Multi-level state space models

Annual DECIDE workshop 2023

September 18-22 at University of Copenhagen
Focus of the workshop

In 2023, focus will be on multi-level (hierarchical) state space models - i.e. where data is monitored at several levels (e.g. herd/section/pen or country/region/farm). In most cases, the state space models will be Dynamic Linear Models (DLMs), but in some cases they will be Dynamic Generalized Linear Models (DGLMs) - typically because the response variable is binary.

In order to work with D(G)LMs at an advanced level, a good understanding of linear algebra is required. In case of multivariate models, also knowledge of the multivariate normal distribution is needed. In order to make sure that all participants have the necessary understanding of these topics, brush-up sessions are included in the workshop. 

It is strongly recommended that participants bring own time series data with them to the workshop. During the week, there will be time allocated to work with own data. For those who are not able to bring own data, a data set will be provided. At the end of the workshop, participants present the results of their own analyses. 

In 2023 we will try to integrate the participants' own data more actively in the plenum sessions compared to 2022. Therefore, on the first day of the workshop, participants bringing their own data with them will be asked to give a presentation of the data so that the organizers and the other participants will obtain a common understanding of the data.

Preliminary list of topics covered in 2023
  • Linear algebra, brush up
  • DLM, basic concepts
  • Estimation of variance components
  • DLMs with structure
    • Trends
    • Seasonal/diurnal patterns
    • Hierarchy
  • Early warning systems based on DLMs
Time schedule

The workshop will start at 10:00 on the 18th of September, and it will close at 14:00 on the 22nd of September. Other days, we typically start at 9:00 or 10:00 and close at 16:30.

A preliminary program for the workshop can be found here.

Lecturers and instructors

Carolina Mendes Galante Merca, PhD student
Leonardo Victor de Knegt, Assistant professor
Dan Børge Jensen, Associate professor
Anders Ringgaard Kristensen, Professor

Registration

Send an e-mail to ark@sund.ku.dk (Anders Ringgaard Kristensen) no later than the 1st of September 2023 and provide the following information:

  • Name
  • Title
  • Affiliation
  • Role in DECIDE 
  • A short description of your own data

The number of participants is limited to 20. Members of the DECIDE project group have priority but others may attend if there is room.

Deadline for registration is the 1st of September 2023.

Literature

A (the) classical introduction to the field is the book by West & Harrison:

West, M. & J. Harrison. 1997. Bayesian Forecasting and Dynamic Models. 2nd Edition. Springer Series in Statistics. Springer, New York.

In the workshop we will use our own textbook notes:

Kristensen, A.R., E. Jørgensen & N. Toft. 2010. Herd Management Science. II. Advanced topics. Academic Books, Copenhagen. ISBN: 9788763461214. 

The textbook will be supplemented with selected articles illustrating the methods discussed. A very brief overview of basic probability theory and distributions may be found in Appendix A of Volume I of Herd Management Science.

Software

The workshop will be based entirely on R and we use the RStudio graphical interface for R. All participants are expected to have both software systems installed. The workshop does not include any formal introduction to R. It is just used as a tool.

Practical information

A separate page provides practical information about transportation, meals and accommodation.