Alternative Social Science

By Nick Adams, Ph.D.

Now is the time for social scientists to take responsibility for guiding societal improvement.

Twenty-first-century societies are rapidly changing. We’re witnessing historic levels of partisan discord and institutional breakdown, and multiple simultaneous sea changes in norms around gender and ethnic identity, sexual expression, and the definitions of criminality. These political and cultural shifts, often amplified and accelerated through Internet platforms, are occurring alongside major economic upheavals, including the deaths and births of entire industries, renewed international trade wars, and inequality levels rivaling those of feudal times. Worse, there is no end in sight for these tumultuous trends. What are people to do? How are we to make sense of all this turmoil and find some working consensus about social reality (if not a social contract) allowing more of us to find a stable and comfortable way in the world?

A naïve observer might suspect that social scientists have a role to play in all this. After all, we can now access immense quantities of social data––digitized historical records and newly created user behavior data generated each day––that could be used to understand and improve the human condition. Shouldn’t, social scientists be more valuable than ever? 

The technical folks see the appeal. Facebook, Google, and Twitter are generating more data about social behavior each month than social scientists could access in any previous century, and they’re hiring nearly as many social scientists each year as are all the leading research universities (R1s) in the United States. Physicists and applied mathematicians are hot to the trend, too. With massive piles of social data going unanalyzed by most traditional social scientists, they’re swelling the ranks of a new discipline called Computational Social Science, and funders are coming out of the woodwork to support their efforts applying models of interaction designed for particles and molecules to various datasets describing human interactions.

More and more, foundations and donors focused on solving social problems are realizing that the challenges of our time will require at least some sociologists, economists, anthropologists, psychologists, and political scientists to engage in a new style of knowledge production free from the constraints of academic career building. This alternative social science will be organized around independent labs that collaborate with universities, involve the public in research, intricately analyze large quantities of social data, publish openly and rapidly, and lead the design of pro-social tools people can use to improve their relationships, organizations, communities, and civil societies.

“Not My Job”

Still, academic social scientists have not seen it as their job to intricately analyze immense quantities of data, nor to directly solve social problems. They might make recommendations to policymakers about the latter. And to be sure, thousands of qualitative researchers have carefully analyzed small datasets, while thousands of quantitative researchers have scoured larger datasets, revealing all the patterns just below the surface. Doing more, however, has been impossible or disqualifying, either because adequate tools did not exist, or because existing approaches required teamwork or multi-year projects.

Ironically considering their focus on social subject matter, team projects are detrimental to most social scientists’ careers unless they’ve already achieved tenure. Pre-tenure academics are often instructed to focus instead on creating single-author publications proving their skills across the entire knowledge-production pipeline –– from research study design and execution, to data analysis and publication. With only a couple years to establish a publication record that will earn them a job, and a few more years until tenure decisions are made, most social scientists can’t take on the high-scale, long-duration research projects that would really tease out intricate and nuanced social behavior, particularly if they also require the development of new tools and methods.

Healthy Change

Shifting the incentives of social scientific production could be thrilling or terrifying depending on one’s perspective. But change doesn’t have to come as some conquering party enjoying spoils at the expense of some deposed old regime. With so much social data, so many new methods, and increasing demand for social science outputs, no one needs to lose for social science to engage in additional research approaches solving more problems and positively impacting people’s lives. In fact, it may be time the social sciences begin to take the same steps other scientific disciplines took as they matured and aroused increasing interest from a solutions hungry public. Doing so will require some conscious effort, particularly from funding organizations and a small minority of social scientists whose motivations extend beyond the tenure track. But with so much new technology, data, and demand for social solutions, the time to act is now.

New Capacities Bridging Old Divides: Data, Tools, and Citizen Scientists

We’re all aware of the social data deluge. But fewer social scientists know that they can now intricately analyze the vast quantities of social data coming online everyday. Even if we have to wait a few more years for the Social Science Research Council to broker data sharing agreements with Facebook, Google, and Twitter allowing us to understand their users’ behavior, we can already analyze the whole of the Congressional Record, Supreme Court Transcripts, and Federal Reserve reports, along with numerous archives digitized by the Library of Congress. Why haven’t we? Why don’t we already have highly accurate predictive and explanatory models of politicians’ machinations, judges’ decision-making, and more? It’s because amassing and analyzing all that observational data in detail is well beyond the scope of solo-author academic social scientists, and we’ve only recently created the tools to make such analyses feasible.

Social interactions—even when relatively structured by ‘Roberts Rules of Order’ and overt political alignments, as in the case of Congressional floor speeches and hearing transcripts—are influenced by numerous variables. Closely analyzing a five-minute conversation can require thirty minutes of work, especially for the expert researcher who can see below the surface of the interaction. So, it has been hopelessly infeasible for social scientists to apply complex and nuanced theories of (e.g. symbolic) interaction to lengthy transcripts of legislative activity, whether they capture the work of Congress or the local school board. It’s simply too much effort. Instead quite often, quantitative researchers attempt to find patterns in politicians’ behaviors by indexing their ideological positions and modeling that alongside their voting behavior. Or, qualitative researchers perform in-depth analyses of one or a few episodes, teasing out some nuanced mechanism of political expression. The latter studies provide social science with additional hypotheses and explanatory fodder, while the former help measure the propensity of the most common behaviors. But neither capture enough rich, comparable data to support the testing and refinement of complex, dynamic theories of political interaction.

For centuries, these very human limitations on data processing have divided social scientists into the complementary (and sometimes rival) camps of ‘quant’ and ‘qual.’ They have also stymied the development, application, and refinement of some of social science’s most promising meso-level theories explaining the group behaviors that emerge somewhere between macro-level economic and historical forces, and the micro-level (social) psychological phenomena we can demonstrate in a lab setting or identify in narrowly-scoped field studies. Without numerous, richly curated examples of meso-level phenomena––the kind we could identify in the masses of social data now available––we have been unable to reliably adjudicate, for instance, whether a particular type of social situation is best described by Goffman’s dramaturgical theory of stigma, Becker’s labeling theory, Ridgeway’s model of implicit privilege, or DuBois’s theory of double consciousness. But all that is changing.

In the last year, my company and I have released a new tool—TagWorks—making it possible for researchers to intricately apply their hard-earned expertise at the scale of big data. It’s not some magic algorithm for parsing social data. Those don’t exist. Even the very best natural language processing tools fail to understand the complex human expressions that emerge from our ambiguous and redundant human languages. They certainly cannot parse a Supreme Court transcript by the conceptual schemata of a subfield like ‘conversation analysis.’ But TagWorks can. It can also parse news reports about protest events by the schemata of the ‘contentious politics’ subfield. And it can parse historical magazine articles according to the schemata of ‘intersectional feminism.’ In fact, TagWorks can apply any conceptual schema, no matter how intricate, to any set of documents, no matter how numerous, allowing researchers to finally perform the high-scale, high-fidelity analyses of observational data they previously only dreamt about.

TagWorks can do all this work (often called content analysis, data labeling, or annotation) because it converts researchers’ intricate and holistic expertise into an assembly line of small evaluation and annotation tasks that lay people can ably perform on their personal computers. TagWorks’ open algorithms find a statistical consensus among independent workers’ annotations of identical documents, then report tried-and-true reliability statistics like Krippendorf’s alpha for each of the labels. The resulting annotations are comparable in quality (and often slightly better) than experts’ and are produced much faster at much higher scales. Even unimaginably large projects analyzing millions of documents can be completed using TagWorks since its output labels can be fed into supervised machine learning models that teach AI to mimic the annotation efforts of humans.

Right now, TagWorks processes textual (natural language) data for top universities, libraries, government agencies, and industry clients. And soon enough, we will be working on video, audio, and image data, too. Some of our clients recruit volunteer citizen scientists to perform their annotation work. Others let us manage paid Internet-based workers for them. (We pay minimum wage or better). By utilizing these vast volunteer/labor pools, TagWorks is able to complete in mere months the sorts of massive, intricate projects that would require years of expert effort using tools like AtlasTI, NVivo, DeDoose, Prodigy, and others. Moreover, our clients don’t have to spend their days recruiting, training, managing, and supervising student research assistants. TagWorks’ annotation assembly lines are designed to provide all annotation instructions, training, and batch management features within its interfaces. So, most of our clients experience a 50% (or more) reduction in their management effort while experiencing a 10x (or more) improvement in document throughput.

What does this new capacity mean for social science? Very little, if researchers are unable to take advantage of the opportunity. We can only lead the herd to water. But if many choose to drink, we may just see a golden age of social science, where complex phenomena are not only explained but predicted with such high reliability that we can begin designing the sorts of social interactions, situations, and institutions that promote human happiness and flourishing.

­­Newly Available & Under-appreciated Audiences

If social science can process massive troves of observational data to develop greater clarity about how humans act and interact, there is little doubt that the public will take interest. Popular psychology is so popular because every person, whether they strive to be or not, is at least an amateur psychologist. They’re also amateur sociologists, economists, anthropologists, and political scientists. These disciplines are only less popular because they have produced fewer findings so ubiquitously applicable as pathological diagnoses and personality trait classifications. As social science gets its arms around larger datasets describing more social phenomena, it can also embrace broader audiences, hungry to understand how we collectively construct our realities, and how we might edit, or make ourselves more comfortable within, them. (If we follow Giddens thesis in Modernity and Self Identity, this pre-occupation with the social will only feature more prominently in the 21st century.)

Engaging people as volunteers on massive data curation projects like those enabled by TagWorks is a great place to start. Such projects do more than educate volunteers about their particular domain, infusing social scientific knowledge into the public consciousness. They also offer insight into the process of science, correcting the misconception of science as a body of knowledge when it is better understood as a set of approaches and techniques for systematic inquiry.

Fortuitously again, current communications technology makes it easier than ever for social scientists to reach public audiences. While publishing in a peer-reviewed journal can take a year or more, research findings can be published to publicly accessible research archives immediately. Blogposts or white paper reports written to engage lay audiences can be circulated through global social media platforms like Twitter and Facebook with only minutes of effort. Recording and distributing brief explainer videos or audio podcasts is simpler than ever, too. And there are even platforms (like Medium and Patreon) designed to collect subscription fees from the audiences who consume such media.

Publishing for the public probably earns a researcher little regard from hiring or tenure committees. But such approaches to publishing ought to be rather valuable to foundation program managers who measure their impact in terms of lives improved. Especially as meso-level research delivers findings helping the public coordinate policy action, accommodate and heal cultural differences, and find common ground across our economic positions, social scientists should learn to step confidently into public fora and speak effectively to diverse audiences through diverse media. This may even be an area where existing social science publishers can lead us into the future. 

New Organizations

For academic social scientists—unless they are already tenured—it will remain rather risky to adopt the goals and values of alternative social science. Taking on complex, big data projects may mean collaborating with other scholars of equal rank, which will dilute the value of resulting publications so necessary for hiring and promotion. Any work deploying new methods can be summarily rejected as ‘insufficiently proven.’ Engaging the public may be openly dismissed as a ‘distraction,’ if not as inappropriate ‘activism.’ And prioritizing the solving of contemporary problems over engagement in decades-long debates can easily be derided as ‘headline chasing.’ Work actually implementing solutions could well be classified as ‘not even social science.’ Given the abbreviated way and narrow path that leads to tenure, the safest approach for academic social scientists probably involves accepting the counsel of senior scholars: to publish two or three tightly scoped, high-impact journal articles that use trusted methods to improve existing theory in a subfield led by well-regarded senior scholars.

Given these risks, alternative social science––as it takes flight––will most likely launch from alternative research organizations that, while collaborative with academic researchers, are not geared to produce them according to the traditional mold. Such organizations will reward staff who work effectively in a larger team, regardless of their ability to demonstrate their independent excellence across every aspect of the research lifecycle. By focusing on collective outputs, these alternative labs will be able to take on much larger projects, intricately curating vast datasets to prepare them for complex analyses that tease out and quantify what previously could only be qualitatively described. Each of their projects will produce not just one or two papers, but dozens, mostly co-authored. These labs will not exclusively publish in academic journals, either. They’ll continually educate their giant pools of volunteers, and publish their findings through video and audio content digestible by the masses, and shareable via social media. And when they are confident in their understanding of some facet of human behavior, alternative social science labs will not shy away from creating interventions––often deployed through Internet software––that will help members of the public voluntarily solve their social problems.

Alternative social science labs are likely to crop up near universities, so they can mentor, and benefit from the voluntary labor of, data science and social science students eager to make a positive social impact. They will Iikely collaborate with traditional academics, too, drawing on their expertise and labor to ensure their projects integrate with the best of basic science even as they engage and educate the public.

Perhaps the best proof of concept for an emerging alternative social science lab already exists in the form of the Goodly Labs, which I lead. Independently founded as a non-profit in 2014 and incubated out of the Berkeley Institute for Data Science, Goodly Labs is currently undertaking six high-impact social science projects and launching three to the public in 2020. Each of these projects have developed software allowing the public to deeply engage in social science research that will positively impact important institutions including the media, police, Congress, the Supreme Court, and the Federal Reserve. Goodly Labs is highly collaborative with researchers, too. It has mentored nearly two hundred UC Berkeley students and post-docs, and its project teams include faculty at New York University (NYU), Simon Fraser University (SFU), Drexel University, and a Nobel Laureate at UC Berkeley, in addition to senior software engineers from government and industry. Goodly has enjoyed funding support from the National Science Foundation, the Alfred P. Sloan Foundation, UC Berkeley, NYU, SFU, the Open Annotation Fund, the Social Science Research Council, the Pritzker Foundation, The Miller-McCune Foundation, and Schmidt Futures.

Though Goodly Labs has mostly flown under the radar as it has developed its projects, 2020 will be a banner year for the organization. In April, it will be inviting Internet-users around the world to join its Public Editor project––a people-powered misinformation detector deploying TagWorks annotation technology to find over 40 types of misleading content in news articles. Public Editor labels argumentative fallacies, mistaken inferences, psychological biases, and other misinforming content in news articles within 30 minutes of its publication to slow the spread and dampen the effect of mis-/dis-information. Later in 2020, Goodly will launch its Demo Watch project, which enlists volunteers to label all the features of police, protester, and city government interactions described in news reports about the Occupy movement. The largest and most intricate project of its kind ever attempted, Demo Watch will be able to tease out complex, multi-level interactive dynamics and identify the decision points that lead to violent outcomes between police and protesters, so they can be prevented. Finally, Goodly will launch its Liberating Archives project in the latter half of 2020. This project has significantly upgraded government transparency to ensure that nearly anyone with an Internet connection, including journalists, social scientists, and citizens can easily evaluate the actions of elected officials to better hold them accountable.

By setting aside the demands and constraints of the academic career path (which too often discourage team projects and methodological innovation), alternative social science organizations are able to take on projects that are unusually large in scale, high in impact, and engaging to a public that is eager to learn about the social world and tackle its major challenges.

Increased Relevance, Power, & Responsibility

Change is often disconcerting, if not downright scary. But alternative social science can succeed and even feed academic social science without imposing change on anyone. The analogy between the physical sciences and their engineering cousins may be instructive, here. Notably, while chemical engineers apply knowledge developed by chemists to solve problems (often for some civil society, government, or industry customer), they have not replaced chemists, nor diverted funds from chemistry departments. If anything, the engineering cousins of the physical sciences have served as a bridge between those more established, basic science disciplines and public and industry actors willing to fund both immediate solutions and basic research. As a result, the boundaries between engineering and basic research are rather healthy. The latter researchers are usually more methodologically conservative since they are dedicated to the scientific norm of publishing only high-certainty findings arrived at by high-certainty methods. But both camps are comfortable where they live. And, if an engineer discovers something relevant for basic research or vice versa, the boundaries are porous enough that either can easily publish wherever their work finds the best audience.

With similarly healthy boundaries between traditional and alternative social science, the latter will be able to pursue and demonstrate newer approaches on a time scale appropriate for its larger projects. And when those approaches work, generating masses of carefully curated data, alternative labs eager to impact today’s social world will also be eager to share their data with any social scientist willing to help analyze it and publish useful findings.

In other ways, too, an alternative social science would likely benefit traditional academic social science. By engaging more closely with journalists, alternative labs could help establish more pathways between social scientists and news media, popularizing not just their own findings, but the scholars who have inspired their work. Alternative labs could even catalyze increasing demand and appreciation for social science, paving the way for those academic social scientists who are interested in reaching broader audiences.

If there is any danger posed by alternative social science it may be this: closer engagement with the public may cause citizens to expect more engagement from all social scientists. To those researchers who entered the field to improve society, such a turn of events may be welcome. But others cultivate a careful detachment from their objects of study. They may not want anybody casting them as a savior. While this position is certainly defensible from a methodological and professional standpoint (and probably healthiest for some of our colleagues of a more introverted disposition) it may not be fair to the public for social science to adopt it wholesale. After all, the public funds most social science research. (Even when it is funded by foundations, those foundations’ largesse depends on publicly mandated tax exemptions.) It is reasonable, too, for a citizenry beset by social tumult to look to relevant experts for guidance and support. If we have the power to engage the public in the improvement of social science and society, perhaps we also have the responsibility to do so. Certainly, we can at least allow some of our colleagues try.

About

Nick Adams is an expert in social science methods and natural language processing, and the CEO of Thusly Inc., which provides TagWorks as a service. He holds a doctorate in sociology from the University of California, Berkeley and is the founder and Chief Scientist of the Goodly Labs, a tech for social good non-profit based in Oakland, CA.