Model choice and goodness-of-fit

Workgroup "Model choice and goodness-of-fit"

Head of group: Dr. Justin Chown


  • Dr. Axel Bücher
  • Prof. Dr. Holger Dette
  • Dr. Michael Hoffmann


In real data sets it is usually unknown how the data were generated. The task of the statistician is to fit a model to the data which reflects their main characteristics best and allows accurate predictions. A model selection strategy supports the statistician in his decision for a well fitting model before the main analysis of the data. A typical problem which is solved by model selection is the identification of relevant covariables in regression models. For example it is interesting to identify the relevant genes which influence the phenotype of some biological attribute. Using model selection strategies it is possible to identify those genes and in the following their influence on the phenotype can be described.
By goodness-of-fit-tests one compares how well different statistical methods fit the data. For example if the relevant genes are identified one could use a linear or a non-parametric curve to predict the phenotype if a genetic profile is given. Which procedure is advantageous can be evaluated using a test for parametric form.
The work group "Modellwahl und Goodness-of-fit" develop and investigate both more classical model selection strategies and modern approaches like penalized regression. Further the work group develops tests for comparing regression curves.