BY

CHARLES BARBEE & DAVID DEAVER

Over the past few decades, public attention has increasingly focused on human error in major tragedies. Some experts have indicated that as true mechanical failures have been eliminated, attention naturally shifts to the human element. In fact, tragedies have always been caused by human error. In the past, however, the negative consequences of an incident were limited to immediate vicinity. As our transportation systems have become larger, moving millions of tons of cargo and hundreds of thousands of people, the consequences of a disaster are greater and reach a much bigger area. Almost everyone is affected by a major transportation disaster. As these consequences reach more people, the implications of human error are clearer. To eliminate these principle causes of incidents, human error must be understood and defended against. Accordingly, one of the most important jobs a marine investigator has is identifying the types or kinds of human errors occurring. As human error is analyzed, it may emerge that certain errors relate principally to specific preconditions (see Enclosure 2). It may also emerge that certain errors are harder or impossible to defend against. These relationships determine what safety recommendations are made and what improvements the Coast Guard undertakes.

In his 1990 book "Human Error," Dr. James Reason integrated two bodies of thought on human error into a single framework, the Generic Error Modeling System (GEMS). GEMS differs from previous error analysis tools in that it integrates the errors that occur at all levels of performance. In creating the annex to the Code for the Investigation of Marine Casualties and Incidents, the IMO adopted GEMS as the preferred tool for investigating human performance. The Coast Guard has also adopted this tool.

One of the most challenging aspects of incident investigation is describing exactly what type of errors the people involved made, and why they made them. To accomplish this, marine investigators need to have a thorough understanding of how people solve problems and perform tasks. The following information provides marine investigators with a basic understanding of how people perform tasks and solve problems by describing Rasmussen's skills, rules, and knowledge (SRK) model of human performance.

Dr. James Reason analyzes human error using the Generic Error Modeling System (GEMS), which is based on the work of his mentor, Dr. Rasmussen. Rasmussen's work explores how people solve problems and do things. It identifies three levels of performance, and suggests that the kind of human error a person makes depends on what level of performance they were engaged in at the time. In 1981, Dr. Rouse succinctly summarized a common theme in psychology: people are furious pattern matchers. As such, the human mind attempts to find patterns and select a "pre-packaged" action rather than analyze each situation and calculate the optimal solution. In 1986, drawing on this observation, Dr. Rassmusen identified three levels of human performance. They are: 1) skill-based performance (SB); 2) rule-based performance (RB); and knowledge-based performance (KB).

　

Usually human functioning is skill-based (SB). In SB performance, the task is so automatic and the surroundings so familiar that the person doesn't think about how to do it. The person merely mentally visualizes the desired state and it happens, largely without conscious monitoring. Unless we are very new to these activities, we do not have to consciously consider how to perform them, or focus attention on them. All the person simply monitors how well (and how far) the activity is progressing. When it's gone far enough, the person turns the automatic function off. Accordingly, SB performance requires that the person pay sufficient attention, at the right times, or else the automatic function may go too far.

　

Complex situations or activities cannot be solely dealt with through automatic functions. While we almost always start in SB performance, people switch up to the next level when an attention check detects a deviation from the "pre-packaged" or planned-for conditions. Usually the discrepancy is minor and correction is found early, so people return to SB performance. Our daily routines largely consist of SB performance with periodic shifts to and from the next level: rule-based performance.

　

While most people don't recognize it, they have an enormous "library" of rules for dealing with every day life, most of which they would have a hard time formulating in words. People acquire some of these rules from formal education, but most are learned from experience. When people detect that SB performance isn't working or wont work, they start trying to figure out the situation, They look for "signs or indicators. This is rule-based (RB) performance. RB performance has two important parts: (1) the perception rule (if you see sign X, then state of the world Y exists); and (2) the action rule (if state of the world Y exists, then action Z is appropriate). To select the right rule and solve the problem, a person in RB performance must select and use the right perception and action rules.

　

As stated before, people are furious pattern matchers, meaning they look at the signs and match stored rules until they find one that fits. Because situations are complex, slight differences between situations may require different solutions. Typically, people discover this by trying a rule that seems to fit, and monitoring how well it works. When one rule fails to solve the problem even when it appeared to fit the situation, people tend to look for another rule (usually a close relation like a parent-child rule). Child rules are exceptions to general (parent) rules when the general rule is mostly true, but needs some slight modification to work.

　

If we can get a rule that fits, we take the "prepackaged" answer: if you see sign X, then state of the world Y is true. If state of the world Y is true, then do Z. Often times, the action Z is simply a skill. For example, take the simple act of adjusting the temperature of a bath. If the water feels hot to the hand (sign), the water is too hot (state of the world). If the water is too hot (state of the world), turn on cold water (SB performance).

　

Although we spend most of our time in SB performance, we rarely have much conscious memory of that performance. With enormous practice and experience, rule-based activity (even very complex RB activity) can be "pushed" below the conscious threshold and become SB.

　

In trying to solve a problem, people match patterns. After trying a number of rules, some very complex situations simply won't fit the rules a person has. Basically, the situation (or problem) is completely new or novel to that person. No prepackaged answers apply, and the person must generate an answer from first principals: knowledge based performance.

　

When they don't have rules to help them, people try to use their understanding of relationships to actively predict the future (without benefit of experience to help). This is a calculated, "do the optimal best thing" kind of operation, which requires the maximum available conscious attention and thought.

　

As a rough rule of thumb, people perform best at the SB level, and very well at the RB level. People are not, however, very good KB performers. As strange as it seems, novices and experts are equally bad at solving truly new and novel problems. A new, novel situation to a junior person may be a common; hence the old adage "Good judgment comes from experience. Experience comes from bad judgment."

　

Some situations appear to be new and novel on their face, but when people analyze, calculate, and try to optimize, hidden similarities between the new problem and others can emerge, allowing the person to try to push the performance back down to the RB or even SB levels. Like the transition between SB and RB, a person may only be in KB performance for a short time before returning to the RB level.

　

Rasmussen's theory suggests that people spend most of their time executing SB tasks, somewhat less time performing RB tasks and decisions, and very little time in KB performance. This relationship can be represented as a pyramid of performance:

The first step in conducting human error analysis is to identify which act or decision to analyze from the actions list on the timeline (see Enclosure 1). During the course of a typical incident, many decisions are made and actions taken. Of those, many will prove to have been errors. Using the processes outlined in Enclosures 1 and 2, each action and decision shall be separated and organized within the model of production. In general, when there is difficulty conducting human error analysis, it is likely that several acts or decisions are being analyzed as one.

　

An Unsafe Act or Decision is a misaction or poor decision taken in the presence of a hazard.

　

Example: The decision not to wear a life jacket on a cruise ship while underway, for instance, (below decks with limited opportunities to reach the edge) is not an unsafe decision because there is no hazard. Deciding not to wear a life jacket in a small vessel while working near the edge or over the side, however, is an unsafe decision.

　

For the purposes of human error analysis, listing or analyzing all the errors or misactions a person commits serves no purpose. Instead, focus should be placed upon those errors and misactions that are Unsafe Acts or Decisions. To do so, the hazard must be clearly identified in whose presence the error or misaction was performed. If the hazard cannot be identified, then the error/misaction does not qualify as an unsafe act or decision, and does not require further attention. These hazards will typically be latent unsafe conditions (LUCs) defined in preconditions or defenses (see Enclosure 2).

　

The second step in analyzing human error requires the decision as to whether the person's error was in execution or planning of the action or decision. Execution errors involve memory and attention failures, whereas planning errors involve mistakes and violations.

　

An Execution Error is the misperformance of an intended sequence of actions (i.e., the person intends to do one thing, but does another). When a person fails to execute the plan they have in mind, an execution error has occurred.

　

If a person does something other than what they intended to do, the person must be using relatively little conscious resources, or in SB performance. If a person is not in SB performance, an execution error was not committed.

　

In some cases, it will appear that a person made an execution error, but was in RB performance. As an example: A mariner may be navigating using the navigation rules (RB-performance), panic and turn to port. In these cases there are two separate errors: (1) the RB unsafe decision that led to the mariner panic; and (2) the SB unsafe act of pushing the rudder to port instead of to starboard. When it appears that a person made an execution error, but was in RB performance, the situation should be examined further.

　

A Planning Error is a mistake or violation that results when the person executes a decision or action as they intended, but that action was inappropriate for the situation. Planning errors are the opposite of execution errors. The error lies not in how the plan was executed, but in the plan itself.

　

Because planning errors involve moderate to high amounts of conscious activity (decisions and problem solving), all planning errors occur in RB or KB performance. If the person was engaged in SB performance, the error is an execution error.

All execution errors are errors in SB performance by definition. However, the person does have to periodically interrupt the preprogrammed behaviors to check on the progress of the activity. These "attention checks" establish (1) whether the actions are running according to plan; and (2) whether the preprogrammed sequence is still adequate to achieve the desired outcome. In essence, the person checks for deviations from his or her expectations. Execution errors occur when something goes wrong with the periodic attention checks. Specifically, attention checks may go wrong because the person pays attention at the wrong time, pays insufficient attention, or completely forgets to perform the attention check In analyzing execution errors, it shall be determined whether the error involved a problem with attention or a problem with memory.