference Hilfinger, Barrington and Lacy⁶⁶ Data gathered from individual, in-depth interviews brought to light: 1) the role of social networks in gathering information, making decisions, and accessing resources; 2) broader structural constraints, including poverty, a lack of transportation, and marginalized status as immigrants; and 3) the emergence of new, if temporary, social networks based primarily on shared nationality, language, and a sense of collective commitment. These and other examples of well-performed qualitative disaster research demonstrate that despite a common misperception that qualitative studies are less scientifically rigorous than quantitative studies, they are indeed important and have been published in prominent journals.

Quantitative data complement qualitative data by expanding the breadth of knowledge about a particular issue. The most popular and efficient method for collecting quantitative data is the use of surveys based on representative sampling. Surveys can be of individuals, households, institutions, or communities, and data in surveys can be collected with questionnaires and record reviews. Surveys of individuals are typically conducted using questionnaires that are self-administered by the respondent or administered by interviewers over the telephone or in person. For surveys of households, organizations, or communities, a representative of the group can be designated to participate in the survey instead of all members of the group.

Survey topics that are common in social science research include: pre-disaster knowledge, attitudes, and behaviors; immediate emotional and behavioral responses to an event; and the course of post-disaster recovery. Commonly perceived limitations of survey use in disaster research include victims’ reluctance to discuss their experiences with investigators and the inconsistent reliability of self-reports, although these concerns have been refuted by several researchers.Reference Bourque, Shoaf, Nguyen and Stallings^{44, 67,} Reference Norris and Kaniasty⁶⁸ Another obstacle to using surveys for disaster research is the recent general decline in participation rates for household surveys.Reference Curtin, Presser and Singer⁶⁹

Surveys of individuals, healthcare providers, and healthcare organizations are heavily utilized in disaster epidemiology to obtain quantitative data about the health status of a population and possible associations between disaster exposure and health outcomes. These data are critical for assessing the immediate and ongoing healthcare needs in a population during and following a disaster. In addition to directly surveying members of the population, epidemiological disaster surveys often collect aggregated data from healthcare providers, emergency response agencies, coroners, and other relevant sources, either prospectively or retrospectively. Public health officials might survey emergency shelters on a weekly basis by reviewing medical records to enumerate shelter residents diagnosed with acute respiratory and gastrointestinal illnesses to detect possible outbreaks of infectious disease.

Standardization of the data collection method is especially important with quantitative data in order to allow comparisons across different events, populations, settings, and times. In this respect, post-disaster rapid health surveys frequently suffer from inconsistencies in sampling methods, data reporting periods, criteria for establishing whether health outcomes are disaster-related, and completeness of data collection for identifying disaster-related injuries and medical conditions. Lack of standardized definitions and survey instruments is one of the major challenges to quantitative data collection in disaster research.

While qualitative and quantitative methodologies complement each other, use of both together in a single investigation (referred to as mixed methods) has yet to become an established approach in disaster research. Mixed methods is broadly defined as research in which the investigator collects and analyzes data, integrates the findings, and draws inferences by using both qualitative and quantitative approaches in a single study or program of inquiry.Reference Tashakkori and Creswell⁷⁰

The Multihazard Mitigation Council of the National Institute of Building Sciences in the United States conducted a mixed method research study to determine the future savings gained from investments by FEMA in hazard mitigation activities.⁷¹ FEMA funded hazard mitigation projects to reduce loses from earthquakes, high winds, and floods. Future savings from these endeavors were measured in two interrelated studies by using different methods to address the common question: What is the ratio of hazard mitigation benefit versus cost? The first study component used benefit/cost ratio analyses and a statistically representative sample of FEMA mitigation grants so that findings in the sample could be applied to the entire population of these FEMA grants. In the second study component, eight communities were selected using purposive sampling to examine if, why, and how mitigation activities percolate through communities. Field studies were conducted in each community by using semi-structured telephone interviews with informants, field visits, and the review of documents. Findings suggest that natural hazard mitigation activities funded by the three FEMA grant programs between 1993 and 2003 were cost effective and reduced future losses from earthquakes, wind, and floods; yielded significant net benefits to society as a whole; and represented significant potential savings to the federal treasury. Specifically, the quantitative benefit/cost analysis found that on average, every dollar spent on natural hazard mitigation saves society approximately $4. The community studies suggest that the 1:4 cost/benefit ratio may be an underestimate because federally funded hazard mitigation often leads to an increase in non-federally funded mitigation programs.

Information Technology Applications in Disaster Research

The application of information technology (IT) in both disaster management and disaster research has remarkably advanced in recent years. The use of Geographic Information Systems (GIS) technology is a prime example. Dash, Thomas, and colleagues have written on the use of GIS technology in disaster management and research.Reference Dash and Stallings^72, Reference Thomas, Kivanç, Kemeç, Rodríguez, Quarantelli and Dynes⁷³ There have also been discussions on the utility of GIS-based spatial analysis in health research and epidemiology.Reference Moore and Carpenter^74, Reference Rushton⁷⁵ The main strength of GIS technology is its ability to integrate geographical data with other information, such as demographic data, extent of physical damage caused by a hazard, morbidity and mortality rates, and access to resources. It also has the capability to analyze data as well as to generate maps and other visual summaries of the data.

FEMA has developed a software program, HAZUS-MH, which uses GIS technology to map and display hazard data and also to produce estimates of potential losses (i.e., physical damage, economic loss, and social impact) from earthquakes, floods, and hurricane winds. GIS-based risk assessment tools such as these are extremely useful to disaster management officials and policymakers who are responsible for developing and implementing disaster mitigation, preparedness, and response strategies for geographically defined areas.

The most common application of GIS technology in disaster epidemiological research is to facilitate post-disaster rapid assessment surveys, which frequently use cluster–random sampling. The cluster–random sampling design, which was originally developed to estimate immunization coverage in a population, allows investigators to obtain expedient and accurate population-based information at relatively low cost.Reference Malilay, Flanders and Brogan⁷⁶ GIS is used to aid the random selection of households, field navigation, data management and analysis, and presentation of results. For example, less than 3 weeks after Hurricane Katrina struck Hancock County, Mississippi, the CDC was asked to conduct a rapid assessment of public health needs. Using GIS, they cluster–random sampled 200 households. Using global positioning system technology to navigate to those locations, they physically surveyed 197 households and completed interviews with 77 of them in 2 days.⁷⁷ The results of the assessment, which indicated a need for water, trash/debris removal, and access to health services, were provided to the state health department and emergency management to guide relief and recovery operations.

There are other applications of GIS in disaster epidemiology that involve more extensive data collection and spatial referencing. Peek-Asa and others used GIS to link separate data elements derived from studying the 1994 Northridge, California earthquake. These included geophysical characteristics (i.e., shaking intensity, strong ground motion, and soil type), individual characteristics of people who were injured in the earthquake (i.e., physical address and demographics), and building data (i.e., damage state, year of construction, structure type), each obtained from a different source.Reference Peek-Asa, Ramirez, Seligson and Shoaf⁷⁸ Their analyses indicated that a person's age and sex, intensity of ground motion, and multiunit building structures independently predict heightened risk for injuries in an earthquake.

GIS has the potential to facilitate data collection, analysis, and presentation for describing or predicting the geographical distribution of various disaster-relevant variables. The usefulness of GIS to disaster research, and especially disaster health research, depends on the quality and availability of spatial data. Health data generally lack spatial attributes (i.e., geocodes) unless they were collected specifically for use in GIS. In addition, there is a legitimate concern about preserving individual confidentiality within spatial information. Researchers have shown, for instance, that a map of Hurricane Katrina-related mortality locations in Orleans and St. Bernard Parishes published in a local newspaper could be reengineered to reveal the actual addresses associated with the points, even though the original map included very little secondary spatial data.Reference Curtis, Mills and Leitner⁷⁹

Examples of relatively novel applications of IT include tracking population movements with mobile phone network data; real-time monitoring of calls made to a disaster communication hub (e.g., 211 calls for community information and referral services in the United States and Canada); crowd-sourced online mapping services; geo-targeted imminent threat alerts and warnings delivered to cell phones and other mobile devices; and use of social media for tracking events and distributing alert and warning messages.Reference Bengtsson, Lu and Thorson^80–85 While not without limitations, these applications may pave the way for further innovations in the use of IT to improve the efficiency of rapid assessments and the coverage of disaster relief services, as well as facilitate research-related activities.

Ethics in Disaster Research

As with any study, ethical considerations are integral to disaster research. The central concern is whether the investigational activity could, directly or indirectly, harm the research participants and the wider community. For example, field observations and interviews of evacuees and emergency responders during or immediately after the disaster might impede the progress of relief operations. Likewise, interviewing disaster victims about their experiences has the potential to cause emotional stress and pain, compounding that already caused by the event. Such an investigation might not be justified by the expected benefits of the study. Other ethical considerations include the ability of researchers to maintain a neutral stance. This situation might emerge when grave human suffering seemingly is attributable to social injustice and an incompetent response by the organizations responsible for protecting people's welfare. Despite a sense of urgency to deploy post-impact, disaster researchers must consider these and other ethical issues in designing their study and before having contact with research subjects. Readers are referred to Chapter 7 of this book, as well as to writings by Stallings, Fleischman et al., and Collogan et al. for further discussion about the ethical issues involved in disaster research.Reference Stallings, Rodríguez, Quarantelli and Dynes^4, Reference Fleischman, Collogan, Tuma, Norris, Galea, Friedman and Watson^86, Reference Collogan, Tuma, Dolan-Sewell, Borja and Fleischman⁸⁷

Disaster Morbidity and Mortality

The discussion of disaster morbidity and mortality describes how these estimates are derived, as well as the many factors that can influence their accuracy and introduce variability across studies.