Key research line: Flexible Modelling and Generalised Regression Techniques
Description
The emphasis in this research line is on the development of statistical methodology based on flexible modelling with the aim to adequately describe (predict) the influence of one of more variables on a response variable. Flexible modelling is achieved by using non- and (semi-) parametric techniques, such as P-splines, wavelets, local polynomial approximation. In this research line the focus is not primarily on one specific area of application; the techniques are used for analysis data from medical applications, econometrics and chemometrics, among others.
The influence of a set of covariates on a respons variable can be described via the mean regression function, but also via the median function or some quantile (percentile) function. Robust statistical techniques are important here. Some covariates can also have an influence on the variability. This leads to the estimation of, e.g., variance functions, dispersion functions, covariances, etc. Flexible modelling needs to be done simultaneously for several functions (e.g. mean and dispersion/variance function). Due to the complexity of stochastic phenomena, it often happens that the functions of interest show behaviours of different degrees (from very smooth to irregular and very fluctuating) in distinct regions. Hence, the nonparametric techniques should be able to discover these different behaviours, based on just the data at hand. Sometimes one has prior information on certain qualities of the function under study (e.g. monotonicity, concavity …). Appropriate restrictions should be built in into the flexible estimation techniques to use this prior information knowledge.
PI-contactperson
Prof. Dr. Irène Gijbels
