Peeking Into The Black Box: Gaining Insight With Recursive Partitioning of GLMs

Beschreibung

Recursive partitioning algorithms separate a feature space into a set of disjoint rectangles. Typically, a constant (e.g., a mean or a pro- portion) is fitted in every segment of the partition. While this is a simple and intuitive approach, it still lacks interpretability as to how a specific relationship between dependent and independent variables may look. Or it may be that a certain model is assumed or of inter- est and there is a number of candidate variables that may nonlinearily give rise to different model parameter values. We want to present an approach that offers a solution to the problem of limited interpretabil- ity of classical trees as well as providing an explorative way to assess a candidate variable's influence on a parametric model: Model-Based Recursive Partitioning (Zeileis et.al., 2008). This method conducts recursive partitioning of a parameteric model such as the generalized linear model by (1) fitting a parametric model to a data set, (2) testing for parameter instability over a set of partitioning variables, (3) split- ting the model with respect to the variable associated with the highest instability. The outcome is a tree where each node is associated with a fitted parametric model. We will describe the procedure and show its versatility and suitability to gain additional insight into the relation- ship of dependent and independent variables by three examples, the link between professors' beauty and their teaching evaluation, the pre- diction of voting behaviour and a failure model for debt amortization.

Zeitraum	27 Juli 2010 → 31 Juli 2010
Ereignistitel	LinStat - International Conference On Trends And Perspectives In Linear Statistical Inference
Veranstaltungstyp	Keine Angaben
Bekanntheitsgrad	International