Stats staff seminar
Philipp Otto & Isa Marques (University of Glasgow)
Friday 8th December, 2023 13:00-14:00 Maths 311B
1. Navigating Spatial Confounding: Understanding Causes and Proposing Future Approaches (Isa): Spatial confounding is a fundamental problem in regression models for spatially indexed data. It arises because spatial random effects, which are included to approximate unmeasured spatial variation, are typically not independent of the covariates in the model. This can lead to significant bias in the estimates of covariate effects. We develop a comprehensive theoretical framework that brings mathematical clarity to the mechanisms of spatial confounding and provides explicit and interpretable analytical expressions for the resulting bias. We then explore the potential of Bayesian methodology to exploit knowledge of how spatial confounding arises and to mitigate it.2. Overview of selected topics of my past and future research (Philipp): In this talk, I will give an overview of my past and ongoing research. My research is mainly centred around spatial and spatiotemporal statistics and network models with various applications in different fields. I will start with spatial and spatiotemporal autoregressive conditional heteroscedasticity (ARCH) models, which we have introduced in Otto et al. (2018, Spatial Statistics). I will also briefly outline the various extensions of this model class that we have done in recent years and show selected examples of applications, e.g., environmental risk modelling (Otto et al. 2023, Econometrics and Statistics) or volatility forecasts on financial networks (Mattera/Otto 2023). Next, I will present my papers on the penalised estimation procedures for entire spatial dependence structures, i.e. all O(n^2) connections between all n observation locations (e.g., Otto/Steinert 2023, Journal of Computational and Graphical Statistics), and regularised estimation procedures for model selection of functional geostatistical models. We have applied these functional models in various fields, e.g., coastal profile modelling (Otto et al. 2021, Coastal Engineering) or bike-sharing usage (Piter et al. 2022, JRSSA). Finally, I will briefly discuss my research in the area of statistical process monitoring, especially for online network monitoring. In this area, I will focus on our paper on monitoring artificial intelligence, in particular, artificial neural networks using data depths (Malinovskaya et al. 2023, Technometrics).