Dienstag, 12.04.2022, 12-13 Uhr
Dr. Peter Pütz
Rounding and Other Pitfalls in Meta-Studies on P-Hacking and Publication Bias
Brodeur, Cook, and Heyes (2020) study hypothesis tests from economic articles and find evidence for p-hacking and publication bias. We introduce a novel method that adjusts for rounding errors. When applying this adjustment, evidence for p-hacking in the data set substantially weakens. In addition, Brodeur, Cook, and Heyes derive latent distributions of z-statistics absent publication bias using two different approaches. We establish for each approach a result that challenges its general applicability.
Dienstag, 26.04.2022, 12-13 Uhr
An expectation-maximization algorithm for hidden Markov models with random effects in the state process
Hidden Markov models are applied to time series data where the observed variables are driven by underlying latent states. For multiple time series, e.g. of individual animals of the same species, including random effects can help to parsimoniously account for the differences between these animals. However, estimation methods for the resulting models are complicated and time-consuming. I explore an EM algorithm that simplifies the estimation by making use of the Viterbi algorithm within the E-step to be able to use methods from simple mixed logistic regression within the M-step. To assess some properties of the estimations with this algorithm and its practicability, a simulation study was conducted and a data example of Caracaras birds is shown.
Dienstag, 10.05.2022, 12:15-13:15 Uhr
Prof. Dr. Dietmar Bauer
The state space error correction model representation (SPECM)
The phenomenon of cointegration features prominently in econometric time series analysis. Many economic time series show a trending behaviour that is not well described using a deterministic trend. When modeling multiple time series jointly it has been noted, that ofentimes different variables trend jointly such that static linear combination are well described as stationary processes.
This phenomenon is most often attacked using autoregressive models which are transformed into the vector error correction model (VECM) form to ease estimation. Soren Johansen and coworkers provided the methods to work with VECMs which subsequently have found a prominent place in the toolset of econometricians.
In this talk we discuss the extension of the VECM to the state space framework which call state space error correction model (SPECM). We show that the well known Johansen procedures for estimating the cointegrating rank as well as the cointegrating space can be applied with slight modifications to the case of state space models.
Dienstag, 24.05.2022, 12-13 Uhr
Prof. Dr. Roland Langrock
Hidden Markov models in ecology: ever more complex?
Over the last 10-15 years, hidden Markov models (HMMs) have very rapidly gained popularity within the statistical ecology community thanks to their versatility to accommodate various types of time series data. HMMs are nowadays routinely applied in particular to animal movement as well as capture-recapture data, and have also been demonstrated to be useful for occupancy modelling, within distance sampling, and for estimating models for population dynamics. The transition from a rather specialised statistical tool to a technique that is now regularly applied by ecologists has been accompanied by a notable increase in the complexity of HMM formulations now commonly used in the ecological literature. In this talk, I will reflect on these developments, showcasing some recent work on extensions of the basic HMM formulation, and using those examples to point out the risks associated with the community using ever more complex HMM formulations.
Dienstag, 07.06.2022, 12-13 Uhr
Priv.-Doz. Dr. Ursula Berger & Prof. Dr. Göran Kauermann
Statistical modelling of COVID-19 data: Putting Generalised Additive Models to work
Over the course of the COVID-19 pandemic, Generalised Additive Models (GAMs) have been successfully employed on numerous occasions to obtain vital data-driven insights. In the talk we substantiate the success story of GAMs and demonstrate their flexibility by focusing on two specific pandemic-related issues. First, we examine the intercedence among infections in different age groups, concentrating on school children. In this context, we derive the setting under which parameter estimates are independent of the (unknown) case-detection ratio, which played an important role in COVID-19 surveillance data. Second, we propose a multinomial model for the weekly occupancy of intensive care units (ICU), where we distinguish between the number of COVID-19 patients, other patients and vacant beds. The modelling implicitly takes into account that COVID-19 patients displaced regular patients. With these two examples, we aim to showcase the practical and "off-the-shelf" applicability of GAMs to gain new insights from real-world data.
Dienstag, 21.06.2022, 12-13 Uhr
Support vector machine quantile regression based ensemble postprocessing
Current practice in predicting future weather is the use of numerical weather prediction (NWP) models to produce ensemble forecasts. Despite of enormous improvements over the last few decades, they still tend to exhibit bias and dispersion errors and consequently lack calibration. Therefore, these forecasts need to be statistically postprocessed.
Support vector machines are often used for classification and regression tasks in a wide range of applications, as e.g. energy, ecology, hydrology and economics. In this study, ensemble forecasts of 2m surface temperature are post-processed using a quantile regression approach based on support vector machines (SVMQR). This approach will be compared to the benchmark postprocessing methods ensemble model output statistics (EMOS), boosted EMOS and quantile regression forests (QRF). Instead of only regarding temperature variables as predictors, other weather variables including time dependence are taken into account as independent variables. The considered dataset consists of observations and forecasts for five years which cover Germany including three different forecast horizons. Despite of a shorter training period for SVMQR in contrast to e.g. boosted EMOS or QRF, SVMQR yields more calibrated quantile ensemble forecasts than the other approaches. Additionally, a comparable performance in terms of CRPS to the benchmark methods and a great improvement in comparison to the raw ensemble forecasts could be detected.
Dienstag, 05.07.2022, 12-13 Uhr
Interventional effects for mediation analysis when the causal structure amongst mediators in unknown