Hintergrundbild
Navigation

HMM workshop


Hidden Markov Models – Versatile Tools for Analysing Animal Movement and Other Time Series Data

Workshop at Bielefeld University, 4-7th September 2018



Overview

Hidden Markov models (HMMs) are flexible statistical models for sequences of observations that are driven by underlying states. In recent years, HMMs have become increasingly popular especially within the ecological community, where they constitute versatile tools for the analysis of animal behaviour data collected over time. Other areas where HMMs have successfully been applied include economics, sociology, psychology & medicine – basically any discipline where time series data are collected!

In ecological applications, an HMM-based analysis is very often natural and intuitive, with the aim of representing biological reality in terms of a fairly simple mathematical model – in these instances, the HMM states can often nicely be interpreted e.g. as behavioural states of the animal considered. The statistical model formulation then opens up the way for various types of biologically interesting inference to be drawn, including e.g. the effect of environmental conditions on animal behaviour, or on individual differences in behaviour. Notably, despite their relatively complex structure, HMMs turn out to be surprisingly easy to handle and are computationally feasible in almost all of the examples ecologists are dealing with on a regular basis.

This 4-day workshop gives a comprehensive introduction to the application of HMMs (in particular, but not only, to ecological data), comprising both theoretical lectures and hands-on practical sessions using R. In particular, participants will have the chance to analyse their own data within the practical sessions, supported by statisticians with considerable expertise on HMMs. In the first two days, the focus will be on the theory, whereas in the last two days we plan to spend more time working with you on your data.

 

Audience

This workshop is targeted primarily at ecologists working with time series data on animal behaviour, including animal movement. However, the general theory will be covered such that anyone interested in HMMs may find this workshop useful. It will be expected that participants are familiar with very basic statistics and probability theory (e.g. what is a probability distribution? what is a random variable? what is conditional probability?). Basic knowledge of the free software R would be advantageous, but is not required. In the afternoon of the day before the actual workshop, we will offer an introduction to R for participants with little to no previous experience with the software.

 

Content

In the theoretical sessions, the following topics will be covered:

  • motivation, overview & basic model formulation
  • fitting an HMM to data
  • model selection & model checking
  • state decoding
  • incorporating covariates, random effects and seasonality
  • extensions of the basic model formulation (e.g. multivariate time series)

The theoretical sessions will be accompanied by practical sessions using the statistical computing software R, where the theoretical concepts are implemented and illustrated using a real-data case study provided by us (or, optionally, the participants’ own data). The practical sessions will largely be based on the R package moveHMM, though we may occasionally also write code from scratch, depending on the participants’ interests.

 

Presenters

The workshop will be led by Vianey Leos Barajas & Roland Langrock (Bielefeld University).

              

Both have given workshops on HMMs before – here is some of the feedback we got:

"It is always a challenge for ecologists to transcend the gap between data and statistical techniques, which was recognised by Roland (likely because of his previous work with ecologists), and who tailored his explanations accordingly to the audience."

"Overall, I found the workshop helped progress my understanding of the application of Hidden Markov Models to my data. It gave me an overview of the possibilities, as well as the limitations. It also provided a much needed refresher on general statistical methodologies, and even helped clarify some overall concepts in statistics which previously I had not understood that well."

"I have attended several movement ecology / statistics workshops, and this was one of the best ones. The focus on a single analysis (with increasing layers of complexity) was good because it allowed delegates to fully understand the method. The course instructors were of exceptional caliber, and it was a privilege to learn from such skilled scientists."

 

Logistics & Computing Facilities

The workshop will take place at Bielefeld University, Germany (room W9-109 in the main building). Workshop participants are encouraged to bring their own laptops if possible. Unfortunately we cannot provide lab computers, but it will be possible to participate in the workshop also if no laptop is available.

The workshop is completely free of charge, but accommodation needs to be organised by the participants themselves, and also we do not provide any meals or drinks (except coffee!).

 

Registration

If you're interested in attending the workshop, or if you already want to register, just send an e-mail to Roland (roland.langrock@uni-bielefeld.de).

Aktuelles

New Paper

Paper accepted at the Journal of Sports Economics on the drivers of betting volumes in the English Premier League.

New member of the group

Sina Mews has joined the group as a PhD student.

New paper

Paper accepted at Movement Ecology on foraging strategies of reef sharks.

New member of the group

Vianey Leos Barajas has joined our group.

New paper

Paper accepted at Statistica Neerlandica on nonparametric inference in HMM-type models.

New paper

Paper accepted at Applied Economics on betting market inefficiencies in German football.

New paper

Paper accepted at Quantitative Finance that uses Markov-switching GAMs for forecasting operational losses in banking.

SFB

Roland is a PI within a new "Sonderforschungsbereich".

Karl Peter Grotemeyer Prize

Roland Langrock will be awarded the Karl Peter Grotemeyer Prize.