Quality & Quantity 34: 237–258, 2000.
© 2000 Kluwer Academic Publishers. Printed in the Netherlands.
Neural Network Used to Analyze Multiple
Perspectives Concerning Computer-Based
PÄIVI MAARIT HANNELE HÄKKINEN
University of Jyväskylä, Insitute for Educational Research, PO Box 35, FIN-40351 Jyväskylä,
Abstract. The aim of this study is to explore the possibilities of neural networks to support the
analysis and representation of the complex qualitative data in behavioral sciences. In this study
for testing the methodological possibilities we analysed data of designers’, teachers’ and students’
interpretations of the same educational software. The intentions of three designers concerning the
interaction with their own software were compared with the interpretations of three teachers’ anti-
cipations of the interaction, and with the actual learning situations of three pairs of students. The
particular kind of neural network used for the data analysis was TS-SOM (Koikkalainen, 1994),
which is a variant of a self-organizing map SOM algorithm (Kohonen, 1984). On the basis of the
results it can be concluded that the method seems to be promising to handle and visualize the data
reduction in a systemic manner without oversimplifying the complex data. Furthermore, the method
supports the researcher in ﬁnding the most essential places where to focus more detailed qualitative
analyses. The visualization tools also allow us to verify the interpretations between independent
raters, which increases the reliability of qualitative data analysis.
Key words: qualitative and quantitative data analysis, neural network, self-organizing map, computer-
based learning environments.
Qualitative research has suffered for a long time from acute difﬁculties in visu-
alizing and representing relationships among data, as well as in distinguishing
and communicating the results (Miles and Huberman, 1995). There are two main
challenges in qualitative data analysis that made us search for new tools to support
the analysis process. First, the analysis of large empirical data sets is always time-
consuming and labor-intensive. The main problems are related to perceiving the
data and to data reduction at the beginning of the analysis. A particular challenge
is to scrutinize the data derivation possibilities in their entirety. In order to per-
ceive the whole coded data, some kind of quantiﬁcation is needed (cf. Chi, 1997).
Second, in social sciences there is often a need to capture the relationships and dy-
namics between different variables within complex data in a systemic manner. This
requires tools for representing the data in a comprehensible but not oversimpliﬁed