Models are representations of reality. They can be detailed or abstract, complex or simple, accurate or misleading. Whether we realize it or not, everything we perceive is processed using models. Therefore, it is important for us to understand how models can help us to understand reality, yet may also mislead us if not used with appropriate care and attention.
Models are used extensively in root cause analysis. Probably the most fundamental of these is the model of causation. There are models based on manipulability, probability, counterfactual logic, etc. This is an area of considerable complexity, as no single model seems to address all possible situations.
The counterfactual logic model of causation is used most often in root cause analysis, as it is the easiest to grasp and is generally the most useful. It is the model that gives us the Necessary and Sufficient test, and for this alone, its usefulness to the investigator or analyst is boundless. However, even this model fails under certain circumstances.
Consider the statement "smoking causes cancer" -- can this statement be proven (or disproved) using the necessary and sufficient test? Not really. However, despite its difficulties in certain areas, the counterfactual logic model of causation is sufficient in the overwhelming majority of cases. This is because it:
- easily guides our thought processes in a predictable way,
- provides rules that can be applied unambiguously and repeatably,
- helps us ensure completeness in causal reasoning, and
- becomes unworkable in those special cases where it does not provide good answers.
This last point might initially seem to be a disadvantage. How can a model that becomes unworkable ever be beneficial? Consider it this way -- what if we used an alternate model that happily gave us answers, well outside it's range of applicability? We might very well continue using the model without realizing that it no longer applied. Even worse, we would believe the wrong answers that the model helped us find!
What other types of models do we employ in root cause analysis? In some cases, we may develop engineering models for physical processes, in order to understand how a failure occurred. In others, we might model an industrial processes to show where bottlenecks are constraining throughput. These types of models are used quite frequently, and generally require specialized knowledge to use properly. However, the difficulty of developing and using such models may actually pale in comparison to the modeling of human behaviour.
We need models of human behaviour because humans are so incredibly complicated. Such models must account for information input and processing, communication, motivation, learning, decision, fatigue... the list goes on and on. Then, on top of models for individual human behaviour, we must add models for group, organizational, and societal behaviour and interaction. The problem seems intractable. Nonetheless, several generalized models do exist.
One step above the models of human and organizational behaviour are models of accident initiation and propagation. The driver for research interest in this area is obvious, as industrial accidents are potentially the most damaging events that can occur. Death and destruction, possibly on a large-scale, are the consequences. It is hoped that by understanding how accidents occur, we can find strategies to reduce the risk of such events.
Accident models, in fact, tend to be models of human and organizational behaviour. What makes accident models different is the sharp focus on failure propagation. The underlying assumption tends to be that accidents start as relatively simple, minor events that eventually spiral out of control. In fact, most recently developed accident models tend to be system models that focus attention on complex interactions between multiple, lower-level failures or infractions.
In the end, we are left with models upon models upon models... each with their own rules and assumptions, strengths and weaknesses. As stated previously, models are useful because they help us abstract away unimportant data so we can increase our focus on useful information. This is the strength of using models; unfortunately, it is also the main weakness. If models are used without knowledge of their assumptions and limitations, we could end up discounting potentially important facts and misdirecting our investigations.
There is no single "model of everything" we can rely upon to provide good answers in all cases. However, we shouldn't be fooled into thinking that the various models can't help us achieve better root cause analysis results. Models can guide us to possibilities we might have missed, and provide insights that we might not have seen. The key success strategy may well be to have knowledge of a wide variety of models that can be used in a variety of situations. Then, as with anything else in life, we must simply ensure that we understand the tools we use, before we use them.
by Bill Wilson