Fakultät für Mathematik, Informatik und Statistik - Digitale Hochschulschriften der LMU - Teil 02/02

Online Soziale Netzwerke (OSNs) repräsentieren das vorherrschende Medium zur computergestützten Kommunikation und Verbreitung persönlicher, geschäftlicher oder auch wissenschaftlicher Inhalte. Eine Reihe von Vorkommnissen in der jüngsten Vergangenheit hat gezeigt, dass die Bereitstellung privater Informationen in OSNs mit erheblichen Risiken für die Sicherheit und den Schutz der Privatsphäre seiner Nutzer verbunden ist. Gleiches gilt für die Bereiche Wirtschaft und Wissenschaft. Ursächlich dafür ist die zentralisierte Verwaltung der Nutzer und ihrer publizierten Inhalte unter einer singulären administrativen Domäne.

Mit Vegas präsentiert der erste Teil dieser Arbeit ein dezentrales OSN, das mit seiner restriktiven Sicherheitsarchitektur diesem Problem begegnet. Oberstes Ziel ist die technische Umsetzung des Rechts auf informationelle Selbstbestimmung. Dazu schränkt Vegas den Zugriff auf den sozialen Graphen und jeglichen Informationsaustausch auf die Nutzer des eigenen Egonetzwerks ein.

Neben der Möglichkeit zur Kommunikation und der Bereitstellung persönlicher Informationen erlauben einige OSNs auch das Browsen des sozialen Graphen und die Suche nach Inhalten anderer Nutzer. Um auch in sicheren und die Privatsphäre schützenden OSNs wie Vegas vom akkumulierten Wissen des sozialen Graphen zu profitieren, beschäftigt sich der zweite Teil dieser Arbeit mit der Entwicklung und Analyse intelligenter Priorisierungsstrategien zur Weiterleitung von Suchanfragen innerhalb dezentraler OSNs.

Im Kontext von OSNs werden neue Algorithmen und Protokolle zunächst simulativ evaluiert. Die Grundlage bildet in der Regel der Crawling-Datensatz eines OSNs. Offensichtlich ist das Crawling in sicheren und die Privatsphäre schützenden dezentralen OSNs wie Vegas nicht möglich. Um diesem Problem zu begegnen, beschäftigt sich der dritte Teil dieser Arbeit mit der Entwicklung eines generischen Modells zur künstlichen Erzeugung sozialer Interaktionsgraphen. Neben den strukturellen Besonderheiten zentralisierter und dezentraler Systeme wird erstmals auch das Interaktionsverhalten der Nutzer eines OSNs modelliert. Die Eignung des Modells wird auf der Grundlage gecrawlter sozialer Graphen evaluiert.

This dissertation explores methods to improve the quality of data about sensitive labor market topics, such as undeclared work and receipt of basic income support in Germany, using surveys of the general population. Due to the sensitive nature of both topics, respondents may choose to misreport and adjust their answers in accordance with social norms.

Over the past decades, special strategies—particularly targeted to reduce misreporting on sensitive topics—have been developed. One such class of data collection strategies are so-called ‘dejeopardizing’ techniques, out of which the randomized response technique (RRT) and the item count technique (ICT) are the most popular and best investigated ones. The goal is to elicit more honest answers from respondents by increasing the anonymity of the question-and-answer process. These techniques provide prevalence estimates as well as estimates of regression coefficients, regressing dependent variables generated by means of RRT or ICT on a set of covariates of interest.

However, these dejeopardizing techniques have not been applied to collect data on undeclared work or receipt of welfare benefits in German surveys. This dissertation aims at closing this gap using an experimental design that allows us to compare the performance of these dejeopardizing techniques to direct questioning. In 2010 we conducted two telephone surveys on undeclared work and welfare benefit receipt. We experimentally tested whether the RRT, the ICT, or the newly developed item sum technique (IST) reduce bias due to social desirability compared to direct questioning (under the ‘more-is-better’ assumption and using validation data in one study).

Our results suggest that neither the RRT nor the ICT provide unambiguous results with respect to improving the accuracy of reports of the socially undesirable behavior, while the IST results were more promising. This dissertation provides insights into a variety of practical and theoretical factors contributing to a successful implementation of the RRT, the ICT and the IST in labor market surveys.

Various experiences in modern life are in one or another form connected to car rides. However, the automotive industry so far only regards driving as the only relevant experience, a perspective which consequently dominates the field of interaction design for vehicles. Yet, the car is an exceptionally potential space for experiences that go beyond the actual driving task, such as intensive conversation or exploration. So how is it possible to design for special situations like those, and thereby create positive emotions for drivers and passengers alike?

To meet this objective it is necessary to use the human, instead of the function, as the starting point. The design approach of "Experience Design" defined by Marc Hassenzahl provides exactly this focus on the human and concentrates first on their experience. Here, positive emotions are specifically created through the fulfilling of psychological needs.

Experience Design enables the precise analysis of experiences in advance of the design by clarifying with the help of psychological needs why a considered experience is viewed as positive. Furthermore, Experience Design supports the composition of an Experience Story, which is attuned to the desired psychological needs and which defines the experience to be designed. This experience can then gradually be translated into an interaction design. Finally, with the help of technology, the created experience can be lived through in situ by participants and later analysed. Based upon this design approach and by means of methods drawn from the fields of human machine-interaction as well as psychology, four studies on the design of experiences through interaction products in the automotive domain are presented. The created experiences are divided into "Experiences at Group Drives in the Car for Pleasure" and "Experiences While Commuting Alone". These experiences take place in different scenarios, namely: in a motorcade, an exploratory cruise, a commuting ride, and while driving considerately. Out of the practical application of Experience Design in these studies a best practice for the use of the employed methods is developed. Thereby, this work brings to light the possibilities for using technology to design experiences that go beyond the mere act of driving. Furthermore, the challenges of designing experiences in usability-focused environments are shown. Thus, this work is aimed at offering inspiration to designers and researchers particularly in the automotive domain for designing experiences and thereby furthering innovation.

The aim of this thesis is to develop new statistical methods for the evaluation of assumptions that are crucial for reliably assessing group-differences in complex studies in the field of psychological and educational testing. The framework of item response theory (IRT) includes a variety of psychometric models for scaling latent traits such as the widely-used Rasch model. The Rasch model ensures objective measures and fair comparisons between groups of subjects. However, this important property holds only if the underlying assumptions are met. One essential assumption is the invariance property. Its violation is extensively discussed in the literature and termed differential item functioning (DIF). This thesis focuses on the methodology of DIF detection. Existing methods for DIF detection are briefly discussed and new statistical methods for DIF detection are introduced together with new anchor methods. The methods introduced in this thesis allow to classify items with and without DIF more accurately and, thus, to improve the evaluation of the invariance assumption in the Rasch model. This thesis, thereby, provides a contribution to the construction of objective and fair tests in psychology and educational testing.