Research by Zachary Steinert-Threlkeld, assistant professor of public policy, measuring the political opinions of Russian-speakers in Ukraine at the time of Moscow’s 2014 incursion into Crimea has been published online by the journal Post Soviet Affairs. The study, co-authored by Steinert-Threlkeld and Jesse Driscoll of UC San Diego, uses a vast collection of social media data to demonstrate that many self-identified Russians living in Ukraine would not have favored a continued campaign to expand Russia’s borders. “Our supposition is that if Russian strategists were considering expansion beyond Crimea, they would have been able to use social media information to assess, with a great deal of precision and in real time, the reception that they would likely receive,” the authors wrote. While there is no evidence that Russian leaders took advantage of this type of analysis, the authors conclude that tapping into social media traffic could provide a useful source of intelligence for those planning military campaigns. “Social media data are straightforward to analyze systematically and can be collected at a relatively low cost,” wrote the authors, whose team, including research assistants in Kiev, used a data set of 6.8 million tweets to gauge social attitudes shared by Russian-speakers. “The prevalence of overtly political behaviors on social media provides important clues about the political dispositions within communities,” they said.
Gary Segura, dean of the UCLA Luskin School of Public Affairs and an expert in polling and public opinion, was quoted in a Pacific Standard article dissecting President Trump’s announcement to cancel foreign aid to El Salvador, Honduras and Guatemala. Trump has made multiple threats in the past to cut off the three Central American countries due to his dissatisfaction with their respective governments’ failures to stop people from leaving. After his recent announcement that funds would be withheld from the three nations, experts objected, explaining that the funds help combat crime and violence, ultimately serving U.S. interests. Segura maintained that ulterior motives were behind the policy decision, which would fuel the asylum crisis. He tweeted, “Pay attention folks. This is an INTENTIONAL act to drive MORE asylum seekers to the U.S. border to help [Trump] maintain his crisis. It’s ugly, devastating in impact, and bad policy.”
The tweets of Donald Trump are not known for factual accuracy, and Jorja Leap of UCLA Luskin Social Welfare told PolitiFact that his recent claims about ICE “liberating” towns from MS-13 and other gangs are an “outrageous” example of his tendency to exaggerate. “This is hyperbolic and misleading language,” said Leap, who is also director of the Health and Social Justice Partnership at UCLA Luskin. “Liberation is usually the terminology of military forces — as in, the Allies liberated France from the Nazis.”
By Stan Paul
Zachary Steinert-Threlkeld has long been fascinated by crowd dynamics, especially among those drawn to mass demonstrations. As a Ph.D. candidate in political science, Steinert-Threlkeld knew that social media generated at protests were a rich source of data — but he could find few tools to help him analyze it.
Now, in a world awash with popular uprisings and social movements — from Tahrir Square in 2011 to the Women’s March following the 2017 presidential inauguration — the assistant professor of public policy at the UCLA Luskin School of Public Affairs has used data generated by millions of posts on Twitter to learn more about crowd behavior and mass motivation.
Steinert-Threlkeld created a guide for acquiring and working with data sets culled from Twitter, which has more than 320 million global accounts generating more than half a billion messages every day.
His efforts culminated this year with the publication of “Twitter as Data,” the first guide in Cambridge University Press’ new Elements series on Quantitative and Computational Methods for Social Science. The series provides short introductions and hands-on tutorials to new and innovative research methodologies that may not yet appear in textbooks.
“When I was learning this as a graduate student, there was a lot of piecing together this information,” said Steinert-Threlkeld, who said he relied on sources such as Twitter documentation and online Q&A forums such as Stack Overflow. “I was able to do it, but it would have been a lot nicer if I had a textbook to show me the lay of the land.”
Steinert-Threlkeld, whose work combines his interest in computational social science and social networks with his research on protest and subnational conflict, said the book includes an interactive online version that allows users to click on links to download information and even sample data.
“It is differently comprehensive than a book,” Steinert-Threlkeld said. He described it as a “more interactive book experience — the first in social science that does this.”
In the book, Steinert-Threlkeld writes: “The increasing prevalence of digital communications technology — the internet and mobile phones — provides the possibility of analyzing human behavior at a level of detail previously unimaginable.” He compares this to the development of the microscope, which “facilitated the development of the germ theory of disease.”
He adds: “These tools are no more difficult to learn and use than other qualitative and quantitative methods, but they are not commonly taught to social scientists.”
To remedy this, Steinert-Threlkeld provides a systematic introduction to data sources and tools needed to benefit from them.
For example, people always want to know who’s protesting and how that influences others who might protest, Steinert-Threlkeld said. Most information has been restricted to surveys, which have limitations. “And so the researcher either gets lucky and happens to have scheduled a survey that occurs during a protest, but usually it’s after the fact.”
That is what’s exciting about using big data to study crowd behavior. “It’s like people always answering surveys,” he said. “Basically, every second you’re giving me survey data. Now we can tell in real time who’s protesting.”
One application of Twitter data is estimating crowd size, Steinert-Threlkeld said. In the past, he has had to rely on reports from organizers, police and the media to gauge the size of protests. “But I’m collecting tweets with GPS coordinates so I can say, ‘Oh, there are these many tweets or these many users from L.A. at this time or Pershing Square at this time, and explain whether that’s a reliable estimate or not of actual protesters.”
Twitter information can also be used to create data based on images shared from protests, Steinert-Threlkeld said. “The work I did before was all text based: What are people saying? Who’s saying it? When are they saying it? That sort of thing. But people share a lot of images online. They share more than they did three or four years ago. It’s really where the space is moving.”
Steinert-Threlkeld said that getting data into a form that a researcher can use requires a different skill set than designing and administering a survey. “But it’s still in some ways survey-like at the end of the day,” he said.
And “it’s fun,” he said. “Now we can tell in real time who’s protesting. We don’t know where the person lives, or their income, or their name. It’s still anonymous. We don’t know if the person who shares the image was there so we’re not incriminating anyone, but we can get a lot of information about protesters that we couldn’t before.”
In the final section of his guide, Steinert-Threlkeld writes: “These data are not a ‘revolution.’ Instead, they represent the next stage in the constant increase in data available to researchers. To stay at the forefront of data analysis, one needs to know some programming in order to interface with websites and data services, download data automatically, algorithmically clean and analyze data, and present these data in low-dimension environments. The skills are modern; the change is eternal.”
By Stan Paul
If your 2016 Thanksgiving dinner was shorter than usual, the turkey on your dining table may not have been to blame.
Who you had dinner with and their political affiliations following last year’s divisive election may have shortened the holiday get-together by about 25 minutes — or up to an hour depending on how many campaign/political messages saturated your market area. It’s all in the data.
“It’s not that conservatives and [liberals] don’t like eating Thanksgiving dinner with each other; they don’t like eating Thanksgiving dinner together after an incredibly polarizing period,” said Keith Chen, associate professor of economics at the UCLA Anderson School of Management. Chen was among a group of scholars and data researchers who presented recent findings on Aug. 25, 2017, at a daylong conference about computational social science and digital technology hosted by the UCLA Luskin Center for Innovation.
Information gleaned from social media and from cellphone tracking data can reveal and confirm political polarization and other topics, such as poverty or protest, said researchers who gathered at the “The Future of Humans as Sensors” conference held at the UCLA Luskin School of Public Affairs.
The event brought together social scientists and data researchers to look for “ways to either extend what we can do with existing data sets or explore new sources of ‘big data,’” said Zachary Steinert-Threlkeld, assistant professor of public policy at UCLA Luskin and the leader of the program.
Steinert-Threlkeld presented his latest research, which was motivated by the Women’s March in the United States, as an example of measuring protest with new data sources that include geo-located Twitter accounts. While conducting research, Steinert-Threlkeld has observed that working with social media data has actually become more difficult of late.
“While Facebook lets you use data from profiles that are public, most people have private profiles,” Steinert-Threlkeld said. Seeing private data requires researchers to work directly with Facebook, which has become more cautious in the wake of a controversial 2014 paper, thus impacting what scholars can publish. In addition, Instagram previously provided much more data, but since 2016 it has followed the Facebook model and that data has been severely restricted despite Instagram’s norm of having public profiles, he said.
“This workshop will discuss how ‘humans as sensors’ can continue to yield productive research agendas,” Steinert-Threlkeld told conference attendees.
Talking about new and innovative ways to do this, Michael Macy, a sociologist and director of the Social Dynamics Laboratory at Cornell, began his presentation by pointing out the innate difficulties of observing human behavior and social interaction, as well as both the potential and the limitations of social media data.
“There are privacy concerns; the interactions are fleeting. You have to be right there at the time when it happens.” He added, “They’re usually behind closed doors, and the number of interactions increases exponentially with the size of the population.”
But, Macy said, new technologies in various scientific fields have opened up research opportunities that were previously inaccessible.
“We can see things that we could never see before. In fact, not only can we see things, the web sees everything and it forgets nothing.”
He tempered the potential of digital data with the fact that for the past several decades the main instrument of social science observation has been the survey, which comes with its own limitations, including unreliability when people recount their own behavior or rely on memories of past events. But, he said, “In some ways I see these social media data as being really nicely complementary with the survey. They have offsetting strengths and weaknesses.”
Macy provided examples of ways that tracking of political polarization can be done, not by looking at extreme positions on a single issue but by inferring positions on one issue by knowing the position that individuals hold on another. This can range from their choices of books on politics and science to their preferences for cars, fast food and music.
“The method seems to recover something real about political alignments … political alignment can be inferred from those purchases, and then we can look to see what else they’re purchasing,” Macy said.
“What I think we’re really looking at is not the era of explanation, at least for now … it’s the era of measurement, and what we are now able to do is to test theories that we could not test before because we can see things that we could not see before.”
The day’s presentations also included the ways in which data can be used to provide rapid policy evaluation with targeted crowds and how demographic sampling weights from Twitter data could be used to improve public opinion estimates. Data could also help fight poverty worldwide.
The world seems awash in information and data, but “most of world doesn’t live in a data-rich environment,” said presenter Joshua Blumemstock, an assistant professor at U.C. Berkeley’s School of Information and director of the school’s Data-Intensive Development Lab.
“You can use Twitter data to measure unemployment in Spain. The problem is that these methods don’t port very well in developing countries,” Blumenstock said. “There’s these big black holes in Africa for Twitter.”
Blumenstock discussed how data from billions of mobile phone calls in countries such as Rwanda could be used in conjunction with survey data to create a composite of where individuals fall on the socioeconomic spectrum. In turn, the information collected could be “aggregated up” to a much larger regional or national level.
“And when you aggregate up, you start to get things that might be conceivably useful to someone doing research or some policymaker,” such as being able to respond instantaneously to economic shocks, Blumenstock said. In addition, instead of costing millions of dollars and taking years, he said this methodology could potentially cost thousands of dollars and be conducted in weeks or months.
“For researchers like me who are interested in understanding the causes and consequences of poverty … just measuring the poverty is the first step. For people designing policy for these countries, their hands are tied if they don’t even know where poverty is,” Blumenstock explained. “It’s hard to think about how to fix it.”
By Stan Paul
The massive Arab Spring protests that began in late December 2010 and spread from North Africa to the Middle East generated huge crowds and had quick and profound effects — including the overthrow of Egyptian President Hosni Mubarak, who had held a firm grip on the country for decades.
Was this the work of people at the core of networks trying for years to create such a movement? Not according to research by Zachary Steinert-Threlkeld, assistant professor of public policy at the UCLA Luskin School of Public Affairs.
In a paper published in “American Political Science Review,” Steinert-Threlkeld argues that individuals not central to a social network may be more responsible for generating collective action and driving protest than those at the center. Steinert-Threlkeld calls his theory of the periphery’s ability to mobilize “spontaneous collective action.”
“Protests occur as a result of decentralized coordination of individuals, and this coordination helps explain fluctuating levels of protest,” Steinert-Threlkeld wrote in the study, “Spontaneous Collective Action: Peripheral Mobilization During the Arab Spring.” Although not intended to explain the Arab Spring, Steinert-Threlkeld’s paper presents the first large-scale, systematic evidence of how individuals behaved in each country.
Unlike simple models such as disease contagion or transmission of news, which need only one exposure to infect or inform the next individual, the spreading of protests is more complex, making it less certain, said Steinert-Threlkeld, who teaches courses on these topics at the undergraduate and graduate levels at UCLA Luskin.
“Having someone tell you, ‘Hey, I’m going to protest tomorrow’ is much less impactful than having multiple people tell you they are protesting tomorrow,” he said. Large groups of people, as opposed to a few central individuals, are able to discuss “where to go, how to get there, when to go,” as well as what is going on once there, Steinert-Threlkeld added. In addition, individuals debating whether or not to protest must receive a credible signal that large numbers of people are protesting.
“Protests are a complex contagion phenomenon because increasing participation makes others more likely to join,” Steinert-Threlkeld said, pointing out that because more individuals are on the periphery than in the core of a network, protest is more likely to occur where the larger group is located. “The decision to participate in a protest appears to be driven by normal people taking cues from each other, not from elites.”
In testing his theory, Steinert-Threlkeld took advantage of the Integrated Conflict Early Warning System, a dataset of daily protests across 16 countries in the Middle East and North Africa during 14 months in 2010 and 2011. He then combined the dataset with geocoded, individual-level communications (nearly 14 million tweets from the same period), to measure the numbers of connections of each person.
“These 13,754,988 tweets show what was being said, when it was being said, where and how many connections each tweet author had,” he reported in the study. “Combining these datasets and using a wide range of models and operationalizations, mass mobilization is shown to occur through peripheral individuals.”
Said Steinert-Threlkeld: “Egypt’s January 25th protests surprised everyone — activists, bystanders and state authorities — with its large mobilization and brief occupation of Tahrir Square.”
He added that many Muslim Brotherhood leaders were summarily jailed — even though they did not sanction protests — because the Mubarak regime assumed only they could have mobilized such a crowd.
“This paper demonstrates the contributions big data can make to understanding processes of social influence in social networks,” he said.