We have a new website go to gov.scot

Paper 25th May 2004 - LGRAS(04)_presentation

Taxonomies and Categorisation - The implications for safety statistics

Taken from a PowerPoint presentation.

Everything is taxonomical

  • Any categorisation of anything is part of a taxonomy.
  • For example: Red, Orange, Yellow, Blue, Indigo and Violet are a taxonomic categorisation of colours.
  • Notice the categories have to be Exhaustive and mutually Exclusive or the taxonomy would be useless.

Databases and Reliability

  • Any method of categorising data is to put it into a database (be it electronic or otherwise). E.g. a library.
  • The key question is: can two or more people classify the same event/object/situation under the same taxonomic categorisation (e.g. colours)?

Traffic Taxonomies

  • The current taxonomy was developed in a Scottish police force and was tested via a reliability trial.
  • There were six classifiers given the SAME event to code/classify: one code per event.
  • Reliability was calculated by comparing each classifier with each other classifier.

Results (percentages)

  • 1\2 41.6
  • 1\3 45.2
  • 1\4 41.9
  • 1\5 58.1
  • 1\6 48.4
  • 2\3 61.3
  • 2\4 41.9
  • 2\5 45.2
  • 2\6 45.8
  • 3\4 41.9
  • 3\5 35.5
  • 3\6 32
  • 4\5 32.2
  • 4\6 25.8
  • 5\6 32.3

Total Reliability or Consensus

  • 43.2%
  • Examples of lack of consensus
  • Coder X: Unsafe Right Turn
  • Coder Y: Visibility
  • Coder X: Disobeying Sign
  • Coder Y: Side Road Etc etc. etc. etc. etc……


  • Noise in database will rise and rise until database is effectively worthless.
  • NOW: important to remember these were experienced classifiers/coders
  • This is NOT a training problem.
  • What is the problem then?
  • The problem is the taxonomy itself.
  • Look at the codes: you may well ask yourself: I don't see a taxonomy here! This is just a list of 'causation codes'!
  • Well, exactly.

Taxonomic Theory

  • Taxonomies must be MEE
  • Mutually Exclusive and Exhaustive.
  • In the list shown, there are 'confounds'.
  • For example: to 'change lane without ascertaining that the road is clear' (10) can clearly be the same as 'overtaking improperly' (11)

Other Examples

  • 'Mechanical Fault' (37)
  • 'Tyre Defective' (38)
  • But a defective tyre IS a mechanical fault surely…..
  • And so on.

Nested Hierarchy

Ideal taxonomies are nested hierarchies

Classifiers move 'through' them via AND/OR gates.

Representation of and/or gates

A Taxonomy from Nature.

A taxonomy example from nature