• en
  • ru

Repeatability, detection and forecasting of events

April 13, 2013

This article does not pretend to the novelty of ideas or a new approach to safety management. It just presents some considerations and arguments that are already known but, nevertheless, are subjected to debates in the aviation world.

The main objective of the article - provide the reader with information that AVEX Bureau considers logical and useful for understanding of the safety management fundamentals.

Since ancient times, it is known that any event will not be repeated in all details. For example, the phrase is attributed to Heraclitus that "it is not possible to enter twice into the same river" - everything flows, everything is changing.

Formally speaking, the likelihood of recurrence of any event in the smallest details ("precise event") is equal to zero.

This is quite obvious when you consider the infinite number of possible details of any precise event.

Furthermore, these details are randomly allocated in time and space. Zero probability in this case is the result of dividing the infinitesimal to the infinitely high. This is an absolute zero.

Impossibility of repeating of precise events deprives them of any cognitive value, as events with zero repeatability prevent realization of the fundamental property of intelligent life - the ability to make a probabilistic choice. The mind is incompatible with absolute precision.

For this reason, in everyday life, we are only dealing with inaccurate events that are perceived without unnecessary details. This type of events is somewhat "blurred", but this disadvantage is compensated by their main property - repeatability. You could even say that blurring - is a payment for recurrence of events.

From a formal point of view, to get an inaccurate event it is necessary to "cut infinite tail" of unnecessary details from a precise event, so that the inaccurate event could be adequately recognized.

The practical difficulty is that before you cut, it is necessary to think twice - events with a “long tail" will be too rare, and with a “short tail” will be too blurred.

And again, recurrence of events is helpful in this issue - experience will tell what details in the event description would be the most useful.

In addition to the probability of occurrence for an inaccurate event there is a probability of its recognition.

Let us illustrate this with an example. Suppose among the thousands of coins lying on the table, it is necessary to choose some specific coins. Mandatory selection criteria are:

  1. leave those coins that lie obverse up, then
  2. leave those coins that were released not later specific year, then
  3. leave coins with a small scratches, then
  4. select coins that have been inspected for scratches at the first second of each minute.

It is obvious that there exist some probability of positive selection. If your selection was somewhat negligent, the probability of positive selection will change - among the selected coins there will be some "wrong" coins or some “right” coins will be missing.

Thus, there is a probability of recognition, which may vary over a wide range and will affect the subjective probability of recurrence of inaccurate events.

The very specific type of events is formed by so-called "consequences".

Events of this type seem to be the most accurate, despite the very short "tail”. For example, as a result of some action, a cup has been broken, which had a certain cost and belonged to someone.

What could be more precise? But it is just seeming accuracy. In fact, on the basis of this result, it is not possible to say how event will develop in the future.

It is convenient to assume that the actual inaccuracy of consequences is compensated by the uncertainty of future events.

It looks as if the events of the past are projected into the future through the "aperture of consequences" - the larger the aperture (more details in the description of the consequences), the smaller the depth of field (smaller variety of possible scenarios of future event).

In the extreme case, if the consequence is a precise event (described with the utmost precision), there is only one scenario of events - fate.

Arguing thus, we are coming to event forecasting, but first let's sum up above discussion:

  1. Any recurring event differs in the details.
  2. Detailed elaboration of events reduces their recurrence.
  3. Recurrence of events and probability of their recognition are interrelated.

Speaking about forecasting, it should be recalled that only inaccurate events may be forecasted. However, the forecast can be accurate. For example, with reasonable certainty it can be predicted that "each time when any trouble will happen, there always will be someone who knew that the trouble will happen exactly as it has happened" (Evans and Bjorn).

Along with that, the forecast accuracy is not its main characteristic. From a practical point of view, it is more reasonable to estimate the "utility" of forecast, i.e. the potential effect of a forecast on the future course of events. The utility of forecast is determined by the following characteristics:

  • timeliness;
  • significance;
  • practicality;
  • accuracy.

Untimely forecasts (delayed or premature) have little or no value. Forecast of events that is not considered significant, is of little value. If prevention of forecasted event requires resources which are not available, the practicality of the forecast will be low. And finally, the inaccurate forecast may harm, i.e. have a negative value.

In addition, for accurate forecasts there is an unpleasant surprise - if reasonable measures were taken in relation to the forecast, the prognosis will not be realized, and may give the impression that the forecast was wrong. There's a choice - either to do nothing and verify the prediction, or do something, and to doubt forecast accuracy. This is somewhat like "uncertainty principle".

Fortunately, for frequently recurring events you can compare the frequency of unwanted events before and after the implementation of reasonable actions.

But the forecast of rare events (severe consequences) will always be questioned, especially since it is never accurate - no one can say the flight and exact date when the plane will overran the runway.

But let’s return to the value of a forecast. Forecast of what events can be maximally useful?

From the most general considerations, the cognitive value of the observing inaccurate events depends on frequency of their occurrence - it tends to zero for very rare or too frequent events.

For example, if an event occurs very often in airline, the event does not bring significant harm and it would be unwise to pay too much attention to it. Forecast of such events may be characterized as timely, accurate, but not significant.

Extremely rare events are not a major source of information, as well. As a rule, they are considered retrospectively and are aimed at finding the causes and circumstances of the event. Large-scale measures that are taken after each such case lose their value over time and by the moment of the next similar event they become totally obsolete, bringing more harm than good. Forecast of such events may be characterized as not timely, not accurate and not practical.

As the object of forecasting and management there remain events with medium recurrence level, the effects of which are perceptible, but not critical to the airline. Such events are well recognizable and forecast may be characterized as timely, significant, practical and accurate to the maximum degree.

Intermediate conclusion for event forecasting:

  1. Forecast utility depends on its timeliness, significance, practicality and accuracy.
  2. Inaccurate forecast could be harmful.
  3. Forecast of rare events is always questioned.
  4. Forecasting of too rare or too frequent events does not have practical sense.

Finally, it remains to discuss one important question - is it possible to prevent rare events (that cause severe consequences), at the expense of management of middle-level consequences?

This question can be formulated otherwise –are there stable causal relationships between events of different frequency? The answer is the following - the model of causal relationships is widely used in the investigation of accidents, it is considered obvious and confirmed by experience.

In terminology used in this article, it means that an infinite "tail of details" of any event can be cut into pieces (clusters) which are sufficiently stable and can be used for constructing of events.

With this approach, any sufficiently rare event may be represented as a chain of interconnected more frequent events. Interruption of such chains destroys the causal connection and prevents the event of a higher level.

Intermediate conclusion for prevention of significant consequences:

  1. There exists short enough event-clusters, which can be formed and used in forecasting of rare events.
  2. Prevention of events with middle-level consequences breaks the causal connection to severe consequences (rare events).

Throughout the article much has been said about the uniqueness and repeatability of events, the possibility of their recognizing and prevention. However, to complete the article let’s say that stability is good because it happens to be broken, leading to the need of large and small changes.

Tel: +380930322525
Email: avex@avex.pro