Provided by Cognitive Sciences ePrint Archive
AARE2002: Paper CHO02101
Improving behaviour classification consistency:
a technique from biological taxonomy
Serene Hyun-Jin Choia, Timo A. Nieminenb, Mark Bahra and Nan Bahra
a Graduate School of Education, The University of Queensland, QLD 4072, Australia
[email protected], phone +617 3365 6234, fax +617 3365 7199
b Centre for Biophotonics and Laser Science, Department of Physics,
The University of Queensland, QLD 4072, Australia
[email protected], phone +617 3365 2422, fax +617 3365 1242
Abstract
Quantitative behaviour analysis requires the classification of behaviour to produce the
basic data. In practice, much of this work will be performed by multiple observers, and
maximising inter-observer consistency is of particular importance.
Another discipline where consistency in classification is vital is biological taxonomy. A
classification tool of great utility, the binary key, is designed to simplify the classification
decision process and ensure consistent identification of proper categories.
We show how this same decision-making tool - the binary key - can be used to promote
consistency in the classification of behaviour. The construction of a binary key also ensures
that the categories in which behaviour is classified are complete and non-overlapping. We
discuss the general principles of design of binary keys, and illustrate their construction and
use with a practical example from education research.
1 Introduction
Quantitative data analysis is a research tool that it would be difficult to overstate the usefulness
of. Its power and generality as a method for testing the validity of hypotheses can be seen by the
breadth and depth of its application in social, biological, and physical sciences. However, by its
very nature, quantitative data analysis requires quantitative data. This, then, is the challenge
often faced in education research - the reduction of observations to numerical data.
To analyse behaviour - for example, the frequency of, or duration of particular behaviours,
or correlations between behaviours - the observed behaviour must be identified. This is fun-
damentally a classification process. The development of the categories into which observed
behaviours are classified is widely discussed in the literature (Barlow & Hersen, 1984; Gittle-
man & Decker, 1994; Herbert & Attridge, 1975; Slater, 1978; Whitley, 2002). The categories
should be:
1. Mutually exclusive. There should be no overlap between the categories - no behaviour
should be classifiable into two separate categories.
2. Complete. The categories should form a complete or exhaustive set of the possible be-
haviours. It must be possible to classify every observed behaviour into a category. This
does not mean that a large number of categories is required - a small number of suffi-
ciently broad categories can be complete.
3. Usable. The categories must be understandable - terms used must be clearly understand-
able and well-defined. Definitions should be concisely and clearly stated. The names