In his book “Human Error”?, Professor Reason discusses two types of human error: variable and constant errors.
It can be seen in the below figure that variable errors in (A) are random in nature, whereas the constant errors in (B) follow some kind of consistent, systematic (yet erroneous) pattern.
The implication is that constant errors may be predicted and therefore controlled, whereas variable errors cannot be predicted and are much harder to deal with. If we know enough about the nature of the task, the environment it is performed in, the mechanisms governing performance, and the nature of the individual, we have a greater chance of predicting an error.
Target patterns of 10 shots fired by two riflemen. Rifleman A’s pattern exhibits no constant error, but large variable errors; rifleman B’s pattern exhibits a large constant error but small variable errors. The latter would, potentially, be easier to predict and to correct (e.g. by correctly aligning the rifle sight).
However, it is rare to have enough information to permit accurate predictions; we can generally only predict along the lines of “fatigued pilots are more likely to make errors than alert pilots”?, or “The SOPs for task X on aircraft type Y is known as being ambiguous and likely to result in pilot error”. It is possible to refine these predictions with more information (e.g. The SOPs in Operator Z’s QRH are known as being ambiguous), but there will always be random errors or elements which cannot be predicted.