BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Flexible Density-on-Scalar Regression Models in Bayes Hilbert Spac
 es
DTSTART:20201204T141500
DTEND:20201204T153000
DTSTAMP:20260509T142055Z
UID:3137f4eeba47a42a43735c092687e9e461197d5d3e99fd95a622ae9f
CATEGORIES:Conferences - Seminars
DESCRIPTION:Sonja Greven\, Humboldt-Universität zu Berlin\n\nWe introduce
  functional additive regression models with probability density functions
  as response variables and scalar covariates. Existing functional additiv
 e models are not directly applicable\, as the non-negativity and integra
 tion to one constraints of densities are not preserved under summation an
 d scalar multiplication.\n\nWe thus formulate the regression model in a 
 Bayes Hilbert space with respect to an arbitrary measure. This enables us 
 to not only consider continuous densities\, but also discrete and mixed 
 densities. Estimation is based on a gradient boosting algorithm that allo
 ws for a variety of flexible effects.\n\nWe apply our framework to a mot
 ivating data set from the German Socio-Economic Panel Study (SOEP). We an
 alyze densities of the woman’s share in a couple’s total labor incom
 e\, including covariate effects for year\, federal state and age of the y
 oungest child. We show how to handle the challenge of mixed densities wi
 thin our framework\, as the income share is a continuous variable with di
 screte point masses at zero and one for single-income couples.\n \nThis 
 is joint work with Eva-Maria Maier\, Almond Stöcker and Bernd Fitzenberg
 er
LOCATION:zoom https://epfl.zoom.us/j/85710428400
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR