Trondheim Symposium in Statistics 2021

The symposium will take place on November 5 and 6 at the Ørland Kysthotell.

Invited speakers:
Solveig Engebretsen (Norwegian Computing Center),
Mollie Brooks (National Institute of Aquatic Resources , Danish Technical University),
Bjørn Heine Strand (Norwegian Institute of Public Health),
Pierre Druilhet (Université Clermont Auvergne),
Nils Lid Hjort (University of Oslo).

Organising committee: Gunnar Taraldsen and Jarle Tufto.

List of participants.

To sign up, contact the organisers no later than October 19 21.

14:45–15:55: Boat from the Trondheim hurtigbåt terminal to Brekstad Check-in. Please be there 2030 mins before departure (at 14:2514:15).
16:00-16:30 Check-in and coffee break.
16:30–17:30: Solveig Engebretsen: Spatial modelling of COVID-19 for situational awareness and forecasting
17:30–18:30: Mollie Brooks: Template Model Builder Applications in Ecology
19–: Dinner and socialising

7:30-8:45: Breakfast
8:45-9:35: Bjørn Heine Strand: Statistical tools in epidemiology and public health
9:35-9:50: Coffee break
9:50-10:40: Pierre Druilhet: Bayesian inference with improper distributions
10:40-11:30: Nils Lid Hjort: The II-CC-FF paradigm for combining information across diverse sources: theory and applications
11:30-12:30: Lunch
12.40-13:30: Boat to Trondheim

Presentations are about 45 min, and then there is 5 minutes time for questions.

Solveig Engebretsen: Spatial modelling of COVID-19 for situational awareness and forecasting

In this talk, I will present the model that the Norwegian Institute of Public Health has been using for situational awareness and forecasting of COVID-19 in Norway. The model is informed by real-time mobile phone mobility data and seeded with all the registered imported cases from abroad. The model is calibrated to daily hospital incidence and laboratory-confirmed cases. Several aspects of the epidemic vary over time: transmission, import, behaviour, intervention policies, mobility, viral variants, vaccination coverage, testing regimes, hospital load, and treatment. These aspects may affect the two data sources differently, challenging consistency of the two data sources under the model, and hence also inference based on both data sources simultaneously. We therefore need to make sure that the model captures all these changes, to make sure that it is consistent under both data sources over time. In addition to estimating reproduction numbers for situational awareness, we provide short-term hospitalisation predictions to the hospitals for planning purposes. By comparing the reproduction number when different restrictions were in place, we can assess the effect of the interventions. The mobility can also be used as a surveillance tool, to monitor compliance with mobility-reducing policies and import.

Mollie Brooks: Template Model Builder Applications in Ecology

In this talk, I will give a short introduction to TMB, assuming that the audience potentially has some knowledge of INLA. Then, I’ll show a few uses of TMB from my own research in demography. We used a very flexible form of an autoregressive model to model sheep growth as a state-space model because of missing observations and observation error. This allowed us to account for transient environmental effects as well as random variation in individual quality. I will introduce the R package glmmTMB which is a flexible method for fitting GLMMs, potentially with zero-inflation. Then, I will demonstrate using glmmTMB with two uncommon distributions: the Conway-Maxwell-Poisson distribution and the Tweedie distribution.

Pierre Druilhet: Bayesian inference with improper distributions.

In the first part of this talk, we discuss some results about improper priors. According to Jaynes (2003, Probability theory : the logic of Science,p. 487) "If we wish to consider an improper prior, the only correct way of doing it is to approach it as a well-defined limit of a sequence of proper priors." We propose here a convergence mode adapted to improper distributions and show how the convergence can be transmitted to the posterior distributions. Despite Jaynes' claim, we also propose to consider improper (and proper) distributions by themselves as objects in an appropriate space, allowing also improper posteriors. In the second part, we consider the problem of the estimation of animal abundance by removal sampling. We show that the usual default priors lead to improper posteriors. Then, we discuss a class of priors with satisfactory frequentist properties.

Nils Lid Hjort: The II-CC-FF paradigm for combining information across diverse sources: theory and applications

We introduce and develop a general paradigm for combining information across diverse data sources. In broad terms, suppose \(\phi\) is a parameter of interest, built up via components from different (and perhaps very different) data sources. The proposed scheme has three steps. First, the independent inspection (II) step amounts to investigating each separate data source, summarising and translating statistical information to a confidence distribution for the component of the problem. Second, confidence conversion (CC) techniques are used to translate the CDs to confidence log-likelihood functions. Finally, the focused fusion (FF) step uses relevant and context-driven techniques to construct a confidence distribution for the primary focus parameter. The II-CC-FF scheme represents significant extensions of various meta-analysis methods. Theory is outlined and applications are discussed. This is joint work with Céline Cunen; see our recent SJS 2021 paper.