http://hdl.handle.net/10900/42132 2026-11-10T05:06:38Z http://hdl.handle.net/10900/185300 Beyond Matching: Designing, Evaluating, and Orchestrating Adaptive Instructional Technologies in Authentic Classrooms Wellert, Luisa Die Dissertation ist gesperrt bis zum 21. Oktober 2028 !; Classrooms are becoming increasingly heterogeneous, and large-scale educational assessments document students’ declining performance in key competence domains. Adapting instructional offers to match students’ heterogeneous learning potentials (e.g., prior knowledge, motivation, language proficiency) and thereby optimally supporting students’ learning poses a substantial challenge for teachers. Digital technologies, particularly intelligent tutoring systems (ITS) and Large Language Model (LLM)-based tutors, provide new opportunities to support teachers in adaptive instruction. Despite empirical evidence for the effectiveness of adaptive instructional technologies for learning, the widespread adoption in schools remains limited. This can be attributed to three main challenges: the complex development and limited contextual transferability of conventional ITS, the resource-intensive evaluation of large volumes of dialogue data produced by LLM tutors, and the complex embedding and orchestration of adaptive instructional technologies in real-world classrooms. The present dissertation addresses these challenges through four studies adopting diverse methodological and interdisciplinary perspectives, with a focus on proximity to school practice. Study 1 examined the development and effectiveness of technology-based adaptive practice systems. Co-design teams of researchers and teachers developed adaptive practice systems with simplified underlying assessment strategies and investigated their effectiveness in a ManyClasses field study across different school contexts. Although adaptation itself did not improve students’ learning outcomes, working on more high-difficulty tasks did show positive effects. These findings raised the question of how adaptive instructional technologies can be designed to provide instructional offers that match students’ learning potentials in a way that fosters students’ learning. Hence, Study 2a turned to LLM-tutors, which can be prompted with evidence-based instructional strategies and may offer a scalable alternative to rule-based ITS. Study 2a, an exploratory online study with university students, compared two versions of an LLM-tutor: one using a baseline tutor prompt and one using a prompt specifying evidence-based tutoring strategies. An epistemic network analysis (ENA) showed that evidence-based prompting led to more frequent cognitively activating tutoring behaviors, while the baseline tutor version received more positive ratings for satisfaction and acceptance. Study 2b emerged from Study 2a as a methodological contribution. The Theory Prompting Validation (TPV) approach was developed to enable LLM-assisted analysis of large volumes of verbal data. Evaluation of the dialogue data from Study 2a showed high agreement between LLM-based and human-based coding. The approach thus opens possibilities for the scalable, reliable, and valid analysis of verbal data beyond the tutoring context. Because adaptive instructional technologies can unfold their potential only when embedded in real-world classroom contexts, Study 3 examined how teachers orchestrate them. This exploratory online study with mathematics teachers aimed to identify criteria that teachers use to decide which students to support first when a mixed-reality system displays the learning states of students working in an ITS. The results revealed similar heuristics that reflected a balance between immediate support needs, student engagement, and preventing individual students from falling behind. This dissertation makes a comprehensive contribution to the design, evaluation, and orchestration of adaptive instructional technologies. For theory on adaptive instruction, it suggests that adaptivity should not be conceptualized merely as matching instructional offers to learners’ current learning potentials, but as using this information to provide challenging yet attainable instructional offers that support learners in reaching their next level of development. For practice, this dissertation highlights the importance of close collaboration between research and practice for the design of adaptive instructional technologies, the need for teachers to critically evaluate the adaptive instructional technologies they select, and the necessity of preparing teachers to orchestrate adaptive instructional technologies in classrooms. Methodologically, the TPV approach opens new possibilities for the scalable analysis of extensive verbal data beyond the tutoring context. Taken as a whole, the dissertation combines technological innovation with empirical evidence and a practice-oriented perspective, providing a foundation for developing adaptive instructional offers that better respond to heterogeneous learning potentials and thereby contribute to more equitable educational opportunities for all students.; Klassenräume werden zunehmend heterogener und großangelegte Schulleistungsstudien verzeichnen einen Rückgang der Schüler:innenleistungen in zentralen Kompetenzbereichen. Instruktionale Angebote an die heterogenen Lernpotenziale der Schüler:innen (z. B. Vorwissen, Motivation, Sprachkompetenz) anzupassen und sie dadurch optimal in ihrem Lernen zu unterstützen, stellt eine erhebliche Herausforderung für Lehrkräfte dar. Digitale Technologien, insbesondere intelligente tutorielle Systeme (ITS) und auf Large Language Models (LLMs) basierende Tutoren, eröffnen neue Möglichkeiten, Lehrkräfte im adaptiven Unterricht zu unterstützen. Trotz empirischer Evidenz für die Wirksamkeit adaptiver instruktionaler Technologien für das Lernen bleibt ihr breiter Einsatz in Schulen bislang begrenzt. Dies lässt sich auf drei zentrale Herausforderungen zurückführen: die komplexe Entwicklung und begrenzte kontextuelle Übertragbarkeit herkömmlicher ITS, die ressourcenintensive Auswertung umfangreicher Dialogdaten, die von LLM-Tutoren erzeugt werden, sowie die komplexe Einbettung und Orchestrierung adaptiver instruktionaler Technologien im realen Unterricht. Die vorliegende Dissertation untersucht diese Herausforderungen in vier Studien aus unterschiedlichen methodologischen und interdisziplinären Perspektiven mit Fokus auf die Nähe zur Schulpraxis. Studie 1 untersuchte die Entwicklung und Wirksamkeit technologiebasierter adaptiver Übungssysteme. Co-Design-Teams aus Forschenden und Lehrkräften entwickelten adaptive Übungssysteme mit vereinfachten zugrunde liegenden Diagnostikstrategien und untersuchten deren Wirksamkeit in einer ManyClasses-Feldstudie über verschiedene Schulkontexte hinweg. Während sich Adaptivität selbst nicht positiv auf die Lernergebnisse der Schüler:innen auswirkte, zeigte die Bearbeitung anspruchsvollerer Aufgaben jedoch eine positive Auswirkung. Diese Ergebnisse warfen die Frage auf, wie adaptive instruktionale Technologien gestaltet werden können, um instruktionale Angebote bereitzustellen, die an die Lernpotenziale der Schüler:innen angepasst sind und zugleich deren Lernen fördern. Vor diesem Hintergrund wandte sich Studie 2a LLM-Tutoren zu, die mit evidenzbasierten instruktionalen Strategien geprompted werden können und somit möglicherweise eine skalierbare Alternative zu regelbasierten ITS darstellen. Studie 2a, eine explorative Online-Studie mit Studierenden, verglich zwei Versionen eines LLM-Tutors: eine mit einem allgemeinen Tutor-Prompt und eine mit einem Prompt, der evidenzbasierte Tutoringstrategien spezifizierte. Eine epistemische Netzwerkanalyse (ENA) zeigte, dass evidenzbasiertes Prompting zu häufigeren kognitiv aktivierenden Tutoring-Verhaltensweisen führte, während die allgemeine Tutorversion hinsichtlich Zufriedenheit und Akzeptanz positiver bewertet wurde. Aus Studie 2a entwickelte sich Studie 2b als methodischer Beitrag. Mit dem Theory-Prompting-Validation (TPV)-Ansatz wurde ein Verfahren entwickelt, das die LLM-gestützte Analyse umfangreicher verbaler Daten ermöglicht. Die Evaluation der Dialogdaten aus Studie 2a zeigte eine hohe Übereinstimmung zwischen LLM-basierter und menschlicher Codierung. Der Ansatz eröffnet damit Möglichkeiten für eine skalierbare, reliable und valide Analyse verbaler Daten auch über den Tutoring-Kontext hinaus. Da adaptive instruktionale Technologien ihr Potenzial nur entfalten können, wenn sie in reale Unterrichtskontexte eingebettet sind, untersuchte Studie 3, wie Lehrkräfte sie orchestrieren. Diese explorative Online-Studie mit Mathematiklehrkräften zielte darauf ab, Kriterien zu identifizieren, nach denen Lehrkräfte entscheiden, welche Schüler:innen sie zuerst unterstützen, wenn ein Mixed-Reality-System den Lernstatus von Schüler:innen anzeigt, die in einem ITS arbeiten. Die Ergebnisse zeigten ähnliche Heuristiken, die eine Balance zwischen unmittelbarem Unterstützungsbedarf, Schüler:innenengagement und der Vermeidung individueller Lernrückstände widerspiegelten. Diese Dissertation leistet einen umfassenden Beitrag zur Gestaltung, Evaluation und Orchestrierung adaptiver instruktionaler Technologien. Für die Theorie des adaptiven Unterrichts legt sie nahe, dass Adaptivität nicht lediglich als Anpassung instruktionaler Angebote an die aktuellen Lernpotenziale der Lernenden verstanden werden sollte, sondern als Nutzung dieser Information, um herausfordernde, aber erreichbare instruktionale Angebote bereitzustellen, die Lernende dabei unterstützen, ihre nächste Entwicklungsstufe zu erreichen. Für die Praxis verdeutlicht diese Dissertation die Bedeutung einer engen Zusammenarbeit zwischen Forschung und Praxis bei der Gestaltung adaptiver instruktionaler Technologien, die Notwendigkeit, dass Lehrkräfte die von ihnen ausgewählten adaptiven instruktionalen Technologien kritisch bewerten, sowie die Notwendigkeit, Lehrkräfte auf die Orchestrierung adaptiver instruktionaler Technologien im Unterricht vorzubereiten. Methodisch eröffnet der TPV-Ansatz neue Möglichkeiten für die skalierbare Analyse umfangreicher verbaler Daten über den Tutoring-Kontext hinaus. Insgesamt verbindet die Dissertation technologische Innovation mit empirischer Evidenz und einer praxisorientierten Perspektive und schafft damit eine Grundlage für die Entwicklung adaptiver instruktionaler Angebote, die besser auf heterogene Lernpotenziale reagieren und dadurch zu gerechteren Bildungschancen für alle Schüler:innen beitragen. 2028-10-21T00:00:00Z http://hdl.handle.net/10900/184616 Technology, Politics, and A Tale of Two Outcomes: How Digitization Weakens Democracy and Strengthens Autocracy Sirikupt, Chonlawit There is an emerging scholarly consensus that unprecedented access to information and lower barriers to political participation, ushered in by digitization and digital technologies, have not strengthened democracy. From mob lynchings in India to the 2021 US Capitol insurrections, social media recommendation algorithms have played a role in amplifying conspiracy theories, disinformation, and propaganda. Outside liberal democracies, the use of digital technologies in autocratic regimes has come to be seen as a mixed blessing. While digital technologies have assisted democratic activism in overthrowing dictators in Tunisia and Egypt, they have also evolved into powerful tools for regime entrenchment by strengthening, among others, state surveillance and censorship capabilities. How do we explain these outcomes, that is, the weakening of democracy and the strengthening of autocracy amid the digital changes taking place across political regimes? This cumulative dissertation locates these trends that belie the initial euphoria about the democratic potential of digital technologies in the center of a largely underregulated global digital sphere and socio-technical designs that leave internet users vulnerable to mis- and disinformation, as well as online harassment and abuse. It advances understandings about these patterns by weaving feature-based, regime-centric, and micro-level approaches into a unified synoptic framework that sheds a closer and more nuanced light on different dimensions of digitally-induced changes that drive the outcomes of my interest. Each approach inspired different theoretical lenses and cases employed to subject my/our arguments to empirical scrutiny. The first article offers a new theoretical lens on how attitudinal and behavioral effects associated with the three key features of digital technologies, namely, the abundance of contradictory information, the decreasing ability of consumers to assess its truthfulness, and AI-assisted algorithmic targeting of users with personalized content, undermine democracy as a regime type. An empirical probe into the two case studies of COVID-19 demonstrations in Germany and the US Capitol attack in 2021 demonstrates that digital technologies can foster misbelief and, subsequently, anti-democratic behavior or agnosticism, which consequently translates into political apathy. The second article theoretically situates digital astroturfing at the center of autocratic regime threat perception and adopts a mixed-methods approach to granularly examine 14,443 tweets linked to the Thai military in February 2020. The findings demonstrate that astroturfing constitutes an important legitimation tool – one that autocrats employ in response to offline events that spark online discourses with collective action potential. The third article goes beyond (and below) the regime-centric paradigm, examining how state-aligned countermovement actors support repertoires of repression. Leveraging a diachronic comparison of protests during Bahrain’s 2011 Pearl Roundabout uprising and Myanmar’s 2021 military coup, it shows that prominent state-aligned social media users exploited social media affordances to document posts with anti-regime leanings, publicize identities of dissident posters, and call on authorities to arrest the latter. Altogether, this cumulative dissertation contextualizes these dynamics within broader comparative political regimes that shape the participation and inclusion (or exclusion) of their citizens. While digital technologies amplify socio-political processes that weaken democracy’s institutional guardrails, their use in autocracies has further contributed to the deepening of authoritarian traits in these regimes. These insights not only extend the frontiers of digitally-adjacent political science research but also signal the need to incorporate more technical understandings of current and emerging digital tools into the research agenda amid the ever-increasing pervasiveness of digital capital in authoritarianism research. 2026-10-22T00:00:00Z http://hdl.handle.net/10900/184346 Gender Equality in Health: Its Measurement, Determinants, and Long-Run Evolution Jung, Sophia Gender inequality remains one of the most persistent and consequential forms of social disparity worldwide, yet its long-run development and determinants are poorly understood. Existing gender equality indices are often limited to contemporary data and depend on administrative sources that are unavailable for most historical and pre-literate societies. This dissertation develops and applies a long-run measure of gender equality in health using adult stature as a proxy. Height reflects cumulative net nutrition and disease exposure during early childhood. After isolating the biological component from the height gap between males and females, the remaining variation captures gender-discriminatory differences in intrahousehold resource allocation and can be used as a measure for gender equality in health. This measure is validated through comparisons with the Gender Development Index and relative life expectancies and applied to three different historical and geographic contexts to analyze the development and determinants of gender equality in health. The first study addresses whether socialism improved gender equality using a sample of 88 countries in Europe and Asia for the period from the 1910s to the 1980s. Regression analysis reveals a significant and positive relationship between socialist governance and gender equality in height. The results are paralleled by improvements in gender equality in education. Robustness checks using staggered difference-in-differences estimation and instrumental variable estimation, using distance to Moscow as an instrument, confirm the positive effect. The second study analyses the long-run development of gender equality in health and the role of subsistence systems relying on male and female height estimates for precolonial America spanning several thousand years. Two-way fixed effects regressions indicate that subsistence systems influenced the development of gender equality. A specialization in maize cultivation, compared to mixed agriculture, had an initially positive relationship with gender equality; in the later period, when population pressure intensified competition for resources within households, the relationship turned negative. Similar trends are observed for hunting. Overall, the results suggest that women’s contribution to subsistence mattered for the development of gender equality. The third study examines how European settlement reshaped gender equality in health among Indigenous populations in the Americas. The dataset from the previous study is expanded to the post-contact period and restricted to the last 2,000 years, combining bioarchaeological evidence with historical anthropometric sources. Exploiting variation in settlement timing and location, two-way fixed effects regressions indicate that European settlements negatively influenced gender equality among Indigenous populations. The results are robust to controlling for subsistence systems, climate, land use, and population density. The relationship is heterogeneous across colonial activity types and institutional environments, consistent with differential disruptions to Indigenous access to land and labor. 2026-10-15T00:00:00Z http://hdl.handle.net/10900/184114 Belief dynamics in sport: Exploring economic and behavioral consequences of suspense, surprise, and expectation violations Richardson, Travis William Sport is an entertainment product, and a central premise in sports economics holds that outcome uncertainty drives its appeal: spectators are drawn to contests whose results are in doubt. Yet despite decades of investigation, the evidence for this relationship remains strikingly inconsistent, particularly in European football. This dissertation traces part of that inconsistency to how uncertainty is measured. Conventional measures are static, capturing pre-match conditions while saying little about how spectators experience a game moment by moment, and they bundle together two distinct mechanisms of entertainment utility that respond to uncertainty in opposite ways. The dissertation instead builds on belief dynamics, the processes through which spectators form, update, and react to beliefs about a contest as it unfolds. Drawing on the framework of Ely, Frankel, and Kamenica (2015), it separates suspense, the forward-looking anticipation of an unresolved outcome, from surprise, the backward-looking jolt when events depart from expectations. It is the conflation of these two mechanisms that lies behind the inconsistent empirical record. Across three empirical studies, the dissertation examines how belief dynamics shape consumer demand and behavior at three levels: within the live event, across competitions and seasons, and beyond the stadium. The first study analyzes minute-by-minute television audience data for UEFA Champions League matches in the United Kingdom and Spain. Suspense and surprise emerge as the main drivers of real-time demand, and their relative weight varies across the two markets. The second study develops match-level suspense and surprise into diagnostic tools for competition attractiveness, deriving a benchmark through simulation and applying it to more than 25,000 matches across Europe's top five leagues. It finds no uniform decline in competitive balance, but heterogeneous trends over time often driven by a small number of dominant clubs, patterns that conventional measures overlook. The third study turns to behavior beyond sport, using fine-grained vehicle speed data around Florida stadiums. Game outcomes that violate pre-game expectations leave a measurable imprint on post-game driving, with the strongest effect following close games that end in disappointing home-team losses. Together, the three studies establish suspense, surprise, and expectation violations as empirically consequential forces, whose reach extends from a single minute of a single match, to the attractiveness of entire competitions, and into everyday behavior beyond the sporting event itself. 2026-10-07T00:00:00Z