The views expressed are those of the author and not necessarily those of the National Transportation Safety Board.
I am an accident investigator. Let us suppose you are my program manager. You have asked me to investigate an accident that has just occurred.
What do you want me to deliver to you at the conclusion of my investigation? What standards should I use to determine if my deliverables are acceptable or unacceptable to you?
What should be the scope of my investigation? What data should I seek, and why should I seek it? How will I recognize the right data when I see it? When will I have enough data so I can go on to my next tasks? How should I organize and present the information I find during my investigation? How many "facts" are enough to give you what you want? How "thoroughly" do you want me to investigate this specific accident? Do you expect me to make recommendations as a part of my deliverables? How do you want me to organize any narrative that I turn in to you?
These are not just hypothetical questions. As an investigator, I am faced with these questions every time I investigate a new accident.
If you don't give me instructions with which I can answer these questions for you, I will answer them the best way I know how. Without standards, however, what assurance do you have that my deliverables will be satisfactory? Or consistent? Or understandable? Or usable?
Without standards, I will answer each question the best way I can in each accident. My answers. may change from accident to accident, the nature, content and "quality" of my deliverables will be affected. That will also affect the value of my deliverables to their users. If other investigators are working on the accident, and we do not begin the investigation with a common set of answers, how do we arrive at agreement about the methods to use, the data and conclusions to report, and the use of the outputs. Which raises other questions - who is going to use my deliverables, and for what purposes, and what are THEIR specifications for my deliverables? If you gave me instructions, where did you get the standards that you want me to use to judge the acceptability of my deliverables? What assumptions, laws, principles or rules of procedure form the basis for those standards, so I will know how to interpret your standards when I encounter questions?
Three sets of criteria are possible: the investigators', the program managers' and accident data users. Let's look at existing practices, to see if there is agreement or disagreement about criteria for answering these questions.
Accident Investigation Purposes
Good management practice dictates that Investigators work toward some objectives: i.e., that they try to define what they want to have accomplished when their tasks are finished. Thus they should diligently seek objectives that they can use as criteria for evaluating accident investigation deliverables. That's where I ran into a problem.
During my investigations, I encountered numerous investigators. Each investigator seemed to have a slightly different idea about the purpose of the investigation. This
diversity of purposes needs to be discussed because differences in purposes led me to my awareness of the differences in the way people perceive the nature of the accident phenomenon.
Accident investigation purposes were as varied as the people interested in a specific accident investigation. Media personnel wanted to know immediately what "caused" the accident. Their quest for cause probably reflected the public's curiosity, so one purpose was to determine cause, and another to satisfy public curiosity. An injured employee wanted to be made whole again for the personal losses incurred. An attorney was interested in culpability and litigation considerations for a client. The regulatory representative wanted to find out if the regulations were adequate or if someone should be prosecuted for violating a regulation. The insurance representative wanted to determine claim settlement or subrogation possibilities. A designer wanted to learn if a design should be changed. A number of policemen wanted to get their forms completed and turned in. A trainer wanted materials for improved training programs. A statistician wanted statistics to analyze. The accident researcher wanted specific data in connection with a safety evaluation project. And the investigator? The investigator was trying to keep everyone happy.
Why were there so many purposes to satisfy - all emanations from a single accident? Why did so many investigators have to investigate the same accident to satisfy so many purposes? Why couldn't one investigative output satisfy everyone? After all, each accident happened only once, IN ONLY ONE WAY. The answers seemed to depend on the reasons for the diversity of purposes. Perhaps, if these reasons were better understood, one accident investigator could describe and explain an accident in sufficient detail to satisfy all these interests. Alternatively, perhaps if an accident were adequately described and explained, the diversity of purposes might be reduced.
As the diverse interests and reasons for differing investigation purposes were studied, one common difference was observed. This difference was the view of the nature of the accident phenomenon held by the various individuals with whom I was working. Once I became aware of these different views I began to see the same differences among other persons involved in safety-related work.
Further observations and analysis of these perceptions, and the actions resulting from each, suggest that there are at least five distinct perceptions of the nature of the accident phenomenon. Each perception results in an accompanying body of assumptions, principles and rules of procedure that - taken together - seem to constitute five differing theoretical bases for accident investigation. In addition to influencing accident investigations, these perceptions seem also to influence views about safety and the nature of safety programs, safety research, and other safety matters.
Let's examine these five distinct perceptions in what appears to be their ascending order of complexity.
Single Event Perception. This perception treats an accident as a single event, or perhaps assumes that an accident can be simplified by thinking of it as a single event. This perception is rooted in primitive history. If an unusual natural phenomenon occurred, and there was no obvious explanation for it, the survivors sought a simple way to explain the phenomenon. Often the search for satisfaction led to a "scapegoat" that "caused" the phenomenon. Find the "scapegoat" (read "cause"), take care of the scapegoat, and the problem was resolved. History offers numerous examples of this perception, including the still-prevalent "act-of-God" reservations in present day insurance policies. Anyone who has observed the media's initial handling of an accident story will recognize evidence of this perception. While largely discredited by the scientific community, vestiges of this perception still surround us. Accidents are still frequently defined as an "event" in safety publications and accident investigation manuals.  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" legislation. Publication of accident cause statistics reinforces this view, and the uses of accident cause statistics in safety research perpetuate this view. It is a world-wide perception, as evidenced by the World Health Organization's classification of the causes of death 
Based on this perception, investigators look for the "cause" of the accident.
Chain-Of-Events Perception. This view treats an accident as a chain of sequential events, as is illustrated by the legendary "for want of a nail the shoe was lost; for want of a shoe the horse was lost;" etc. The perception was formalized by Heinrich, who seems to have introduced the "domino" term to the safety literature in 1936. His premise was that if an 'unsafe condition" (hazard?) set up a row of vulnerable dominos, an "unsafe act" would start them toppling. The perception may be rooted in our written language forms; events can only be presented one at a time, in sequence as one writes about a phenomenon using sequential arrays of letters to make words, then sequential arrays of words to make sentences, and then sequential arrays of sentences to make paragraphs.
Based on this perception of the accident phenomenon, investigators look for information that will permit the "reconstruction" of the chain of events in the accident. The search usually focuses on "unsafe acts" or "unsafe conditions." "Cause" is modified by terms such as proximate, primary, remote, etc., to accommodate the increased complexity of the perceived phenomenon addressed in the investigation. The perception seems widespread in the legal field.
Determinant Variable Or Factorial Perception. This perception is perhaps best understood in the accident investigation field by referring to a report by Thorndyke, in which he described "the search for the experimental ideal of the single independent variable." He formulated "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 a variable in a particular factor on the probability of an accident." The perception assumes that some common "factors" are present in accidents, and that they can be discerned by statistical manipulation of the "right" accident data.
This perception leads to the admonition to investigators to "get all the facts" so they can be examined later to isolate those "factors" that are not due to chance. Reported accident "causes" and "causal factors" are widely analyzed and discussed.
The Logic Tree Perception. This perception presumes that converging chains of events lead to an undesired event. About 1960, the need to predict accident events in defense missile programs stimulated development of the "fault tree"
analytical method for analyzing inadvertent missile launches. The method is generally credited to Watson. It was based on the perception that an accidental launch would occur with some predictable likelihood along each possible pathway if several pathways to firing were available. The events could flow in chain-like fashion from a variety of origins in the system toward the undesired, accidental launch event. Methods for displaying the branched events chains or pathways to the "top event" in the "fault tree" pyramid provided a way to make predictions about the "safety" of these systems.
The perception and the resultant display procedures provided for the logical organization of the events and conditions data into a visible, easily critiqued and readily understood display. Relationships among events and conditions could be observed simultaneously in these displays, without the narrative language limitations mentioned earlier. The displays provided a way to test the predicted events against their sequential logic, which was not possible earlier. Displays showed accident investigators what specific data was needed to verify the speculations on the "fault tree."
The Multilinear Events Sequence Perception. An accident report published in 1962 suggests the perception of accidents as a segment of a continuum of activities. This aircraft accident report presented the aircraft performance data from a flight recorder in parallel rather than in series. This showed how many events were occurring simultaneously before and during the accident being reported. The display requires a process view of the phenomenon, and suggests that the phenomenon be viewed as the transformation process by which a homeostatic activity is disrupted with accompanying unintended harm. The accident process can be described in terms of specific interacting actors, each acting in a sequential order with discrete temporal and spatial logical relationships. By breaking down the events seen as the accident into increasingly more definitive sub-events, the understanding of the phenomenon increases with each successive breakdown. Procedures based on this perception have been formalized, based on building blocks consisting of events (event = 1 actor + 1 action) arrayed into multilinear strings of events much like a musical score A set of investigative principles for preparing the building blocks and arraying them has been developed and used for teaching the investigative method.
This perception borrows heavily from the perception of homeostasis common in the medical field, as applied by Pask to teaching machines and incorporates many of the crash trauma concepts advanced by Haddon.
Strengths And Weaknesses For Accident Investigation
As each perception and resultant investigative methodology was applied, during investigations, strengths and weaknesses of each were observed. These strengths and weaknesses are summarized below.
Single Event Perception. This perception has only one possible strength; it tends to concentrate attention on a single corrective measure. If the correction is - by chance - defined properly so it will be effective, this concentration can help bring about a beneficial change.
The principal weakness of this view is that it demands an overly simplified explanation of accidents, which encourages .misunderstandings about accidents and safety work. Technically, the perception demands conclusions especially about "cause" - that cannot be reconciled with investigators' observations during accident investigations. The perception requires the investigator to search for "the" cause of one aspect of the accident, usually the crash or collision or fall, etc. The scope of the investigation is limited to this aspect of the phenomenon and setting; as
soon as the investigator has enough evidence to support his or her conclusion as to the "cause", the investigation is terminated and the report completed.
The method of investigation is usually an informal, single-person interview-dominated data search, and the conclusions are reported on a predetermined check list or accident reporting form. A comprehensive explanation of the accident is rare, discovery of safety problems even rarer, and the data are essentially presented as ''facts" when in reality they are investigator's conclusions.
Chain Of Events Perception. The reconstruction technique provides some disciplining of the data search, in that the chain concept calls for sequential ordering of data. The methods used to implement this perception demand the use of sequential time-logic tests for the events selected. However, criteria for the selection of data used during the reconstruction activity are imprecise and very unlikely to lead to reproducible results.
The investigative problems are also similar to those with the single event perception. Designation of the beginning of the "chain" or the end of the "chain" is left to the discretion of the individual investigators, or it is implicitly specified by the entries called for on reporting forms. When this perception is driven by a desire for "cause" the selection of specific unsafe acts or conditions is essentially subjective, especially when the "chain of events" is lengthy. As soon as the chain of events is complete in the eyes of the investigator, the effort is terminated. When it was developed, as now, "unsafe conditions" lacked technically valid criteria that would lead investigators to reproducible conclusions about the conditions selected as unsafe. The same holds true for "unsafe acts." The conditions or acts called unsafe by the investigator represent the investigator's conclusions, rather than observed data generated by the phenomenon. How investigators arrive at these conclusions is essentially undisciplined by criteria that produce complete or replicable explanations of an accident. The conclusions are descriptive and usually symptomatic, rather than etiologic  As evidence, the reader is encouraged to review the taxonomy and. choices of entries in the American National Standards Institute's standards for reporting accidents for their technical precision and consistency of their nature (events vs. conditions or states vs factors vs conclusions).
Factorial Perception. A strength of the Factorial perception is its openness to the possibility of discovering previously undefined relationships, and its effort to distinguish determinant from chance relationships. The effort is based on statistical methodologies.
A major weakness is the total dependency upon data reported by accident investigators. Although the "facts" used for analyses should have a probability of 1, the lack of tests for the accident "facts" recorded during the investigation undermines the validity of their subsequent use. During investigations, the primary emphasis is on data gathering rather than analysis.
Hypotheses are only identified after one scans the data reported from a statistically significant number of accidents. "Factors" are vulnerable to equivocation logic defects, and the data furnished to support the ideal depend heavily upon forms which reflect the analysts' hypotheses and assumptions about the nature of the accident phenomenon. The investigative data seem to be driven by statistical methodologies rather than the nature and logic of the data or observations recorded. Accident scope and data specifications or tests are usually not prescribed for investigators. The result is that investigators establish their own bounds for the
phenomenon they will investigate, and establish their own criteria for the "factors" they record. In effect, these factors then represent the conclusions of the individual investigators, rather than observed value-free data generated by the accident phenomenon.
The determinant variable perception differs from the single event perception in that in practice it seems to consider conditions in addition to events as "causal" in accidents. Close inspection of accident "data" generated under this perception reveals indiscriminate mixing of events, conditions, phases, causes, times, places, and almost any other characteristics one can imagine. Because any "factor" found in an investigation may identify a potential determinant variable, all the "factors" must be recorded by the investigator. This unstructured view lacks disciplining criteria for investigators, with the further result that data reported are not independently reproducible. Every "factor" interpreted to be a part of the accident mechanism becomes an individual's judgment call. Every bit of data in a given accident becomes a judgment call: is it the "right" data and should the investigator record it? Of all perceptions, this determinant variable view provides the least guidance for an investigator.
Because of its reliance on statistical methodologies for analyses, the perception obscures the need for investigative methodologies that will provide an understanding of the accident phenomenon from one or a few accidents, in a way that will expose control actions directly rather than by statistical identification of the problem. Handling of time relationships among "factors" is another major problem area when operating under this perception; lack of these relationships may be a fatal weakness of the accident data generated under these methodologies.
Logic Tree Perception. The greatest strength of the "fault tree" concept is that it provides an approach to organize speculations about accidental courses of events, and displays data monitoring requirements so one can watch for the initiation of an accidental events sequence pathway.
For accident investigators, the perception provides valuable guidance for identifying and organizing data generated by the accident and recovered by the investigator. It forces the investigator to discipline the search for data by arraying it into logical branched pathways leading to the 'top event." In this way, it discourages irrelevant hypotheses about the accident being investigated, reducing controversy among investigators. Entries on the "tree" can be tested for their sequential logic along a pathway as soon as they are entered on the tree.
However, the perception has important weaknesses that have created serious problems for investigators and safety analyses based on the resultant methodologies. No criteria exist for identifying and selecting the undesired "top" event or events to be charted. This means that the beginning and end of the phenomenon investigated are again left to the judgment of the individual investigator, and the purpose and scope of this investigation are confined to the "top" event and predecessor pathways.
Most importantly, the methodologies as practiced do not accommodate the timing or duration of interactions during an accident. Nor do they provide a way to trace systematically the behavior of each actor involved in determining the outcome of the phenomenon and the relationship of that actor's actions to the actions of other actors during the accident. As practiced, the "fault trees" are not immune to equivocation logic fallacies, such as the "less than adequate" (LTA) in the MORT investigative "fault tree" check list.
MuItilinear Events Sequence Perception. A major strength of the multilinear events sequences perception and the method for arraying data it generated is that they facilitate discovery by the way they compel the structuring of the data into logical arrays. It encourages efficiency during investigations, because data must be fit into the accident data array as soon as the data become available, thus moving the analysis function to the point of origin of the investigator's data sources. The perception also demands an accounting of the actions of each involved actor during the entire duration of the accident sequences.
Probably the most valuable feature of the perception and the resultant investigative methodology is the ability to display - simultaneously - the RELATIVE TIMING of the actions by each actor - animate or inanimate - that can be identified during the investigation. This capability generates insights into accident phenomena that are unobtainable under any of the other perceptions. The method for arraying the events in parallel, proceed/follow, spatially -and time-disciplined events sequences for each actor helps investigators identify and TEST hypotheses about the nature of the accident AS DATA BECOME AVAILABLE. If data are not available, logic tree charting methods can be used to organize speculations by using the latest know event and working backward in time until a predecessor event is reached in one of the data strings.
Multilinear events sequences displays also provide a rational method for selecting candidate countermeasures that would reduce risks. The principle of intervening between events in a sequence, employed extensively by Heinrich , is supplemented by other principles for controlling risks, such as reordering the time relationships between events, or retiming relationships among events by different actors engaged in the activity. The range of countermeasure choices that can be identified using this method far exceeds the range of choices identified using other approaches. By working with events sets, estimates of accident probabilities during the conduct of an activity are possible. The methodology resembles PERT-charting techniques by dealing with proceed/follow logic tests during the data gathering steps in an investigation. It differs in its handling of events-sets time displays.
An important benefit of the method is that it easily leads to generalized, orderly models of safety problems in accidents. Another benefit is that the methodology provides a way of relating accidents to activities, thus facilitating instruction of participants.
A possible weakness is the perceived complexity of the MES perception and resultant methodologies, which discourages their use. A second concern is the present lack of formal criteria for ending the "break-down-events" procedure. A third concern is possible misrepresentation that the perception is being employed by displaying events in parallel, without cross-linking and timing the events "tracks" in the displays.
Experience with the methodologies suggests that the degree of improvement over other methodologies at this time will provide "breathing room" to formalize the "break-down" end point procedures.
Implications For Investigators
The above discussion barely begins to describe the variety of investigative practices attributable to the perceptual differences. The author has found at least 7 differing investigative processes, 44 reasons for investigating accidents, 6 fundamentally different methodological approaches, 3 distinctive types of investigative outputs, and - significantly - a total absence of criteria for establishing the beginning or end of the phenomenon being investigated(26)
Probably the most significant implication of the differing perceptions of the nature of the accident phenomenon and resultant methodological differences for investigators is that each investigator develops a personalized investigative methodology. These personalized investigative methodologies are not likely to result in replicable outputs 1) by an investigator in successive investigations, or 2) among different investigators working on the same accident. The effects of outputs that are unique to the investigator include controversy, conflict, and questionable effectiveness of the outputs in advancing "safety."
To overcome this basic deficiency in investigative methodological disciplines, adversary techniques have been introduced into some investigative processes. The investigation organization discussion in a Department of Transportation textbook provides an excellent description of this technique. The adversary process provides a forum for participants to air their viewpoints during an investigation, and motivates competing participants to bring out arguments and evidence that best serve their perceived interests. Logic weighs heavily in the process, but unless the process is augmented by adequate technical "tests", differences among the participants may be resolved by debating skills, rather than the strength of scientifically supportable technical arguments. The process can also end with incomplete understanding of the accident if it is not used in conjunction with proper technical methodologies.
A second area of implications for accident investigators is the difficulty of linking investigations to predicted safety performance of an activity. Logic trees and MES Charts both provide promising approaches to overcoming this difficulty, but until they are made available to investigators routinely, no way for investigators to demonstrate links between predictions and performance are available. The difficulties being experienced by OSHA and the Department of Transportation in its highway safety evaluation programs are clear examples of this problem for accident investigators.
A third area of implications for accident investigators relates to the qualifications of investigators. As long as .there is no scientifically derived and supported investigative methodology for accident investigators, anyone from any disciplinary background can hold himself or herself out to be an investigator. The career ramifications for "safety professionals" are obvious; the need for development of an investigative discipline becomes equally apparent when one considers the diversity of existing perceptions .
Given these circumstances, what can be done? Two steps seem to be indicated. First, a requirement to explicitly identify the investigator's perception of the accident phenomenon which forms the basis for each investigative work product is imperative. Secondly, a specific undertaking within the accident investigation community to assess the effectiveness and utility of present perceptions and their resultant methodologies is long overdue; some measure for achievements flowing from different perceptions would be immensely helpful for evaluating relative utility.
Professional safety societies such as the System Safety Society seem to be logical places to initiate such steps. One concrete action by the SSS could be to require papers submitted to HAZARD PREVENTION to carry a statement asserting the author's perception of "accident." For example, this author uses the Multilinear Events Sequences perception in his work; that perception formed the basis for development of the criteria used in this paper.
The author would welcome readers' critiques and suggestions.
 Frenze, R.J.. "The Process of Hazard Control", Kendall/Hunt Publishing Co., Dubuque, IA 1978
 "Accident Investigation and Reporting, Instruction 5102.1" Chief of Naval Operations, Department of Defense, 18 May 1977.
 Kuhlman, R., "Accident Investigation", International Loss Control Institute, Loganville, Ga., 1977.
 "International List of Diseases and Causes of Death" Eighth Revision, 1968.
 . Heinrich, H.W., "Industrial Accident Prevention", McGraw Hill, New York, 1936.
 Prosser, W.L., Wade, J.W. & Schwartz, V.E., "Cases and Materials on Torts"; (see Model Jury Instructions, p 359), The Foundation Press, Inc., Mineola, NY 1976.
. Thorndyke, R.L., "The Human Factor In Accidents", a project report for the US Air Force with special reference to Aircraft Accidents, reprinted by the US Dept. of Health, Education and Welfare, Public Health Service, Washington, D.C., 1951.
 "Fault Tree Analysis as an Aid to Improved Performance", AMC Safety Digest, US Army Material Command, May 1971.
 "Fault Tree Analysis as an Aid to Improved Performance", AMC Safety Digest, US Army Material Command, May 1971.
 "Fault Tree for Safety", D-6-75133, The Boeing Company, Seattle, WA, 1966.
 . US Civil Aeronautics Board, "Aircraft Accident Report SA 361, United Airlines, Inc. DC-8 N 8013 U, and Transworld Airlines, Inc. Constellation 1049A N 6907 C, near Staten Island, NY, December 16, 1960", Published 1962.
 Benner, L., ''Accident theory and Accident Investigation", Proceedings of the Society of Air Safety Investigators Annual Seminar, Ottawa, Canada, October 1975.
 Benner, L., "Four Accident Investigation Games" Lufred Industries, Inc. 1979.
 Haddon, W., Jr., Suchman, E.A., and Klein, D., "Accident Research Methods and Approaches", Harper & Row, NY 1964.
 Aetna Life and Casualty Co., "Accident Investigation", Catalog No.426482, Form c-3383, 8-76.
 Pask, G., "Learning Behavior in Nerve, Brain And Memory Models", in Progress In Brain Research (Vol 2), Elsevier Publishing Co., Amsterdam 1963.
 US Department of Transportation, National Highway Traffic Safety Administration, "Fatal Accident Reporting System (1979)", HS Form 214(Rev 10/78), Form Approved O.M.B. No. 004R-5634.
 Leplat, J., "Reconstruction and genesis of accidents -advantages and difficulties", Research on Occupational Accident, Swedish Work Environment Fund, Stockholm, 1976.
 Haddon, W., Jr., "The Changing Approach to Epidemiology, Prevention and Amelioration of Trauma: the Transition to Approaches Etiologically Rather Than Descriptively Based", American Journal of Public Health 58:8,1968.
 "US Standard Method of Recording Basic Accident Facts Relating to the Nature and Occurrence of Accidents", Z-16.2-1962 et seq., American National Standards Institute, New York 1962.
 King, K., "Feasibility of Securing Research - Defining Accident Statistics", National Institute for Occupational Safety And Health Publication No. 78-180, 1978.
 US National Transportation Safety Board, "Improving Survival In Hazardous Materials Accidents", Report HZM 79-4,1979.
 Benner, L., "Hypothesis Generation in Rare Events Research", Workshop on Rare Events Research Methodologies, National Bureau of Standards Gaithersburg, MD, 1976 (NBS Special Publication 482).
 Mucciardi, A.N., Orr, E.C., & Chang, J.K., "Highway Safety Programs Effectiveness Model", Final Technical Report; Adaptronics, Inc., McLean, Va., 1977.
 US Atomic Energy Commission, "Reactor Safety Study", WASH 1400, August 1974.
 Benner, L., "Understanding Accidents: A Case For New Perceptions and Methodologies", Accident Causation, Special Publication SP-461, Society of Automotive Engineers, Inc., Warrendale, PA, February 1980.
. US Energy Research and Development Administration, "A Contractor Guide To Advance Preparation For Accident Investigation", ERDA-76-45-3, SSDC-3, NTIS, Springfield, VA, 1976.
 . Benner, L., "Accident Investigation: Multilinear Events Sequencing Methods", Journal of Safety Research, 7:2, 1975.
 Benner, L., "Four Accident Investigation Games: Simulations of the Accident Investigation Process", Lufred Industries, Inc., Oakton, VA, 1979.
 Gibson, D.A., "Herbert W. Heinrich: First engineer elected to the International Insurance Hall of Fame", Professional Safety, 25:4, April 1980.
 Benner, L., "D.E.C.I.D.E. in Hazardous Materials Emergencies", Fire Journal, 69:4, July 1975.
 Benner, L., "Hazardous Materials Emergencies" (textbook), Lufred Industries, Inc., Oakton, VA, 1976
 U.S. Department of Transportation, Transportation Safety Institute, "Aircraft Accident Investigation Procedures and Techniques", Federal Aviation Administration, Oklahoma City, OK, 1976.
 Lehner, U.C., "Georgia Fruitcake and OSHA Sticks" Wall Street Journal, April 14,1980.
 National Highway Safety Advisory Committee, letter to Secretary of Transportation, June 19, 1979.
 Lorge, B F., Personal communication about significance of accident perceptions.
These illustrations were developed for use during investigation training courses, separately from this paper, to illustrate the concepts.
About The Author
Ludwig Benner, Jr., 12101 Toreador Lane, Oakton, VA 221 24; 703-620-2270; Professional Safety Engineer, CA No.379. Mr. Benner is Chief, Hazardous Materials Division, Bureau of Technology, National Transportation Safety Board, 800 Independence Ave., Washington, D.C. 2O594, and Adjunct Professor, University of Southern California, Eastern Region. He teaches graduate level course, "Investigation of Accidents".