8. Nachwuchsworkshop

29. November 2019 fand an der Universität Bielefeld der 8. Nachwuchsworkshop des Zentrums für Statistik statt.

Hannah Busen
Helmholtz Zentrum München & Universität Bielefeld, Fakultät für Wirtschaftswissenschaften

Modelling the impact of spatial proximity on scientific collaboration considering network structures

Spatial proximity between researchers may lead to more frequent or more intense collaboration than between scientists who work at large distance from each other. In this project, we aim to investigate the impact of spatial proximities within research institutions, virtual research networks and targeted funding schemes on the implementation of interdisciplinary collaborations. To assess the impact of spatial networks, virtual networks and cooperation support on the strength of interdisciplinary cooperation we will use a variety of data science techniques, mainly based on statistical testing and (regularized) regression analysis. Defining collaboration as the number of common publications between two researchers, regression models will be used to answer the question if researchers are more likely to work together if they have a small distance to each other. The focus lies on different possibilities to account for the dependency structure in network data. Different approaches, upcoming problems and ideas will be discussed. Data contains publications and building distances from the Helmholtz Zentrum Munich and the Bielefeld University for the years 2015 to 2018, and detailed information about citations, key words and author affiliation.

 

Georg Kessler
Universität Bielefeld, Fakultät für Soziologie

’Late-Bloomers’: Fact or Artifact? - Insights Into Delinquent Trajectories of Young Adults

While the bulk of research conducted on trajectories of delinquency covers the periods of childhood and adolescence well, less is known about delinquent behavior during the transition from adolescence to young adulthood. Total delinquency decreases until about the age of 19 years and
then stabilizes. Recently, a debate on the existence of a possible adult-onset trajectory is discussed among Development and Life Course criminologists. Longitudinal studies comprising the age periods of adolescence and young adulthood did not identify this group through semi-parametric
modeling techniques when using the entire panel, which is not surprising: after all, the overall tendency with increasing age is desistance. However, applying only the time period of young adulthood, a trajectory of what could be claimed to be adult-onset can be distinguished. The paper focuses on criminal trajectories of young adults aged between 18 and 28, the most common drivers into and out of crime and how well these predict belonging to a delinquent trajectory in this specific population. Trajectories will be explored and estimated via growth mixture modeling (GMM). The population under study are participants of the German panel study "Crime in the Modern City" with annual and biannual surveys from 13 to 30 years of age. The following
questions are addressed in the paper: Beside a general decrease of crime in late adolescence, can we identify of shifting of crime patterns? How does this relate to the opportunity structure changes in young adulthood? How well do common explanatory factors of delinquency trajectories for adolescents perform among young adults? Can we propose a reasonable and refined way to capture the concept of a "cocoon" during adolescence for late bloomers and does it apply as an explanatory factor for the hypothetical trajectory of adult-onset within the study?

 

Clemens Mahlmeister
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften & BearingPoint, Düsseldorf

High Resolution Motion Prediction with HMMs

Mobile Robots conduct more and more transportation tasks in the digital factories of automotive OEMs and suppliers. Motion prediction can alleviate current challenges such as traffic jams and deadlock scenarios by making autonomous delivery vehicles or robots aware of their surroundings and their surroundings’ motion intent. Common traffic participants in a factory are pedestrians as well as vehicles and bicycles. The Geolife dataset gathered by Microsoft features high resolution movement data and is utilized to learn the motion intent of these participants. This allows for predicting short-term trajectories and efficiently steer mobile robot through factories. This presentation aims to discuss two approaches with regards to modelling motion. Both approaches are based on Hidden Markov Models that are often used to model long-term animal movements but are new to modelling short-term trajectories in the area of seconds. Initial results are presented.

 

Patrick de Matos Ribeiro
Universität Dortmund, Fakultät für Statistik

Stochastic trends and economic fluctuations reconsidered

In their seminal paper King, Plosser, Stock and Watson (American Economic Review, 1991; KPSW) investigate the empirical performance of the neoclassical stochastic growth model by vector autoregressive (VAR) cointegration analysis using quarterly observations from 1949:1 to 1988:4. In addition to the real variables, private consumption, private investment and GNP net of government spending, they also consider nominal quantities, i.e., money (M2), prices (GNP deflator) and a short-term interest rate (federal funds rate). Economic theory predicts three cointegrating relationships, the two great ratios (consumption-ouptut and investment-output) as well as a money demand relationship. Furthermore, the relative importance of permanent and transitory shocks for the dynamic behaviour of the US economy is investigated. As is well-known, the solutions of dynamic stochastic economic models are generally vector autoregressive moving average rather than VAR processes, with the solutions typically given in state space format. Building upon recent advances in state space cointegration analysis, this paper reassesses, and extends by using data potentially up to 2018:4, the KPSW analysis from a state space perspective.

 

Lukas Matuschek
Universität Dortmund, Fakultät für Statistik

Pseudo maximum likelihood analysis of I(2) processes in the state space framework

Nominal macroeconomic time series are regularly found to be adequately described as I(2) processes, with cointegration analysis typically performed in the vector autoregressive (VAR) framework. The VAR framework may be too restrictive: First, VAR processes are not closed under marginalization or aggregation, where in both cases the resulting processes are in general vector autoregressive moving average (VARMA) processes. Second, the solutions of dynamic stochastic economic models are typically VARMA processes rather than VAR processes. To overcome the limitation to VAR processes we develop estimation and inference techniques for I(2) cointegrated VARMA processes cast in state space format. In particular we derive consistency as well as the asymptotic distributions of estimators maximizing the Gaussian pseudo likelihood function. As usual, the parameters corresponding to I(2) and I(1) variables are estimated super-consistently at rates T² and T respectively, whereas all other parameters are estimated at rate T^1 / 2. The limiting distributions of the parameters corresponding to the integrated components are mixtures of Brownian motions, the parameters of the stationary subsystem are asymptotically normally distributed. Furthermore, we discuss hypothesis tests for the cointegrating ranks as well as for the cointegrating spaces.

 

Sina Mews
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften

Investigating the hot hand effect in continuous time

In general, the so-called hot hand effect refers to the idea that a person is in a state in which his success probability is greater than usual. The term is, for example, often used in sports like basketball, where people assume that a player has a “hot hand” when he hits several shots in a row. We model this phenomenon using a continuous-time state-space model to investigate serial dependence between observations, which correspond to binary indicators of success. In particular, the observations depend on a latent state process that serves as a proxy for the player’s current form. As observations may not be regularly spaced in time, we use the Ornstein-Uhlenbeck process to model how the state process evolves continuously over time. The feasibility of our approach and especially its sensitivity to parameter constellations of the Ornstein-Uhlenbeck process are explored in simulation studies. Finally, our approach is applied to data containing free throws from basketball.

 

Marius Ötting
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften & Fakultät für Psychologie und Sportwissenschaft, Abteilung Sportwissenschaft, Arbeitsbereich V - Sport

A regularized hidden Markov model for analysing the "hot shoe" in football

Although academic research on the "hot hand" effect (in particular, in sports, especially in basketball) has been going on for more than 30 years, it still remains a central question in different areas of research whether such an effect exists. In this talk, the potential occurrence of a "hot shoe" effect for the performance of penalty takers in football is investigated based on data from the German Bundesliga. For this purpose, hidden Markov models (HMMs) are considered to model the (latent) forms of players. To further account for individual heterogeneity of the penalty taker as well as the opponent's goalkeeper, player-specific abilities are incorporated in the model formulation together with a LASSO penalty. The results suggest states which can be tied to different forms of players, thus providing evidence for the hot shoe effect, and shed some light on exceptionally well-performing goalkeepers.

 

Jennifer Pohle
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften

Flexible estimation of the state dwell-time distribution in hidden semi-Markov models

Hidden Markov models (HMMs) are flexible time series models for observations that are driven by an underlying latent state sequence. For mathematical convenience, the state sequence is usually assumed to be a (homogeneous) first-order Markov chain. This, however, implies that the state dwell time, i.e. the number of consecutive time points spent in a given state, follows a geometric distribution. Consequently, the mode of the dwell-time distribution is one, a property that in practice is not always desirable. Hidden semi Markov models (HSMMs) overcome this limitation by allowing for an arbitrary discrete dwell-time distribution. Typical choices in applications are parametric distributions such as the Poisson or the negative binomial distribution. However, this again a priori restricts the overall shape of the dwell-time distribution. In this talk, we propose a fully data-driven penalized maximum likelihood approach to explore the dynamics of the state process. The approach is illustrated using animal movement data from muskoxen in Greenland.

 

Houda Yaquine

Universität Bielefeld, Fakultät für Wirtschaftswissenschaften

Stochastic modeling of childhood Leukemia cells mechanisms

In this project, we investigate a Leukemia burden measurements data set from a pediatric patient-derived xenograft (PDX) model. Our goal is to build different stochastic models (e.g. diffusion models, switching diffusion models) that can describe this data set to help us understand the dynamics of the Leukemia cells and their different behaviors. A model choice study is aimed to choose the best appropriate model; the understanding of the mechanisms of these cells will allow creating new therapeutic strategies that can reduce the possibility of the disease relapse.