GMS Medizinische Informatik, Biometrie und Epidemiologie (Jun 2005)
Smoothing methods in biometry: a historic review
Abstract
In Germany around 25 years ago nonparametric smoothing methods have found their way into statistics and with some delay also into biometry. In the early 1980's there has been what one might call a boom in theoretical and soon after also in computational statistics. The focus was on univariate nonparametric methods for density and curve estimation. For biometry however smoothing methods became really interesting in their multivariate version. This 'change of dimensionality' is still raising open methodological questions. No wonder that the simplifying paradigm of additive regression, realized in the generalized additive models (GAM), has initiated the success story of smoothing techniques starting in the early 1990's. In parallel there have been new algorithms and important software developments, primarily in the statistical programming languages S and R. Recent developments of smoothing techniques can be found in survival analysis, longitudinal analysis, mixed models and functional data analysis, partly integrating Bayesian concepts. All new are smoothing related statistical methods in bioinformatics. In this article we aim not only at a general historical overview but also try to sketch activities in the German-speaking world. Moreover, the current situation is critically examined. Finally a large number of relevant references is given.
