10. Dezember 2019 fand an der Universität Bielefeld der 7. Nachwuchsworkshop des Zentrums für Statistik statt.
Im Rahmen des Nachwuchsworkshops haben sich Doktorandinnen und Doktoranden der am Zentrum für Statistik beteiligen Bereiche gegenseitig ihre Forschungsfelder vorgestellt und darüber diskutiert.
Marvin Bürmann
Universität Bielefeld, Fakultät für Soziologie
Standardized Occupations versus Unstandardized Qualifications – Undereducation of Immigrants in Germany
The paper investigates the effect of occupational standardization on employees’ probability of getting high profile jobs without having the necessary formal certificates. Based on the Socio-economic panel from the years 2013 to 2016 it is shown for the first time, that such undereducation becomes more unlikely, the more standardized an occupation is with regard to its certificates. This closure effect is in line with expectations drawn from the literature about occupational closure. It was also unclear whether this effect is independent of the certificates’ origin. The results reveal in a robust manner, that lower qualified migrants from third countries are not affected by this closure effect. Apparently employers take into account that a standardized vocational education system rarely exists outside Germany and that most of the practical skills are therefore informally acquired on the job. To make these skills transparent to the whole labor market, it would need recognition processes for informally acquired skills additional to the equivalence testing of foreign certificates.
Sebastian Büscher
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften
Using Motifs in person-centred mobility simulation models
Person-centred mobility simulation models simulate the mobility choices of each individual in a population over a full day. The realisation of these choices leads to trips which consequently determine the demand for different transport infrastructures. Such simulation models need to encode the mobility plans of the whole population in a way that is representative for the investigated population. Recently mobility motifs have been shown to provide a representation of certain features of the daily activity. The distribution of the motif choice is remarkably stable for different regions. Here we want to discuss how the synthetic populations are typically generated and how motif choice can be used to enhance this process while preserving the representativity of the synthetic population in terms of different observed characteristics of the investigated population. To do so, the dependence of motif choice on a number of underlying sociodemographic features is investigated.
Jan Klostermann
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften
How Sponsored Posts effect Audience Engagement in Social Media
Due to their often high credibility and expertise, social media opinion leaders have the power to influence the behavior of other consumers. Marketers have recognized this high persuasiveness and consequently pay these opinion leaders to create positive electronic word-of-mouth for their brands and products. As sponsored posts need to be disclosed, social media opinion leaders and brands have to consider possible negative outcomes of this practice. While current research investigates the effect of sponsorship disclosure on, for example, perceived credibility and authenticity, there is no examination of how sponsored posts impact follower engagement (i.e. likes and comments) of real social media opinion leaders. In an empirical analysis of more than one million posts created by Instagram opinion leaders, we can show that sponsored posts create less follower engagement and that a high frequency of sponsored posts has a negative impact on the engagement of sponsored posts, such that they are liked less often. We compare these findings across opinion leaders in different areas of interest and discuss theoretical and managerial implications.
Alexander Max
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften
Emojis and Brand Dimensions
Measuring hedonic attitudes of consumers towards a brand has been a challenging matter in marketing research for many years. With the increasing popularity of social networks, the expressed content of customers can be seen as a valuable and accessible source to derive customer’s perception of the hedonic dimensions of a brand. While current literature focuses mainly on user-generated text to predict this perception, we provide an extended approach by integrating emojis as a powerful expression of consumer’s affective attitudes. In a first step, we identify a set of brands and their hedonic characteristics. Secondly, we develop a statistical model to predict hedonic properties based on a wide range of collected emoji data from brand related user-generated content. We compare several modifications of the used data with respect to the performance regarding the prediction of hedonic attitudes and contribute to open up user-generated content as a base of managerial marketing decisions.
Sina Mews
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften
A multi-state capture-recapture model in continuous time for the annual movement of bottlenose dolphins
Our modelling approach is motivated by individual capture histories of bottlenose dolphins off the east coast of Scotland. For these individual histories, standard capture-recapture methods are not readily applicable as they are designed for regular sampling protocols, whereas in our case, the capture occasions are not regularly spaced in time. Therefore, we consider a continuous-time model formulation. The capture-recapture setting can be regarded as a special case of a (partially) hidden Markov model (HMM), with the observed capture history of an individual as the state-dependent process and an underlying, partially observed state process related to the movement of the individual between different sites. In particular, we can exploit the efficient HMM-based forward algorithm for evaluating the likelihood and hence for parameter estimation. The main interest of the present analysis lies in the annual movement patterns of the dolphin population between different sites. Therefore, we model the dolphins’ movement rates, expressed as state transition intensities in our model, as a function of time of year. However, incorporating such time-varying covariates into the continuous-time Markov state process is rather challenging as the corresponding likelihood function then becomes intractable. We suggest an approximation using piecewise constant state transition intensities, which renders the likelihood evaluation feasible. The approximation can be made arbitrarily accurate by using an increasingly fine resolution of the approximating step function. The latter is demonstrated in simulation studies, in which we also investigate the effect of different parameter values on the estimation accuracy.
Marius Ötting
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften & Fakultät für Psychologie und Sportwissenschaft, Abteilung Sportwissenschaft, Arbeitsbereich V - Sport
Performance when stakes are high: do professional darts players keep their cool?
Understanding and predicting how individuals perform in high-pressure situations is of importance in designing and managing workplaces. In this talk, performance under pressure in professional darts as a near-ideal setting with no direct interaction between players and a high number of observations per subject is investigated. Analyzing almost one year of tournament data covering 44,071 dart throws, we find contrary to what would be expected given the evidence in favor of a choking phenomenon, that professional darts players excel under high pressure. These results could have important consequences for our understanding of how highly skilled individuals deal with high-pressure situations.
Jennifer Pohle
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften
Functional data analysis for modeling distinctive shapes in high-resolution state-switching time series
Hidden Markov models (HMMs) are widely used in ecology to analyse animal movement data like GPS location or diving data. The underlying hidden states of the model are usually assumed to be proxies for the behavioural modes of the animal (e.g. foraging or traveling), which underlie the observed movement patterns. However, due to substantial developments in tagging technology, it is nowadays possible to collect movement data at a very fine temporal resolution and HMMs can struggle to capture the strong serial correlation of the observed time series. Therefore, instead of the complete data, often subsamples or summary statistics across fixed time intervals are analysed. For instance, in the context of diving data, the maximum depth and duration of each dive could be considered. This, however, can lead to a loss of information because only some observations enter the analysis. We therefore propose to combine functional data analysis methods and hidden Markov models to model characteristic shapes of the dives. This allows us to take into account the complete observed time series and could lead to better insights into the movement patterns and behavioural modes of the animals.
Dorian Tsolak
Universität Bielefeld, Fakultät für Soziologie
Detecting stereotypes and extremist views in big online data
Spaces of online communication such as social media platforms are known for heated political discussions especially with regards to topics such as immigration or gender equality (Müller & Schwarz 2018). For instance, right-wing political parties in Europe have been able to gain a lot of traction in recent years by sharing populist views related to immigration on social media (Engesser et al. 2017). A main challenge when analyzing specific (e.g. extremist) discourses on such platforms consists of extracting cases relevant to the research question at hand from the full data set. Our research makes use of word embeddings models, which allow for unsupervised topic modeling when combined with a clustering algorithm of choice. Using the resulting clusters, data for specific topics can be retrieved from the full data set and analyzed in depth.
Mattias Ulrich
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften
Dynamic model selection in demand forecasting using supervised learning on data set characteristics
Retailers usually supply a wide range of stock keeping units (SKUs), which may differ for example in terms of demand quantity, demand frequency, demand regularity, and demand variation. Given this diversity in demand patterns, it is unlikely that any given method for demand forecasting will be cost optimal across all SKUs. To minimise overall costs, there is thus a need to match any given demand pattern to its associated cost-optimal method.
Here, we propose a dynamic model selection framework for demand forecasting using supervised learning. Specifically, we consider model selection as a classification problem, where the classes correspond to the different methods available for forecasting. Based on the most recent data for the SKU of interest, we identify past occasions with similar characteristics and classify according to the different methods' performances for those test instances. The performance is measured taking into account asymmetric underage and overage costs, as specified by the retailer. As benchmark for our approach, we consider several methods that are commonly applied, namely individual selection, aggregate selection, and combination forecast. The models are evaluated in a case study using data from an e-grocery retailer, where we compare the out-of-sample predictive performance.
Houda Yaqine
Universität Bielefeld, Fakultät für Wirtschaftswissenschaften
What do we know about switching diffusion models?
In this talk an overview about the switching diffusion models and the basic mathematical results from the literature is given, as well as their importance from a modeling point of view of real-world phenomena.