Ulf Mertens
Universität Heidelberg
Psychologisches Institut
Hauptstrasse 4751
D69117 Heidelberg
Room: F023
email: ulf(dot)mertens(at)psychologie.uniheidelberg.de
Tel: +496221547322
Fax: +496221/547787
I don't have fixed office hours, so just write me a mail
or call in order to make an appointment.
My personal
blog:
Click here
Research interests
 Adaptive Design Optimization
Adaptive Design Optimization (ADO) describes an algorithm that allows choosing for each trial
of an experiment the stimulus or condition that is most informative in terms of model comparison.
After each gathered datum, the ADO algorithm searches for the stimulus that results in the most
pronounced differences among the different model’s predictions. Due to this optimization procedure,
the number of trials needed to choose the true model can be reduced significantly.
 Bayesian Statistics
The algorithm relies on the use of Bayesian Statistics. The parameters of the different models as well
as the respective model probabilites are updated repeatedly via MCMC/Gibbs sampling.
 Machine learning
ADO is closely related to Active Learning. In the context of machine learning, it is often the case that there
is a lot of unlabeled data available (images, handwritten digits etc.). Since labeling of those data is a
timeconsuming and expensive task, simply choosing data points at random is an inefficient way. Active Learning
however picks cases for which the current model (e.g linear classifier) has most uncertainty. This procedure
reduces the required number of labeled cases to reach a certain accuracy.
 Linear mixed models
Linear mixed models (LMMs) are a popular statistical tool to model hierarchical structures
in the data. Especially in psychology, LMMs are often used for the analysis of repeated measurements.
I am currently working on a comparison of some wellknown methods for such designs including ANOVA,
MANOVA and LMMs.
Teaching
 Übung Deskriptivstatistik
An introduction to basic descriptive statistics using the R programming language
 Übung Inferenzstatistik
This is a subsequent course where the focus lies on how to use R for inferential
statistics (linear models, contrast analysis etc.)
 Statistics Toolbox
In this class, we show how to tackle some common problems in empirical theses such as
dealing with violations of assumptions, missing values, marginally significant results
and so on.
 Data visualization with R
This class focuses on how to create publicationquality plots with both base R as well
as the ggplot2package.
