Datalab Seminar, Lucas Kook: "Transformation Models - an introduction" (Kopie 1)
At Thursday, 10th October from 12:00 to 13:00, we have a very interesting talk from Lucas Kook (University Zurich)- The event takes place in the room TH 541.
Title: "Transformation Models - an introduction"
Description: Traditionally, building a regression model is an iterative process. Data scientists start with a simple model and assess its shortcomings by residual analysis. Oftentimes response and covariables need to be transformed to achieve a well performing model. Here I will present a novel approach called "most likely transformations" (mlt) in the context of transformation models. Instead of predefining any transformation of the response, the most likely transformation is estimated from data by maximizing the likelihood. Many well-known models fall under the flexible class of transformation models. Examples include normal linear regression, the Weibull and Cox model for survival responses and logistic regression. This BBS will give a brief introduction to the statistical theory underlying transformation models with a strong focus on applications in the statistical computing environment R. In particular, normal linear regression will be illuminated from a transformation model perspective, paving the way for understanding the more intricate class of conditional transformation models