computational social science

computational social science

The Impact of Computational Social Science on Understanding Human Behavior

1. Introduction to Computational Social Science

The sudden availability of big data about human actions is particularly good news for economic researchers, who have long been envious of the ability of other researchers to analyze large datasets of human behavior. There are, however, many challenges that arise in adapting the tools of computational social science to new economic research contexts. With this paper, we provide a guide for the non-specialists interested in exploring big data, summarizing the various predictive algorithms and the necessary inferential frameworks developed in the computer science literature.

Over the past decade, new databases on human behavior have emerged from digitizing traditional institutions or web-based platforms. Simultaneously, a vibrant new field of computational social science has developed and exploited a variety of social, economic, and technological big data – from sales transactions and financial reports to Twitter and detailed product review and ratings. Some have argued that numerous advances in computer science and mathematics may enable the emergence of an entirely new field of computational social science, focusing on issues that have largely been the domain of qualitative social science and areas like journalism and political science.

2. Theoretical Foundations and Methodologies in Computational Social Science

Understanding Human Behavior: A Data-Driven Analysis of Social Responses to Events from a Natural Experiment on Distributed Volunteering provides a nice example of cutting-edge work in the rapidly evolving empirical literature dealing with social contagion processes. Using several datasets taken from distributed volunteer projects (with the data often collected in real time), the authors plot the timing of computers’ decisions to volunteer during the course of short trials (lasting 30-60 seconds) in relation to the decisions made by other computers that were asked to make similar volunteering choices. Remarkably, the computers making the volunteering choices did not have to personally signal their intention to volunteer; instead, they each had their decision to volunteer recorded automatically as a by-product of using the normal operation of the volunteer-servers. This pure measure of individual action allows researchers to sidestep some of the usual worries about the role of hunches, pro-social value orientation, or a desire to avoid embarrassment when explaining why one’s actions seem to be influenced by contemporaneous social actions.

Network structure (meaning the specific ways people are interconnected) and how these structures impact individuals’ behaviors through “social contagion” processes, is one main trend in this new field. “Text as Data,” another front that has been receiving attention, involves developing techniques for analyzing large volumes of text data and then using these techniques to solve problems that matter to social science and public policy. Other prominent research fields aim to explore how the process of media creation and consumption impacts public opinion, whether and how search engines shape the access of information, and how advertising and peer-to-peer sharing contributes to the success of cultural products.

Toward a better understanding of human social behavior, the emerging field of computational social science aims to combine traditional social science and statistics with large-scale databases and computer-based methods. By crafting theoretical models and then testing those models against big data sources, computational social scientists build models regarding human behavior. Recently, new models and results from this field have created global headlines and have appeared in top social science journals.

3. Applications of Computational Social Science in Various Fields

A recent survey of economics papers that use social media data finds that the most common applications connect social media data to existing economic theory. However, a number of papers attempt to tie social media insights into behavior. This is different than other targeted uses of social media data. For example, Khandani and Kim examine whether social media data convey information about the investment or rating decisions of investors. Further overlapping with the broader domain of misuse of information, Hirshberg, Khandani, and Vissing-Jorgensen propose a real-time index of investor mood derived from the text content of the headline articles from The New York Times. Moreover, depending on how predictions from machine learning analyses are disseminated, the activity could be labeled front running, or when self-performed, insider trading. However, other papers using social media data realize that it offers real-time tracking of social phenomena, such as consumer-level product opinions or corruption. Indeed, it seems likely that social scientists will increasingly use these data for more than testing existing theories of human behavior.

We can categorize the particular fields where computational social science methods are utilized in a few different ways. First, some subfields are focused on understanding and exploiting the proliferation of new sources of observational data that is becoming available. In many cases, this work is naturally labeled as social data science. Second, while the majority of computational social science emphasizes derived data, a number of researchers focus on learning directly from secondary or empirical data to infer social processes based on behavior, preferences, and interactions. Finally, in many cases we also see work going into the development and synthetic modeling of representative populations that maintain privacy and confidentiality. It is worth noting that while there are obvious benefits to large-scale data analysis made possible by new and diverse data streams, we acknowledge that much of social science is inherently small data and focuses on a limited number of well-studied populations and behaviors.

4. Challenges and Ethical Considerations in Computational Social Science

The first group of challenges relates to the quality of the conclusions that can be drawn from analyzing these large-scale internet data. Given the breadth of understanding one might pursue about human social interactions through studying internet behavior, how can we validly distinguish between the very large number of possible statistical generalizations that we might construct? In traditional social scientific research, researchers have tried to create meaningful generalizations via either random selection of subjects to be studied or the random assignment of participants to conditions. Previous work in Computational Social Science has also made use of internet data collected via pre-planned research studies to answer various questions about human behavior online. However, with fewer constraints on the experimentally controlled context of internet data, observational data by nature presents challenges for establishing the conditions necessary to infer causation. Acknowledging the point that sample size in a big data context may not always guarantee representativeness or randomness, could implementing the same safeguards on big data to increase their utility for making conclusions similar to that in smaller scale internet research be useful for the growing field of new opportunities for strengthening the potential for understanding human social interactions using computational methods?

As with any scientific advancement, the exponential growth in the “big data” available to researchers for studying human behavior via the internet and related devices provides both opportunities and challenges. Here we will focus on two main types of challenges that arise in conducting computational social science research: challenges related to the validity of the derived conclusions and the ethical considerations involved in performing this kind of research.

5. Future Directions and Opportunities for Advancement

Areas of focus include the fundamental science focused on developing significant substance and depth for accurate theory on social systems, changes, and individuals; addressing the very high-quality science requirements of policy evaluation research, especially in areas requiring iterative estimation, such as climate and international games; development of social science bridge projects and consulting services that focus on addressing scientific challenges and opportunities that have a strong social component across sciences like biological engineering and neurobiology.

We anticipate the three original contributions of our paper in the discussion, so while our proposed research program is ambitious, it is not totally unexpected. Our expectation is that over the next five to ten years, the maturation of the tools, data, and methods of Computational Social Science (CSS) will enable us to combine the theoretical traditions and empirical practices of the various established social sciences with the development of new rich areas of CSS in ways that will bring significant new opportunities and hold significant new promises about addressing fundamental social behaviors at a depth and richness never before achieved.