PPRuNe Forums - View Single Post - TAM A320 crash at Congonhas, Brazil
View Single Post
Old 16th Sep 2007, 07:06
  #2288 (permalink)  
PBL
 
Join Date: Sep 2000
Location: Bielefeld, Germany
Posts: 955
Likes: 0
Received 0 Likes on 0 Posts
Some sane words from alf once again.

Originally Posted by alf5071h
Can the ‘Why–Because’ group map their thoughts to SHELL?
Yes, more or less.

The SHELL "model", like Hopkins's Accimaps, is an attempt to classify factors into broad groups. Unlike Accimaps, there is some model of "how things happen" behind it. For pure classification, all one has to do is classify the factors in the WB-Graph, and we will shortly have SW that will display classifications. We have been aware of the value of classification for some explanations ever since Ernesto de Stefano's WBA of the Ladbroke Grove rail accident in 2002, at that time the largest WBG that had been put together (somewhat over 90 factors). But unfortunately the layout engine we prefer did not until recently have a layout for classified graphs that we liked along with an interface we could use.

In so far as a given "model" forces a specific architecture for "how accidents happen" ("the X-bone's connected to the Y-bone"-type explanations), then WBA remains neutral, because the Counterfactual Test is a model-neutral test for causality. A study in 2003 by our collaborators at the rail institude in Braunschweig compared a WBA of a rail accident with a model-based analysis, and showed that about one third of the factors in the model-based analysis did not actually occur in the WBG. See Brinkmann/Lemke at
http://www.tu-braunschweig.de/ifev/v...bieleschweig2/
In other words, these factors were introduced by pure theory, rather than being necessary causal factors in the causal net.

I also performed a comparison in 2005 of Hopkins's analysis of the Glenbrook rail accident in NSW with a WBA, with the graphs reproduced more readably in
http://http://www.rvs.uni-bielefeld....ookFigures.pdf
I am somewhat wary of the "extra baggage" that using a specific model brings with it, in so far as it goes beyond a certain purposeful classification of factors, and I prefer to allow, even aid, but not force, such classifications on top of the pure causality as determined by the Counterfactual Test.

PBL
PBL is offline