Overview
On November 1 I began officially as a Lecturer in the Stanford Doerr School of Sustainability at Stanford University. I am developing two courses. The first, starting January 1 next year, is Computational Social Science for Sustainability. The second is Agent-based modeling. So far I have a syllabus for Computational Social Science. I have plans for programming problem sets, lecture material, and I’m excited to see what students pursue for their cumulative midterm and final projects.
I was hired to create a course that teaches the theory and techniques of computational social science to support sustainability with minimum technical and theoretical overhead. That means strategic choices of subject matter, and how it should be presented. Personally I think it is fun and valuable to remix and resample standard ways of introducing material. In this spirit, social interactions will be introduced as occurring in the substrate of social networks. From the first problem set, students will learn to construct, manipulate, and visualize social networks. We will spend the ten-week course studying complementary approaches to developing mechanistic models of emergent social phenomena from repeated social interactions over time. Emergent social phenomena include, for example, viral adoption of sustainable behaviors, political polarization, and sustainable groundwater management and irrigation, or management of other common-pool resources.
We identify and model three types of social interaction: social learning, social influence, and cooperation. Social interactions of interest may include one or more of these types of interaction. See the course syllabus for more details on specific theoretical components of social science I plan to use to model social learning, social influence, and cooperation. Broadly, social learning is informed by cultural evolutionary studies, social influence is framed in terms of opinion dynamics studies, and cooperation is based on evolutionary game theory as its starting point. Social phenomena and constituent psychological and cognitive processes
The course: rigorous, computer-literate social science for sustainability
Sustainability, and similar challenges like pandemic preparedness and prevention, require behavioral science to promote better outcomes. This need provides a selection pressure towards a trim presentation of social science essentials. Computation is essential for social science because we can build simulated groups and societies to perform computational experiments that would be difficult or impossible to perform in the real world. Nonetheless, these simulations can demonstrate fundamental truths in model systems with clear implications for the real world. For example, we can understand exactly how well information percolates through different model social networks, so that we may predict how many people must learn about some sustainable behavior in order for the behavior to go viral, i.e., spread widely to all members of a society.
Social science is one branch of behavioral science, complemented and informed by psychological and cognitive sciences and linguistics. These fields, in various ways, exist to understand how language works and how individuals are influenced differently by different communication strategies. The goal of social science here is to understand how repeated social interactions aggregate into emergent society-scale outcomes. For example, the emergent outcome of how many farmers adopt more carbon-friendly agricultural practices depends on a series of interactions with others such as learning from crop advisers and hearing the opinions of fellow farmers. Computational social science provides theoretical and modeling tools for representing repeated social interactions structured by social networks. Computational social science adds to its knowledge base with new analyses of social simulations based on empirically-motivated assumptions about social interactions work, structured by social networks and subject to uncertainty and stochasticity. Computational social science also produces new knowledge through statistical analysis of empirical data made rigorous by thoroughly accounting for sources of variance. In summary, this course will enable students to create their own analyses of societal change to contribute to a scientific understanding of how to scale up adoption of sustainable practices.
Many students are attracted to sustainability because they care about the planet and its inhabitants. We would do well to welcome and expect sustainability students to have a spectrum of programming experience and interest. To this end, I have begun developing an R library, socmod that will pull together essential computational social science tools, models, and data. The skills students will gain in R are highly transferable. R use is widespread, I belive thanks to the stylish leadership of Hadley Wickham and associates. We will use R Shiny and tidyverse functionality including dplyr and ggplot for analyzing and visualizing model outcomes. I plan to follow the approach of Agents.jl in the functional approach to defining agent-agent interactions and model dynamics for each step, which are then scheduled and run by library code.
This approach to modeling will require the combination of several tools. First, object-oriented programming is most sensible for agent-based models, which will be a primary modeling method. Oftentimes a model, agent-based or otherwise, will need to be run several times over several parameter settings. This quite naturally suggests that Models, too, should be objects. Because R by default assigns by value instead of by reference, and because R is traditionally functional and declarative more than procedural, explicit object-oriented coding is relatively rare. For object-oriented programming I have chosen the R6 package, based on advice from Hadley Wickham’s book, Advanced R. In Chapter 14 on R6, Wickham notes that R6 “uses the encapsulated OOP paradigm, which means that” you call methods using the infix $ operator, calling “them like object$method()”.
It is possible to maintain information about simulated individuals in tabular format, as Acerbi, Mesoudi, and Smolla (2022) did wonderfully in their R-based approach to agent-based modeling (they call it individual-based, but they are identical). However a more declarative, object-oriented approach is easier for beginners, as testified to by the choice in Smaldino (2023) to use NetLogo. NetLogo is highly idiosyncratic, used almost exclusively for pedagogical purposes, and so may have relatively little transferrable value in terms of programming experience. However, there is nothing currently available that makes agent-based computational social science modeling more accessible. If the software library associated with this course succeeds, it may be accessible, help students learn transferrable skills in R and object-oriented programming, and grow into a powerful research tool for computational social science in R.
In making any sort of publicly-available library it’s necessary to make it easy to install and especially for students. It would be a waste of time to rush and accumulate technical debt like poorly-documented code or making code excessively difficult to install. Therefore, I am pursuing a document-driven development approach, where one of the documents are the problem sets themselves. I am using pkgdown for library documentation, and devtools for package development and distribution.
Conclusion
While discussing the course Computational Social Science for Sustainability with the practicum and curriculum manager for a master’s program in the Doerr School of Sustainability, she asked me if the course should be categorized as a science course or an arts course. I briskly replied, “Science.” I understand and deeply sympathize with the question, even though to me the answer is obvious. The confusion may come from the fact that when grade school students study “social science” in the United States, it is really just “history”, maybe with some political science and civics. Furthermore, I am aware that some sociologists and anthropologists, for example, are uncomfortable with a “science” of human behavior because of historical connections with colonialism or colonial powers, and possibly due to implications that humans lack free will. It seems my job is to make it obvious to my audience that Computational social science for sustainability is unmistakeably science just like physics, chemistry, or biology. I believe this science can help steward collective behavior (Bak-Coleman et al. 2021) for a more sustainable world, which to me seems humanitarian enough to overcome potential concerns.
More updates will follow as I create problem sets and continue writing course notes and build the socmod R library to accelerate student learning.
References
Acerbi, Alberto, Alex Mesoudi, and Marco Smolla. 2022.
Individual-based models of cultural evolution: A step-by-step guide using R. London: Routledge.
https://acerbialberto.com/IBM-cultevo/.
Bak-Coleman, Joseph B., Mark Alfano, Wolfram Barfuss, Carl T. Bergstrom, Miguel A. Centeno, Iain D. Couzin, Jonathan F. Donges, et al. 2021.
“Stewardship of global collective behavior.” Proceedings of the National Academy of Sciences of the United States of America 118 (27): 1–10.
https://doi.org/10.1073/pnas.2025764118.
Smaldino, Paul E. 2023. Modeling Social Behavior: Mathematical and Agent-based Models of Social Dynamics and Cultural Evolution. Princeton: Princeton University Press.