Open Access

Online behaviors analysis consists of extracting patterns from server-logs. The works presented here were carried out within the “mBook” project which aimed to develop indicators of the quantity and quality of the learning process of pupils from their usage of an eponymous electronic textbook for History. In this thesis, we investigate several models that adopt different points of view on the data. The studied methods are either well established in the field of pattern mining or transferred from other fields of machine learning and data-mining. We improve the performance of archetypal analysis in large dimensions and apply it to unveil correlations between visibility time of particular objects in the e-textbook and pupils’ motivation. We present next two models based on mixtures of Markov chains. The first extracts users’weekly browsing patterns. The second is designed to process essions at a fine resolution, which is sine qua non to reveal the significance of scrolling behaviors. We also propose a new paradigm for online behaviors analysis that interprets sessions as trajectories within the page-graph. In this respect, we establish a general framework for the study of similarity measures between spatio-temporal trajectories, for which the study of sessions is a particular case. Finally, we construct two centroid-based clustering methods using neural networks and thus lay the foundations for unsupervised behaviors analysis using neural networks. Keywords: online behaviors analysis, educational data mining, Markov models, archetypal analysis, spatio-temporal trajectories, neural network

Network analysis methods have long been used in the social sciences. About 25 years ago, these methods gained popularity in various other domains and many real-world phenomena have been modeled using networks. Well-known examples include (online) social networks, economic networks, web graphs, metabolic networks, infrastructure networks, and many more. Technological development made it possible to store and process data on a scale not imaginable decades ago — a development that also includes network data. A particular characteristic of network data is that, unlike standard data, the objects of interest, called nodes, have relationships to (possibly all) other objects in the network. Collecting empirical data is often complicated and cumbersome, hence, the observed data are typically incomplete and might also contain other types of errors. Because of the interdependent structure of network data, these errors have a severe impact on network analysis methods. This cumulative dissertation is about the impact of erroneous network data on centrality measures, which are methods to assess the position of an object, for example a person, with respect to all other objects in a network. Existing studies have shown that even small errors can substantially alter these positions. The impact of errors on centrality measures is typically quantified using a concept called robustness. The articles included in this dissertation contribute to a better understanding of the robustness of centrality measures in several aspects. It is argued why the robustness needs to be estimated and a new method is proposed. This method allows researchers to estimate the robustness of a centrality measure in a specific network and can be used as a basis for decision making. The relationship between network properties and the robustness of centrality measures is analyzed. Experimental and analytical approaches show that centrality measures are often more robust in networks with a larger average degree. The study of the impact of non-random errors on the robustness suggests that centrality measures are often more robust if missing nodes are more likely to belong to the same community compared to missingness completely at random. For the development of imputation procedures based on machine learning techniques, a process for the evaluation of node embedding methods is proposed.

The world currently faces important issues concerning climate change and environmental sustainability, with the wellbeing of billions of people around the world at risk over the next decades. Existing institutions no longer appear to be sufficiently capable to deal with the complexity and uncertainty associated with the wicked problem of sustainability. Achieving the required sustainability transformation will thus require purposeful reform of existing institutional frameworks. However, existing research on the governance of sustainability of sustainability transformations has strongly focused on innovation and the more ‘creative’ aspects of these processes, blinding our view to the fact that they go hand with the failure, decline or dismantling of institutions that are no longer considered functional or desirable. This doctoral dissertation thus seeks to better understand how institutional failure and decline can contribute productively to sustainability transformations and how such dynamics in institutional arrangements can serve to restructure existing institutional systems. A systematic review of the conceptual literature served to provide a concise synthesis of the research on ‘failure’ and ‘decline’ in the institutional literature, providing important first insights into their potentially productive functions. This was followed up by an archetype analysis of the productive functions of failure and decline, drawing on a wide range of literatures. This research identified five archetypical pathways: (1) crises triggering institutional adaptations toward sustainability, (2) systematic learning from failure and breakdown, (3) the purposeful destabilisation of unsustainable institutions, (4) making a virtue of inevitable decline, and (5) active and reflective decision making in the face of decline instead of leaving it to chance. Empirical case studies looking at the German energy transition and efforts to phase out coal in the Powering Past Coal Alliance served to provide more insights on (a) how to effectively harness ‘windows of opportunity’ for change, and (b) the governance mechanisms used by governments to actively remove institutions. Results indicate that the lock-in of existing technologies, regulations and practices can throw up important obstacles for sustainability transformations. The intentional or unintentional destabilisation of the status quo may thus be required to enable healthy renewal within a system. This process required active and reflective management to avoid the irreversible loss of desirable institutional elements. Instruments such as ‘sunset clauses’ and ‘experimental legislation’ may serve as important tools to learn through ‘trial and error’, whilst limiting the possible damage done by failure. Focusing on the subject of scale, this analysis finds that the level at which failure occurs is likely to determine the degree of change that can be achieved. Failures at the policy-level are most likely to merely lead to changes to the tools and instruments used by policy makers. This research thus suggests that failures on the polity- and political level may be required to achieve transformative changes to existing power structures, belief-systems and paradigms. Finally, this research briefly touches on the role of actor and agency in the governance of sustainabilitytransformations through failure and decline. It finds that actors may play an important role in causing a system or one of its elements to fail and in shaping the way events are come to be perceived. Drawing on the findings of this research, this dissertation suggests a number of lessons policy makers and others seeking to revisit existing institutional arrangements may want to take into account. Actors should be prepared to harness the potential associated with failure and decline, preserve those institutional elements considered important, and take care to manage the tension between the need for ‘quick fixes’ to currently pressing problems and solution that maintain and protect the longterm sustainability of a system.

‚Reallabore‘ erleben als junges Format transformativer Nachhaltigkeitsforschung gegenwärtig eine beeindruckende Konjunktur – ohne das bislang hinreichend geklärt ist, was sie konzeptionell Neues bieten. Die Dissertation arbeitet den Reallabor‐Ansatz aus Perspektive der transdisziplinären Forschung methodisch aus. Die Basis hierfür bildet die Erfahrung mit dem Auf‐ und Ausbau von einem der ersten Reallabore in Deutschland: Das langfristig ausgelegte „Quartier Zukunft – Labor Stadt“ in Karlsruhe transformiert in Kooperation mit der Zivilgesellschaft ein Quartier modellhaft in einen nachhaltigeren Lebensraum. Es setzt dabei gleichermaßen auf Bildung, Forschung und Praxis. Die vorgelegten Texte der kumulativen Dissertation bilden verschiedene Stadien der Entwicklung der Reallaborforschung und der methodologischen Reflexion ab. Die ersten beiden Texte entwickeln eine praxisnahe Definition und ordnen Reallabore ein in verwandte Diskurse. Die folgenden beiden Texte stammen aus der beginnenden Stabilisierung des Reallabordiskurses. Der eine stellt Ziele und Designprinzipien für Reallabore als Rahmen transformativer und transdisziplinärer Forschung dar, der zweite greift aktuelle Diskussionen um Lernprozesse konzeptionell auf. Die letzten beiden Texte fokussieren auf die Ebene der Projekte im Reallabor am Beispiel der Transformativen Projektseminare, einmal in analytischer Perspektive, einmal in methodisch‐didaktischer. Der Rahmentext abstrahiert die Ergebnisse der zuvor publizierten Texte entlang dreier Forschungsfragen und integriert sie zu einem Konzeptmodell transdisziplinärer Forschung im Reallabor, dem „Apfelmodell“. Auf Basis von Diskursen zu Transdisziplinarität, Nachhaltigkeitswissenschaften, Bildungstheorie und Didaktik sowie zu Laboren mit sozialwissenschaftlicher oder interdisziplinärer Ausrichtung werden drei Forschungsfragen verfolgt: Was ist neu am Reallabor‐Ansatz? Welches Potenzial hat ein Reallabor für transdisziplinäre Forschung? Und welche Rolle spielt Lernen im Reallabor? Die methodologische Reflexion führt zu einem Verständnis von Reallaboren als Format zwischen Urban Living Labs und Transition Labs, das sich gegenüber diesen insbesondere durch Langfristigkeit, Bildungsziele und eine klare Trennung zwischen Labor und Experimenten auszeichnet. Aus der kritischen Auseinandersetzung mit Reallaboren wird eine doppelte Bezugnahme auf Transdisziplinarität herausgearbeitet, einerseits als Infrastruktur für transdisziplinäre Projekte, andererseits als in sich transdisziplinäres Unterfangen. Ausgehend von dieser Unterscheidung wird ein Vorschlag gemacht, an welche experimentellen Methodologien jenseits der klassisch‐naturwissenschaftlichen die transdisziplinäre Forschung, die bislang kaum experimentell arbeitet, anknüpfen kann. Das Reallabor unterstützt solche Experimente durch einen Rahmen aus materieller Infrastruktur, durch Kompetenzen der Beteiligten, durch Wissensbestände und soziale Vernetzung. Die Vernetzung über Projektgrenzen hinweg, ein weiteres wesentliches Charakteristikum eines Reallabors, dient dazu, parallele Experimente zu vernetzen und iterative Lernzyklen zu unterstützen. Diese Aspekte werden verbunden zum „Apfelmodell“ transdisziplinärer Forschung im Reallabor, in dem das Reallabor als doppeltes Bindeglied fungiert, einerseits zwischen internen und externen Lernzyklen, und andererseits zwischen wissenschaftlichen, bildungsorientierten und praktischen. Durch die Interpretation der Abläufe im Reallabor als Lernprozesse wird ein Anschluss an Bildungsprozesse auf unterschiedlichen Skalen möglich. Neben Lernprozessen im Reallabor als Lernumgebung lässt sich das Reallabor als lernende Institution und als Kristallisationspunkt gesellschaftlicher Lernprozesse verstehen. Das Apfelmodell kann gleichermaßen im Kontext theoretischer Fragen im Transdisziplinaritätsdiskurs herangezogen werden als auch praktischen Zwecken dienen, insbesondere in der Planung von Reallaboren, in der quervernetzten Konzeption von Projekten darin, in der Evaluation und in der Kommunikation.

My dissertation embraces four empirical papers addressing socio-economic issues relevant to policy-makers and society as a whole. These papers cover important aspects of human life including health at birth, life satisfaction, unemployment periods and retirement decisions, and are intended to provide a contribution to the respective research areas. The analyses are carried out applying advanced econometric methods and are based on data sets consisting of survey data as well as administrative records. The joint paper with Alessandro Palma and Daniela Vuri "Prenatal Air Pollution Exposure and Neonatal Health" in Chapter 2 investigates the causal impact of prenatal exposure to air pollution on neonatal health in Italy in the 2000s combining detailed information on mother’s residential location from birth certificates with PM10 concentrations from air pollution monitors. Variation in local weekly rainfall is exploited as an instrumental variable for non-random air pollution exposure. Using quasi-experimental variation in rainfall shocks allows to identify the effect of PM10, ruling out potential bias due to confounder pollutants. The paper estimates the effect of exposure for both the entire pregnancy period and separately for each trimester to test whether the neonatal health effects are driven by pollution exposure during a particular gestation period. This information enhances our understanding of the mechanisms at work and help prevent pregnant mothers from most dangerous exposure periods. Additionally, the effects of prenatal exposure to PM10 are estimated by maternal labor market status and maternal education level to understand how the pollution burden is shared across different population groups. This decomposition allows to identify possible mechanisms through which environmental inequality reinforces the negative impact of early-life exposure to air pollution. This study finds that average PM10 and days with PM10 level above the hazard limit reduce birth weight, gestational age, and measures of overall newborn health. Effects are largest for third trimester exposure and for low-income and less educated mothers. These findings imply that further policy efforts are needed to fully protect fetuses from the adverse effects of air pollution and to mitigate the environmental inequality of health at birth. The joint paper with Christian Pfeifer "Life Satisfaction in Germany After Reunification: Additional Insights on the Pattern of Convergence" in Chapter 3 updates previous findings on the total East-West gap in overall life satisfaction and its trend by using data from the German Socio-Economic Panel for the years 1992 to 2013. Additionally, the effects are separately analyzed for men and women as well as for four birth cohorts. The results indicate that reported life satisfaction is, on average, significantly lower in East than in West German federal states and that part of the raw East-West gap is due to differences in household income and unemployment status. The conditional East-West gap decreased in the first years after the German reunification and remained quite stable and sizable since the mid-nineties. The results further indicate that gender differences are small. Finally, the East-West gap is significantly smaller and shows a trend towards convergence for younger birth cohorts. The joint paper with Christian Pfeifer "Unemployment Benefits Duration and Labor Market Outcomes: Evidence from a Natural Experiment in Germany" in Chapter 4 explores the effects 2 of a major reform of unemployment benefits in Germany on the labor market outcomes of individuals with some health impairment. The reform induced a substantial reduction in the potential duration of regular unemployment benefits for older workers. This work analyzes the reform in a wider framework of institutional interactions, which allows to distinguish between its intended and unintended effects. The results based on routine data collected by the German Statutory Pension Insurance and a Difference-in-Differences design provide causal evidence for a significant decrease in the number of days in unemployment benefits and increase in the number of days in employment. However, they also suggest a significant increase in the number of days in unemployment assistance, granted upon exhaustion of unemployment benefits. Transitions to unemployment assistance represent an unintended effect, limiting the success of a policy change that aims to increase labor supply via reductions in the generosity of the unemployment insurance system. The single-authored paper "How Older Workers Respond to Raised Early Retirement Age: Evidence from a Kink Design in Germany" in Chapter 5 explores how an increase in the early retirement age affects labor force participation of older workers. The analysis is based on a social security reform in Germany, which raised the early retirement age over several birth cohorts to boost employment of older people and ultimately alleviate the burden on the public pension system. Detailed administrative data from the Federal Employment Agency allow to distinguish between employment and unemployment as well as disability pensions and retirement benefits claims. Using a Regression Kink design in a quasi-experimental framework, I show that the raised early retirement age had positive employment effects and negative effects on retirement benefits claims. The reform did not affect unemployment benefits or disability pensions claims. My results also show that some population groups are more sensitive to a reduction in retirement options and more likely to seek benefits from other government programs. In this respect, I find that workers in manufacturing sector respond to the raised early retirement age by claiming benefits from the disability insurance program designed to compensate for reduced earnings capacity due to severe health problems. The treatment heterogeneity analysis further suggests that high-wage workers are more likely to delay exits from employment, which is in line with incentives but might also indicate an increased inequality within the affected birth cohorts induced by the reform. Finally, women seem to rely on alternative sources of income such as retirement benefits for women, or spouse's or partner's income not observed in the data. All things considered, workers did not adjust to the increased early retirement age by substituting early retirement with other government programs but rather responded to the reform in line with the policy intent. At the same time, the findings point to heterogeneous behavioral responses across different population groups. This implies that raising the early retirement age is an effective policy tool to increase employment only among older people who have the real choice to delay employment exits. Therefore, reforms that raise statutory ages should ensure social support for workers only marginally attached to the labor market or not able to work longer due to potential health problems or other circumstances.