Title: Detecting Shape Heterogeneity in Penalized Regression Splines

Author: Marjolein Fokkema

Abstract:
Penalized regression splines (a.k.a. smoothing splines) are a popular and
powerful tool for non-linear regression modeling. They use a rich spline
basis, with which almost any shape of association between predictor(s) and
response can be approximated, and avoid overfitting through a penalty on the
weights of the spline basis functions. Our aim is to test whether the shape of
smoothing splines differs as a function of covariates. We take advantage of
the equivalence of mixed-effects and smoothing spline models (Wood, 2017).
Next, we employ the derivatives and score-based tests proposed by Wang and
others (2018, 2020, 2022) to test for shape stability. In this presentation, I
will results on the performance of this approach as evaluated on real and
simulated datasets.