Hindsight and AF447


AF A330-200 F-GZCP (Image Source: P. Kierzkowski)

Knowing the outcome of an accident flight does not ‘explain’ the accident

Hindsight bias and it’s mutually reinforcing cognitive cousin the just world hypothesis are traditional parts of public comment on a major air accident investigation when pilot error is revealed as a causal factor. The public comment in various forum after the release of the BEA’s precis on AF447 is no exception.

This post is part of the Airbus aircraft family and system safety thread.

But just knowing that the crew should have done things differently does not equate to the crew’s behaviour being culpable or egregious nor does it provide much explanatory power. Remember that at the time the crew would have had no idea of the outcome, would not have all the details which later investigation revealed and certainly did not understand how all the causal factors were interacting. They were simply responding to the situation as they perceived it at that instant.

The real question to ask is to ask why would an experienced crew, knowing only what the accident crew knew, respond in a similar fashion. To answer this question a 2004 NASA Ames study reviewed the 19 major US accidents attributed by the NTSB to human error from 1990 to 2000.

The first part of what NASA found was that the error patterns clustered into six separate but overlapping areas that correlated to the type of situation facing the crew:

  1. Inadvertent slips and oversights while performing highly practiced tasks under normal conditions
  2. Inadvertent slips and oversights while performing highly practiced tasks under challenging conditions
  3. Inadequate execution of non-normal procedures under challenging conditions
  4. Inadequate response to rare situations for which pilots are not trained
  5. Judgment in ambiguous situations
  6. Deviation from explicit guidance or SOP

Clearly the third, fourth and sixth clusters have significant relevance to the AF447 accident.

NASA’s study found that 4 out of the 19 accidents involved attempted upset recoveries from stalls, spiral dives and windshear, all of which are non-normal procedures that when coupled with challenging circumstances (night, weather or low altitude) can exceed the aircrews ability to cope.

The NASA study also quoted a Veridian report that found many pilots still had trouble recovering even with training and conjectured that this was due to the difference between the training environment of ‘OK we’ll do stalls after lunch…’, and the ‘holy crap..’ environment of experiencing an unannounced event in the real world at one o’clock in the morning.

…training in upset recovery is not sufficient to enable the majority of pilots to deal with all situations they might encounter…

Key Dismukes, Chief Scientist for Human Factors, NASA Ames

(as quoted in Aviation Week & Space Technology, 23 Aug 2002)

So, in the case of AF 447 having got themselves into a unusual attitude condition, how well would we really expect the crew to perform in recovering the aircraft, especially when they were lacking reliable air data?

Similarly having entered into a high alpha low airspeed state outside the flight protection envelope (and for which they have not trained) do we expect that the aircrew would have remembered that applying thrust would exacerbate the pitch up given the low authority of the elevator?

The last cluster involved deliberate deviation from guidelines or SOP, in the case of AF447 the crew recognised a loss of air data but failed to execute the memory items for that incident. But was this wilful? Or was there something deeper going on?

A BEA study of aircrew responses to unreliable airspeed events found that in all the events the memory items were not applied. In other words any other crew would have done exactly as the crew of AF 447 did. In the other incidents this response worked, in the case of AF447 it did not.

The NASA study also identified what they called cross cutting factors that contributed to the likelihood of error:

  1. the situation required a quick response (nearly 2/3 of the events),
  2. the situation required the management of concurrent tasks (the great majority of events),
  3. equipment failures and design flaws (approximately 2/3 the events),
  4. stress (hard to quantify but an expected physiological response),
  5. shortcoming in training or guidance (greater than 1/3 of events),
  6. plan continuation bias, and
  7. organisational or cultural issues.

For AF447 the loss of air data, poor salience of the auto-thrust disconnect and elevator trim, lack of differentiation between stall approach versus stalled state in aural warnings, stall warnings indication loss and elevator trimming nose up (2) could all be seen as design faults that undercut crew actions and added uncertainty to their decision making.

Yet another performance shaping factor would be the stress experienced by the crew and its effects on their cognitive abilities. We can confidently expect that in a high stress and surprise situation they would have experienced narrowing of attention and a loss of working memory capacity required to carry out even skill based tasks. In these circumstances would it be unusual that they overlook the automatic trim moving to a nose up position?

And of course when the traditional FAA/JAA generic approach for stall recovery training is to maintain the nose high and apply power, in the context of recovering from an incipient stall at the approach point (1), would we really expect a crew to recognise the aircraft was stalled and respond with an instinctive push down on the controller having entered into a stall? Really?

As the authors of the NASA study recommend, we should recognise AF447 as a system accident, get past ‘blaming the aircrew’ for errors in hindisght and focus on the opportunity to identify and reduce the vulnerabilities of the system.


Dismukes, R.K., Berman, B., & Loukopoulos, L. D., The limits of expertise: rethinking pilot error and the causes of airline accidents. Presented at the 2006 Crew Resource Management Human Factors Conference, Denver, Colorado, 2006.


1.  The traditional (FAA sanctified) generic approach to stall training has consisted in a controlled deceleration to the Stall Warning, followed by a power recovery with minimum altitude loss. The industry now recognises however that if the pilot tries to prevent altitude loss by maintaining back pressure on controller while going to TOGA this will result in a stall. In other words this is a stall avoidance manouevre rather than a full stall recovery procedure.

2.  A stabiliser trim setting nose up, as if for slow-speed flight, reduces the nose-down pitch elevator authority.