Practical difficulties arise during the investigation and reporting of most accidents. These difficulties include the determination of the scope of the phenomenon to investigate, the identification of the data required, documentation of the findings, development of recommendations based on the accident findings, and preparation of the deliverables at the end of the investigation. These difficulties reflect differences in the purposes for the investigations, which in turn reflect different perceptions of the accident phenomenon. They can lead to questionable data for accident research and other end uses. Five underlying theories about the nature of the accident phenomenon are presented, and the implications of these different theories for accident research are discussed. Increased awareness of accident theories and increased dialogue with investigators is proposed.
I am an accident investigator. What do you want me to deliver to you at the conclusion of my investigations? What criteria should I use to determine if my deliverables are acceptable or unacceptable to you?
These are not rhetorical questions. As an investigator, I am faced with decisions that affect my deliverables each time I investigate a new accident. What should be the scope of my investigative effort? What data should I seek, and why? How will I recognize it when I see it? When do I have enough data? How should I convert the data into my findings? How should I argue and present my conclusions about this accident, and what recommendations should I propose?
The ways that I choose to answer these questions during an investigation will influence my deliverables, and that in turn will affect your research! My answers depend on several considerations, including my views about the purpose of my investigative efforts, my perceptions of the accident phenomenon, and my understanding of the deliverables expected of me. My difficulties are not unique.
ACCIDENT INVESTIGATION PURPOSES
As a proponent of working toward clear objectives (MBO) I have tried to identify the purpose of all my accident investigations. I have also tried to understand the purposes of the other persons involved in these investigations. The diversity of purposes I have found is worthy of discussion, because differences in purposes led me to my awareness of the differences in perceptions of the accident phenomenon.
Accident investigation purposes I have observed are as varied as the number of people interested in a specific accident. The media personnel want to know immediately what “caused” the accident. Their quest for cause probably reflects the public’s curiosity, so one purpose is to satisfy this public curiosity. An injured employee wants to be made whole again or to recoup personal losses by an adequate claim. An attorney wants a basis for litigation and culpability. The regulatory representative wants to find out if the regulations are adequate or if someone should be prosecuted for their violation. An insurer wants to determine claim settlement or subrogation possibilities. A designer wants to learn if a design change is needed. Victims want a basis for recovering their losses. Many policemen want to get the accident form completed. A training man wants new material for his training course. An operator wants to understand his liability, and wants to know if he will have to change his operations. The statistician wants statistics. The accident researcher wants accident data so he can better understand the phenomenon and control it. The investigator wants to satisfy one or more of these purposes.
Why do we have so many purposes to satisfy -— all emanating from one occurrence? Why must we have so many investigators to satisfy these purposes? Why can’t we develop an accident investigation output which would satisfy them all?
The answers depend on the reasons for the diversity of purposes. Perhaps if these reasons were understood, an accident investigator could explain the accident in sufficient detail to satisfy all these interests.
As I have studied the diverse interests and reasons for differing purposes, one striking difference has been observed. This difference is in the perceptions of the accident phenomenon held by these various persons.
My analyses of these perceptions and the actions resulting from each suggest that there are at least five distinctive views of the accident phenomenon held by persons interested in accidents. Each perception involves a differing view of the nature of the phenomenon and is accompanied by a body of assumptions, principles and rules of procedure that, taken together, constitute five differing theories.
The single event theory. This theory is based on the ‘assumption that an accident consists of a single event that has a cause. Find the cause, and you have explained the phenomenon. The investigative task is easy: find the cause and correct it, and you will prevent accidents.
This perception seems rooted in primitive history. If an unusual phenomenon occurred, and there was no ready explanation for it, the survivors sought a scapegoat as the “cause” of the occurrence. Find the “cause” (read scapegoat) and the victims are satisfied.. History offers numerous examples, including acts of God precepts which carry over into present day insurance policies. Anyone who has observed the media’s handling of an accident will recognize some evidence of this perception. While largely discredited by the scientific community, vestiges of this theory can be widely observed. Accidents are still frequently defined as an “event” in safety publications  and military investigation manuals, for example.  Many aspects of our legal system in the highway safety field also provide evidence of this perception, such as police citations for accidents, and some of the arguments about “no-fault” insurance legislation. Publication of accident cause statistics reinforces this view, and the use of these statistics in the media perpetuates it. It is a world-wide perception, as evidenced by the World Health Organizations classification of the causes of death. The principal difficulty is that the view encourages an incomplete examination of the accident phenomenon.
The chain-of-events theory. The perception of sequential events is a popular perception of the phenomenon, widely recognized. Most of us have heard about the sequence that goes “for want of a nail the shoe was lost; for want of a shoe the horse was lost,” etc. The concept was adapted by Heinrich, who gave it the term “domino” theory,  in 1936. His premise was that if a set of “unsafe conditions” (hazards?) set up a row of vulnerable dominos, an “unsafe act” would start them toppling. Under this concept, the investigator looks for information that will help reconstruct the chain of events that constituted the accident. Then as now, the unsafe conditions lack criteria, as do the unsafe acts. These terms represent the investigator’s conclusions, rather than observations of the phenomenon. How investigators arrived at these conclusions was not disciplined by principles or criteria that would provide consistent and reproducible findings. The conclusions were descriptive, and usually symptomatic, rather than etiologic.  As evidence, the interested reader is encouraged to review the taxonomy and choice of entries in the American National Standards Institute’s standard for reporting of accidents  for technical precision and consistency. The indiscriminate mixture of events, conditions, factors and other kinds of entries does not facilitate understanding of the accident phenomenon or development of countermeasures, for this researcher.
The determinant variable theory. The work of Greenwood and Woods  and Newbold  suggested the factorial view of the phenomenon. Their “accident proneness” concept was statistically inferred by examination of available data. The focus on static conditions reflected the view expressed by Thorndike  of the search for the experimental ideal of the single independent variable, and set the goal and ideal of an accident investigation as the gathering of data in such a way that statistical comparisons will permit fair estimates of the influence of variables in a particular factor on the probability of an accident. This included precursor conditions of the actors involved in the accident. The view assumes that some common factors are present in accidents, and that they can be discerned from the right accident data. This view led to the admonition to investigators to “get all the facts” about an accident. Hypotheses can only be generated after the fact. By not prescribing the scope, relevance tests and other data specifications, the theory and its operation require the investigator to make these decisions according to the investigator’s best judgment. The result is that no accidents are reported in a reproducible manner, even when trained investigators use reporting forms. Such reports are almost totally dependent on the conclusionary findings and judgments of the causes or causal factors.
The branched events chain theory. About 1960, the need to predict accident events in the military missile program stimulated development of the “fault tree” analytical method for analyzing missile safety. The method is generally credited to Watson.  It was based on the perception that an accidental launch would occur with some likelihood if a pathway to inadvertent firing were available. The events could flow in a chain-like sequence from a variety of origins in the system toward the undesired, accidental launch event. The method of displaying the events chains that could lead to the “top” event in the “fault tree” pyramid was a significant predictive advance.  This procedure provided for organization of the events and conditions data into a visible, easily critiqued and readily understood display. It also provided for testing the predicted events against their sequential logic, and provided a basis for identifying data needs in the event of a system failure. Its significance is that, while it is an adaptation of the chain-of-events theory, it establishes data requirements that facilitate prediction of the accident possibilities in a given system. It also provides guidance during investigations.
The multilinear events sequences (Process) theory Recent work by the author suggests that accidents are a segment of a continuum of activities, and proposes a process view of the accident phenomenon.  The accident phenomenon is viewed as the transformation process by which a homeostatic activity is interrupted with accompanying unintended harm. The process is described in terms of specific interacting actors, acting in a sequential order with a discrete temporal and spatial logic. The procedures for analyses are defined in terms of changes of state and events (event = actor + action) that produce the change of state, and techniques for generating hypotheses are linked to the procedures. Both the investigative and analytical procedures are based on the premise that “everyone and everything always have to be someplace doing something,” (Benner’s First Law.) This perception formalizes many of the crash trauma concepts advanced by Haddon  and others in the medical field, and borrows heavily from the medical perception of homeostasis. It is also easily integrated into some of the risk-based concepts emerging in the safety field.
These, briefly, are the highlights of the five theories discerned by the author. For convenience, they are summarized as
- The single event perception and “cause theory.”
- The chain—of-events perception and “domino theory.”
- The determinant variable perception and “factorial theory.”
- The branched events chains perspective and “logic tree theory.”
- The multilinear events sequences perspective and “process theory” or “p-theory.”
What do these differing perceptions mean to the accident researcher?
IMPLICATIONS FOR ACCIDENT RESEARCH
From this discussion of accident theories, though abbreviated, several significant implications for accident researchers can be inferred. For convenience of the reader, they are arranged into three categories of interest:
- investigative traps,
- data traps, and
- methodology traps.
Each different theory affects the purpose, scope and method of investigation of accidents. The “cause theory” requires the investigator to search for “the” cause of one aspect of the accident phenomenon, usually the crash or collision. The scope of the investigation is limited to this aspect of the phenomenon and the setting, and as soon as the investigator has “sufficient” evidence to support his conclusion of “cause,” the investigation is terminated and the report prepared. The method of investigation is usually an informal, single person, interview— dominated data search, and the conclusions reported on a predetermined check list or accident reporting form. A comprehensive explanation of the accident is rare, discovery of safety problems even less frequent, and the data are essentially presented as the investigator’s conclusions, except for the accident identifier data, such as time, location, etc.
Much the same comments apply to accidents governed by the “domino theory.” The designations of the beginning and end of the chain of events are usually left to the discretion of the individual investigator, or they are implicitly specified by the entries called for on the report forms. When this perception is combined with the “cause theory,” the selection of the one event in the chain as the cause or a causal factor is purely subjective, even when manuals with indicated entries are available. As soon as the chain is complete in the eyes of the investigator or the investigative team, the investigation effort is terminated. The reconstruction technique provides some discipline for the data search, but criteria for this reconstruction technique are imprecise and unlikely to assure reproducible results.
The “factorial theory” is even less clear with respect to purpose and scope, from the point of view of the investigator. Every “factor” not a part of the accident mechanism is a judgment call. The same beginning and end problems arise. How much is enough data in a given accident is not predetermined nor hypothesized, with few exceptions. Of all theories, this provides the least guidance for the investigator.
The “logic tree theory” focuses on information that helps identify the critical path of events to the undesired event. No criteria exist for designating the undesired event, so the beginning or end of the accident phenomenon are again left to the judgment of the individual investigators. This also has the effect of confining the purpose and scope of the investigation. However, the search and display methods focus investigative effort on relevant data (data needed to complete the tree) as it is being gathered -- a powerful improvement over the other procedures. It also facilitates discovery of new knowledge during the investigation, and organizes speculations if data are not available.
The “P-theory” defines the beginning and end of the phenomenon being investigated, and has a method for identifying the data needed and for testing data relevance as the investigation unfolds. The procedures incorporate the logic tree benefits, and also show time relationships among the events. While the theory provides for a full explanation of what happened, the needs of the investigator may require added information to satisfy extensive “why” questions of an institutional or management nature, and the origins of these influences on the specific phenomenon being studied. *
The traps for both the researcher and the investigator occur when two or more of the theories influence an investigation and its reports. Both parties need to be aware of the effects of shifts from one theory to another during an investigation or research project, and need to watch for such shifts. This is particularly important with respect to the data generated by “shifty” investigations, and to hypotheses generated from such data. An example of such shifts can be found in many research reports if one is alert to the theories and investigative problems described above. One research report of the development of a comprehensive causal network model contained three such shifts on a single page!  The conclusions, by the way, included a call for additional data from new accident investigations, and a study of the time structure to quantify interactions.
Each theory produces different data at the end of the investigation. The causal theory produces conclusions about cause (s). The domino theory produces conclusions about a chain of events and “unsafe conditions” or “unsafe acts” affecting the chain of events —— all judgment calls. The factorial theory produces a list of “factors” which can be events, conditions, changes of state, social circumstances, hereditary characteristics, personal attitudes, actions, failures, errors or anything else judged to be potentially “causal” by the investigator or researcher. The logic tree theory produces a display of the most likely and alternate “critical paths” leading to what the investigator decides is the undesired event or even the most undesired event. The P-theory leads to a display that shows the actions of each actor who or which played a role in the accident, and the relationships among these actions in a logical time and spatial sequence, to a level of detail judged adequate by the investigator.
Thus, each theory has its data promises and problems. Perhaps the most difficult trap for the researcher to avoid is the shift between investigative observations and investigators’ conclusions. For example, the conclusion “following too close” is not useful for identifying countermeasures until the data on which the conclusion was predicated are available in process form. The aggregation of such conclusionary data for the purpose of drawing new conclusions can lead to spurious or worse results.
A second trap is the classification of conclusionary data reported about accidents. For example, the variety of taxonomies used to categorize the role of alcohol in traffic accidents reflects the problems with the handling of conclusions in reports, and until the role of alcohol is understood in the context of the driving actions by drivers involved in accidents, through improved accident analysis or simulations, the most effective taxonomy will elude us.
A third data trap relates to the generation of hypotheses. The use of conclusions of accident investigators and their investigation reports as a basis for new hypotheses seems highly risky, at best, and could be utterly futile -— depending on the data. If the criteria for the investigator’s decisions are vague, arbitrary or non-existent, the likelihood of futile results increases. A review of the quality of the conclusionary data in research reports suggests that this is a serious research problem in the accident research field.
The scientific method is based on observation--of phenomena -- and the subsequent formulation of hypotheses, then experimentation and validation of the hypotheses, and ultimately control of the phenomenon. In the accident research field, the phenomena are observed -—albeit retrospectively--by accident investigators who conduct the investigation. And these investigators operate to at least five differing perceptual sets or combinations of these sets. The criteria for recording observations vary from set to set, ranging from essentially no criteria to tightly defined criteria. The chaotic state of the available accident data base is easy to understand. Accidents are generally considered uninstrumented experiments. However this does not mean that changes of state are unrecorded. They can often be observed or logically inferred from surviving people or things after the “experiment” is over. Until rules of procedure for recording observations of the accident phenomenon are formulated for investigators, users of the data run a high risk of futile research.
Users also run the risk of applying solutions offered by their disciplines to the wrong problems. Accident theories drive research methods too. For example, the factorial theory demands statistical analysis methodologies, while the P-theory demands operations research, task analysis and other methodologies. More subtly, they influence evaluations of alternative methodologies. For example, a strong statistical and factorial theory bias obscures strengths of alternative theories in a research report on methods for studying pre-crash factors in highway safety research.  The methodology used for research must match the underlying theory that drives the investigative data. Most investigators do not think in terms of the subsequent research use or methodologies that will be fed by their data. If they did, their data would surely improve, and more disciplined investigative methods would result.
The bottom line of my research is this: we all need to take another look at the theories that drive accident investigations and research. Then, the accident research community and the accident investigator community must increase their contacts, and begin to exchange understandings of accident theories, methodologies, and criteria for each others’ deliverables. Until this’ happens, I do not foresee significant accident research breakthroughs in the transportation safety field.
This is beginning to occur. As I participate in or observe the development of risk analysis methodologies, the new analysis techniques emerging from this endeavor are beginning to define the answers to the questions I posed at the beginning of this paper.
I urge you to examine your own theoretical basis for your research, and to determine its relationship to the work of the investigators. Then, undertake an increased dialogue with investigators as a part of your future research efforts. I hope this paper will give you some guidance of value.