Experimentalists in political science often face the question of how abstract or concrete their experimental stimuli should be. Typically, this question is framed in terms of tradeoffs relating to experimental control and generalizability: the more context you introduce into your studies, the less control you have, and the more difficulty you have generalizing your results. Yet we have reasons to question this framing of the tradeoff, and there is relatively little systematic evidence experimenters can rely on when calibrating the degree of abstraction in their studies. We seek to make two contributions with this project. First, we provide a theoretical framework which identifies and considers the consequences of three dimensions of abstraction in experimental design: situational hypotheticality, actor identity, and contextual detail. Second, we replicate a range of classic vignette-based survey experiments from political science, varying these levels of abstraction. Our results suggest that apart from a specific set of conditions, there are fewer tradeoffs between abstraction and detail in survey experiment design than political scientists often assume.

Broadening access to both computational and educational resources is critical
to diffusing machine-learning (ML) innovation. However, today, most ML resources and experts are siloed in a few countries and organizations. In this paper, we describe our pedagogical approach to increasing access to applied ML through a massive open online course (MOOC) on Tiny Machine Learning (TinyML). We suggest that TinyML, applied ML on resource-constrained embedded devices, is an attractive means to widen access because TinyML both leverages low-cost and globally accessible hardware, and encourages the development of complete, self-contained applications, from data collection to deployment. To this end, a collaboration between academia and industry produced a four-part MOOC that provides application-oriented instruction on how to develop solutions using TinyML. The series is openly available on the edX MOOC platform, has no prerequisites beyond basic programming, and is designed for global learners from a variety of backgrounds. It introduces pupils to real-world applications, ML algorithms, data-set engineering, and the ethical considerations of these technologies through hands-on programming and deployment of TinyML applications in both the cloud and on their own microcontrollers. To facilitate continued learning, community building, and collaboration beyond the courses, we launched a standalone website, a forum, a chat, and an optional course-project competition. We also open-sourced the course materials, hoping they will inspire the next generation of ML practitioners and educators and further broaden access to cutting-edge ML technologies.

How does naming and shaming affect public support for compliance with international agreements? We investigated this question by conducting survey experiments about the Paris Agreement, which relies on social pressure for enforcement. Our experiments, administered to national samples in the United States, produced three sets of findings. First, shaming by foreign countries shifted domestic public opinion in favor of compliance, increasing the political incentive to honor the Paris Agreement. Second, the effects of shaming varied with the behavior of the target. Shaming was more effective against partial compliers than against targets that took no action or honored their obligations completely. Moreover, even partial compliers managed to reduce the effects of shaming through the strategic use of counter-rhetoric. Third, identity moderated responses to shaming. Shaming by allies was not significantly more effective than shaming by non-allies, but Democrats were more receptive to shaming than Republicans. Overall, our experiments expose both the power and the limits of shaming as a strategy for enforcing the Paris Agreement. At the same time, they advance our understanding of the most significant environmental problem facing the planet.