Summary VENI-Project
"Modelling interaction effects as small trees in regression and classification".
In the behavioural sciences, interaction effects are investigated in many theoretical models and empirical
studies to explain and predict human behaviour, emotions, quality of life, and so forth. Interaction effects are often
called moderator effects. A moderator is a predictor variable that affects the direction and/or strength of the
relationship between another predictor variable and a criterion variable. Moderator effects are almost universally
represented as cross-products of the variables that enter the interaction, forming a product interaction. However,
the product interaction has several important problems. Therefore, Dusseldorp and Meulman (2001) introduced a
new data analysis strategy, the so-called Regression Trunk Approach (RTA). This approach makes it possible to
estimate a radically new type of regression models, namely models including threshold interactions. The
preliminary results are promising. Several important aspects of the approach, however, need further development.
The aims of the proposed research are: 1) to elaborate and extend the properties of RTA into a complete analysis
strategy for interaction research; 2) to investigate in which situations, relevant to the behavioural sciences, a
threshold interaction is more appropriate, and in which situations a product interaction; and 3) to explore the
generalization of RTA to classification problems.
An underlying goal of the proposed research is to stimulate researchers in thinking about which type of
interaction is more appropriate for their problem under consideration, and in this way to contribute to the
development and testing of theories that incorporate interaction effects. The proposed research contributes to
the behavioural sciences by the development of a powerful strategy that facilitates researchers to assess
interaction effects. The research is an innovative extension of the project “Classification and Prediction
in Subject Oriented Multivariate Analysis”. This project was part of the PIONEER program
“Subject Oriented Multivariate Analysis”, awarded to Prof. J. J. Meulman.