NEWS

---

gamtree 0.0.3

• Function splinetree added for partitioning unpenalized or parametric spline models.
• Score-based conditional inference tests can now be used to fit smoothing-spline trees. This much reduces computational load. Function gamtree allows to choose between use of model-based recursive partitioning (method = "mob") and conditional inference trees (method = "ctree").
• Although the previous implementation of gamtree was functional, the parameter stability tests were based on the 'wrong' derivatives, leading to bad overfitting behavior. The current implementation of gamtree has more limited functionality, but uses proper derivatives for testing stability of smoothing splines. It employs gamm4 to fit the GAMs, which provides a full mixed-effects specification of GAMs. It employs merDeriv to obtain valid derivatives or scores. The function is now much slower, but should yield accurate subgroups and exhibit much less overfitting.

gamtree 0.0.2

• Many bug fixes, including:
• Offsets now correctly passed between tree and GAM, which earlier version did not, yielding no changes of improvement when joint estimation was performed.
• Improved argument passing to functions gam() and bam().
• Function gamtree now supports fitting with gam() and well as bam().
• Improved plot, coef and predict methods.

gamtree 0.0.1

• First development version, available from github.