Short courses in advanced statistics
Our individual courses cover advanced statistics on-site at ETH Zurich over three to six half-days. Pick the topics your work requires, register directly and transfer the credits to our CAS or DAS if you take one subsequently.
Upcoming courses
These courses are currently available and you can register for them by completing the form at the end of our page.
6 half-days | CHF 1,800
Dates
- 09.11.2026
- 16.11.2026
- 23.11.2026
- 30.11.2026
- 07.12.2026
- 14.12.2026
Lecturer
Participants learn to describe business and scientific problems as probabilistic models, apply Bayes’ rule to draw inferences from data and use the probabilistic programming language Stan to sample from posterior distributions. The course fits linear models for both continuous and categorical outcomes and covers techniques for handling hierarchical structures in the data. Examples from fields such as insurance, meteorology and marketing are used.
Concepts: Prior and posterior distributions · conjugacy · Markov chain Monte Carlo · hierarchical models · posterior predictive checks
5 half-days | CHF 1,500
Dates
- 16.11.2026
- 23.11.2026
- 30.11.2026
- 07.12.2026
- 14.12.2026
Lecturer
Sample surveys are a fundamental tool for collecting data from populations. The course explains the elements of a sample survey and introduces the most important classical designs – simple random sampling and stratified random sampling – and their estimators, along with estimation methods using auxiliary information and the Horvitz–Thompson estimator. Data preparation, non-response and its treatment, variance estimation and the analysis of survey data are also covered.
Concepts: Simple random, stratified and cluster sampling · design weights · non-response · variance estimation
Unscheduled courses
These courses might be offered again in the future. Register for our info mailing list to be notified when they are scheduled again.
4 half-days | CHF 1,200
Lecturer
High-dimensional data arise when the number of variables is large relative to the number of observations, complicating modelling, prediction and inference. The course focuses on analysing such data, with an emphasis on prediction problems and feature assessment. The first part covers linear regression in high-dimensional settings, overfitting, prediction and the bias–variance trade-off, model selection, and ridge and LASSO regularisation. The second covers logistic regression and regularisation, classification using decision trees, random forests and AdaBoost, multiple testing, p-value adjustment and variance shrinkage.
Course Material: external page https://staedlern.github.io/highdim_stats/
Concepts: LASSO and ridge regression · regularisation · variable selection · multiple testing and the false discovery rate
Lecturers
Modern data analysis often calls for methods that make fewer distributional assumptions. The course covers non-parametric and randomisation tests, resampling techniques such as the jackknife and the bootstrap, and asymptotic approximations of estimators.
Concepts: Rank-based tests · bootstrap · permutation tests · cross-validation
3 half-days | CHF 900
Lecturer
In practice, outliers and deviations from idealised model assumptions affect data. The course covers robust regression methods: influence functions and breakdown points, regression M- and MM-estimation, robust inference, covariance estimation with a high breakdown point, and applications in principal component and linear discriminant analysis.
Concepts: Influence and breakdown point · M-estimators · robust regression · robust covariance estimation · outlier detection
3 half-days | CHF 900
Lecturer
When linear models cannot adequately capture the relationship between variables, nonlinear regression offers a flexible alternative. The emphasis is practical: model specification, parameter estimation and inference in nonlinear settings.
Concepts: Nonlinear regression models · estimation methods · approximate tests and confidence intervals · profile t-plots and profile traces · parameter transformations · prediction and calibration
5 half-days | CHF 1,500
Lecturer
Survival analysis addresses situations in which the outcome of interest is the time until an event occurs. The course presents widely used parametric models and methods, including accelerated failure time models, alongside the key nonparametric approaches: the Kaplan–Meier estimator and the Cox proportional-hazards regression model. The methods apply to data from clinical trials, public health, epidemiology, engineering, economics, psychology and demography.
Concepts: Censoring · Kaplan–Meier estimator · log-rank test · Cox proportional-hazards model · parametric survival models
Lecturers
This course provides an introduction to probabilistic deep learning. Deep learning is the method of choice for complex data such as images, and this course presents it as a continuation of familiar probabilistic models such as logistic regression within the framework of maximum likelihood and Bayesian methods. The course covers neural network architectures, including fully connected and convolutional networks, as well as how to select one for a given task and how to model different distributions of the target variable – Gaussian, Poisson or multinomial – according to the problem. It provides hands-on experience of building probabilistic deep-learning models, tuning them and controlling the training process.
Concepts: Feed-forward networks · back-propagation · regularisation and overfitting · convolutional and recurrent architectures · model evaluation
3 half-days | CHF 900
Lecturer
Repeated measurements arise when several observations are collected on the same experimental unit over time or space, which makes them correlated. The course covers analysis of longitudinal data and other repeated measurements: profile characteristics, compound symmetry and intra-class correlation, split-plot models, covariance structures for measurements separated in time or space, growth curves, multivariate analysis of variance, and cross-over experiments, in which several treatments are applied to the same unit.
Concepts: Within-subject correlation · mixed-effects models · longitudinal data · covariance structures
4 half-days | CHF 1,200
Lecturer
Many phenomena are spatial, making geographical analysis relevant across geology, soil science, agronomy, ecology, forestry, hydrology, climatology, human geography and epidemiology. Spatial datasets require specialised modelling to capture complex underlying structures, including spatial autocorrelation and dependence on explanatory variables. This module explores point-pattern distributions and spatial prediction techniques used to generate continuous surface maps. The course introduces geostatistical methods such as kriging and machine learning, with a focus on spatial prediction and model evaluation.
Concepts: Variograms · kriging · spatial point processes · areal data and neighbourhood models
Lecturer
Content pending.
Confirmation
Thank you for registering!