by HE Brady · 2018 · Cited by 38 — Tukey’s impact on statistics has been immense (Statistical Science 2017), and his concept of data analysis covers much of the same ground as
22 KB – 31 Pages
PAGE – 2 ============
2 The Challenge of Big Data and Data Science Big Data and Data Science fiBig data and data science are being used as buzzwords and are composites of many conceptsfl says the US National Institute of Standards and Technology i n a 2015 fiframeworkfl report on fiBig Datafl (NIST, 2015, 2). The phrase fibig datafl a ppear s frequently in the press and in academic journals, and fidata sciencefl programs have sprouted in academia over the last five years. On March 29, 2012, the White House Office of Science and Tech nology Policy announced the fiBig Data Research and Development Initiativefl (OSTP 2012) that would build upon federal in itiatives firanging from computer architecture and networking technologies to algorithms, data management, artificial intelligence, machin e learning, and development and deployment of advanced cyberinfrastructure (NITRD 2016, 6).fl In just the first six months of 2018, the New York Times published ar ticles telling us fiAI and Big Data Could Power a N ew War on Povertyfl (January 2, 2 018), fiBig Data Comes to Dieting,fl (January 25, 2018), fiHow Democracy can Survive Big Datafl (March 22, 2018), and fiHow Big Data is ‚Automating ™ Inequalityfl (May 4, 2018) . fiBig datafl appeared about 560 times per year in JSTOR from 2014 -2017 even tho ugh it was mentioned less than once a year in the century before 2000 and only an average of about eight times a year between 2001 and 2010. In the last five years, at least seventeen Data Science programs have started at major American research universities ( http://msdse.org/environments/ ), and the Internet is replete with advertisements for data science books and courses, often with the come -on of fiBecome a Data Scientist.fl The phrases have certainly caught on, but th ey mean d ifferent things to different people, and some even d oubt their utility (e.g., boyd & Crawford 201 2; Donoho 2017 ). Despite the imperfection of the se terms and the hyperbole that often surround s them , they point to real changes that are important for political science. fiBig Data ,fl fiData Sciencefl and the related idea s of fiArtificial Intelligence ,fl fiCyberinfrastructure,fl and fiMachine Learningfl have implications for the following developments and trends discussed in this article : Societal and Political Change from Big Data and Data Science ŠThe volume, velocity, variety, and veracity of data being generated by and available to governments, armies, businesses, non -profits, and people ha ve combined with the enormous increases in compu ting power and improvements in data science methods to change society in fundamental ways, creating new phenomena, and raising basic questions about the control and manipulation of people and populations, the future of privacy, the veracity of information, the future of work, and many other topics that matter for political scientists. Increasing Amounts of Data Available to All Scientists, including Political Scientists Œ All the sciences are being affected by these changes. The Thirty Meter Telescope co ming on line in 2022 will gene rate 90 terabtyes every night; genomic data is doubling every 9 months, and it is currently being produced at approximately 10 ter abytes per day; the large Hadron collider at CERN generates 140 terabytes per day . The web prod uces about 1,500,000 terabytes every da y and this flow of data offers social scientists a chance to study the fisinews of societyfl (Weil 2012) and the finerves of governmentfl (Deutsch 1963) in a way that could not be done in the past. Now political scientis ts can observe and analyze (sometimes in real -time) the information that people choose to consume, the information produced by political actors, the environment in which they live, and many other aspects of people™s lives.
PAGE – 3 ============
3 New Ways P olitical Scientists Organize their Work — With this onslaught of data political scientists can rethink how they do political science by becoming conversant with new technologies that facilitate accessing, managing, cleaning, analyzing, and archiving data . New Kinds of Questions Asked by Political S cientists ŒPolitical scientists must ask what they are trying to accomplish with concept formation, description, causal inference, and prediction into the future. In the process, new methods and insights will be devel oped about political behavior , and new designs will be put forth for political institutions . Dealing with Ethical Issues Regarding Political Science Research Œ Finally, political scientists must think about complicated ethical issues regarding access, use, and broadcasting of information, and the possible misuse of their models and results. Before considering these five changes and their implications for political science, I describe the exponential growth in data and computing power that has led to the prominence of fibig datafl and fidata sciencefl followed by definitions of these untidy phrases. Increasing Volume, Velocity, and Variety of Big Data Social scientists must come to grips with the current dramatic tr ansformation s in the communication of information that parallel the striking changes in transportation in the 19 th century. Historians report that with the invention of the steam engine, the time for and cost of travel dropped dramatically in the 1800™s creating new trad ing networks, new opportunities for migration, new kinds of cities with commuter suburbs, and new understandings of the world. In 181 6, using horse -driven stagecoach es, mule -drive n canal -boats , or sailing packets that averaged two to eight miles per hour , a trip between Philadelphia and Quebec (560 miles) took 103 hours Œ over four days at five to six miles per hour . By 1860 with the advent of railroads and steam -boats that went fifteen to thirty miles per hour, the time and cost for travel dropped by over two -thirds , and the same trip took just 31 hours Œ just over one day (estimated from Taylor 1951, Chapter VII, 141) . With the regular use of jet planes in the 1960s, people could fly from New York to London in less than seven hours compared to the two -weeks it took by ship in the mid -ninetee nth century Ša fifty -fold improvement over 100 years. These transportation changes Šalong with electrification at the end of the 19 th century Šrevolutionized society in ways that had enormous implications for politics , economics, and society. Changes every twenty years in information technologies punctuated the hi story of the late 19 th, 20th and early 21 st century : telephones (1870 -1890s ), phonographs ( 1870 -1890s ), cinema ( 1890 -1920s ), radio (19 00-1920s ), television ( 1940-1950s ), mainframe computers ( 1940 -1950s), personal computers ( 1970 -1980s) , Internet and World Wide Web (1980 -2000s ), cell phones ( 1980 -2000s ), and smart phones (2000s -Present) . The most fundamental innovation came with the move from analog devices to digital ones starting in the 1950s and proceeding dramatically in the 1990s and thereafter. These changes brought extensive digital datatification in which myriad events are now digitally recorded , widespread connectedness in which events and people are identified so that they can be linked up with one another, pervasive networking where people are embedded in a community of users who interact with one another and become nodes in larger network s, and ubiquitous computerized authoring where computers c reate new information that becom es part of the social system and its culture . Political scientists led the way in studying these changes. Harold Lasswell and Karl Deutsch were early students of communications and their impacts on societies . In 1983, MIT political scientis t Ithiel de sola Pool first looked at the production of fiwords fl in the American mass media (e.g., radio, television,
PAGE – 4 ============
4 records, movies, newspapers, books, etc.) and point -to-point media (telephone, first -class mail, telegrams, facsimile, and data communication) from 1960 to 1977. Pool found that words in these media doubled every eight years , growing at about nine per cent per year . He also found that that fiprint media are becoming increasingly expensive per word delivered while electronic media are becoming cheaper,fl so that figrowth in both mass and point -to-point media has been greatest in the electronic ones.fl Furthermore, fialthough the largest flow of words in modern society is through the mass media, the rate of growth is no w fastest in media that provide information to individuals, that is, point -to-point media.fl Finally, fithe words actually attended to from those media grew at just 2.9 percent per yearfl so that fieach item of information produced faces a more competitive mar ket and a smaller audience on average (Pool 1983, 609) .fl Pool predicted much of what we know about modern communications: they are growing fast, they are increasingly electronic and point -to-point, and people experience information overload and fragmented information flows. Perhaps most presciently, Pool also said that fiComputer networking is for the first time bringing the costs of a point -to-point medium, data communication, down to the range of costs characteri stics of mass media (611) .fl Subsequent studies by political scientists and others (Lyman & Varian 2003 ; Bohn & Short 2012 ; IJC, 2012 ) focused on the volume or stocks of information (e.g., the number of books in a bookstore ) as well as on the flows or velocity (the daily sales of books) and the variety of information (subject matters of books). They also measur ed inf ormation in digital bytes instead of words so that the measures reflect the proliferation of images such as video which communicate many more bytes per second than do words through text or speech (Bohn & Short 2012, 986) . Hilbert & Lopez (2011 , 63, Table 1 ) found that the world™s storage capacity in bytes per capita doubled every 40 months between 1986 and 2007 . The bulk of the world™s flow of communications was still in broadcas t communications which grew at the rate of 6% per year per capita but (point -to-point) telecommunications grew at the rate of 28% and could conceivably exceed broadcast communications within ten to fifteen years. Finally, they computed a new quantity Œ the growth in the world™s computational power in Millions of Instructions Per Second or MIPS , and they found that the world™s human ity guided general -purpose computation grew at an impressive compound annual growth rate of 58% per capita between 1986 and 200 7. Embedded applications -specific computation grew even faster, at 8 3%. This research identifies several notable trends that have produced the fiBig Datafl revolution. First, there is the tsunami of data about societal events, and digital communications are overtaking analog. This extensive digital datafication (Cukier & Mayer -Schoenberger 2013, 29) creates data in a format that can be readily stored and processed by computers . fi Recordingfl might be used instead of the ugly neo logism fidatification,fl but it seems too passive for processes that are trans mogrifying human interactions into data. Even though some of these data are relatively unstructured as text, audio , networks, or images , data scientists are figuring out ways to analyze them. Second, there is widespread conne ctedness because point -to-point telecommunications can be, in principle, more easily tracked than broadcasting. For example, w hereas broadcasters traditionally required elaborate survey operations (such as Nielson™s media -use diaries) to track their audience , Netflix has immediate data on the download of its movies. More generally, we can now record and connect data on individual postings, purchases, police encounters, and even perambulations. Dataficati on and connect edness mean that once ephemeral events c an now be identified and studied . Third, one feature of the changing information environment is especially important for social scientists. Whereas once communications were classified as either per son -to-person (e.g., conversation, letters, or telephone) or mass communications from one -source -to-many people (e.g., books, newspapers, cinema, radio, or television), modern communications involve mediated social networks Šnetworking Šthat
PAGE – 5 ============
5 combine features of both modes (Neumann 2016; Schroeder 2018): Twitter involves individual communications sent to many followers using hashtags that define self -mediated areas of concern . Facebook involves individuals with customized profiles who have networks of fifrie ndsfl and who have affiliations with common -interest user groups that share information . Google involves a query by an individual who is provided with a list of relevant websites . Amazon involves a search for a particular product that results in suggestio ns about other relevant products that can be bought online. In all these media, knowledge about people™s characteristics and their search behaviors is used to suggest and sometimes impose particular actions or relationships. The implications of these new modes of communication are not clear, but they probably operate differently in the three important spheres of politics, markets, and culture (Schroeder 2018) . They may also have important impacts such as increasing the chance for political polarization i n politics through the creation of networks that are closed with respect to dissenting opinions (Neumann 2016). Finally, whereas the communication of information traditionally involve d sending messages from one place to another in the most verisimilar f ashion possible even when the message was transformed along the way (e.g., from voice into electrical signals in a telephone), an increasing fraction of information is partly, if not entirely, computer authored by programs and algorithms that transform inputs into quite different outputs. Computers use program s to produce new product s that combine inputs in novel ways : A Google search takes a request and delivers plausible fianswersfl to that search; a computer game produces a fantasy virtual environment for entertainment; a Computer Automated Design program produces a design that meets certain specifications; and so forth. For the first time in history, aside from naturally produced information from the environment, there is non -human product ion of new information . Nature and humans no longer have a monopoly on authoring. W e now live in an era where computers can author, publish, and supply new forms of information. Another job of social science is to improve and understand these processes . Definition s of Big Data and Data Science The growth of data and the creation of large databases in business, government, daily life, and scientific research launched many efforts to understand and utilize data. fiData mining ,fl fiknowledge discoveryfl (Maimon & Roach 2005, 2010) and fibusiness intelligence and analyticsfl (Chen et al. 2012) became popular terms in business describing statistical and logical rule -based efforts to extract knowledge from large databases. Within engine ering, a seventy year tradition continues of building computers and robots with fi Artificial Intelligence (AI)fl (Russell & Norwig 2009 ) that can perform human -like tasks such as playing games of chess or driving cars. Some of the methods developed by AI re searchers have been combined with traditional methods of statistics to produce methods for fipattern recognitionfl (Ripley 1995), fimachine learningfl (Bishop 2011) and fistatistical learningfl (Hastie et al. 2016 ). During the first decade of the 21 st century the need for better ways to process and use data, especially in the sciences, were discussed under the rubric of ficyberinfrastructurefl (Atkins et al. 2003; Berman & Brady 2005), but more recently the terms fiBig Datafl and fiData Sciencefl have become popular. fiBig Datafl — For those of us who remember when computer memories were measured in kilobytes instead of terabytes (a factor of a billion more ), fiBig Datafl seems like a moving target, especially given Moore™s filawfl which successfully predict s the doubli ng of the number of transistors per square inch every eighteen months , but the term has arisen despite the advances in computer power because data seem to be growing faster than our ability to process them. The total volume in number of bytes, the variety in terms of text, images, audio, video, sensor, social media, and other forms of data, and the
PAGE – 6 ============
6 daily velocity (Laney 2001) of data are growing even faster than computing power. The large volume leads to problems of s toring and managing data. The growth in terms of variety adds the difficulties of translating data from one form to another , and the growth in terms of velocity leads to the need to edit data fion -the -runfl and to choose what is important. More recently a fourth concern, checking on the veracity of the data, add s another layer of complexity on top of volume, variety, and velocity. Size, complexity, and technological challenges provide one definition of big data (Ward and Barker, 2013 ; National Research Cou ncil 2013 ), but they do not seem like a sufficient basis for heralding a sea -change in our data environment since the race between data set size and computer capabilities goes back to the advent of computing. The National Institute of Standards and Technol ogy has more usefully proposed that fifundamentally, the Big Data paradigm is a shift in data system architectures from monolithic systems with vertical scaling (i.e., adding more power, such as faster processors or disks, to existing machines) into a paral lelized, ‚horizontally scaled™, system (i.e., adding more machines to the available collection in order to deal with volume, variety, and velocity ) that uses a loosely coupled set of resources in parallel (NIST 2015, 5). fl But the statistician David Donoho objects that fithe new skills attracting so much media attention are not skills for better solving the real problems of inference from data; they are coping skills for dealing with organizational artifacts of large -scale cluster computing (Donoho 2017, 747 ).fl We also do not know whether this new architecture is permanent or transient. Beyond the sheer amount of data the truly distinguishing features of the Big Data revolution are the new technologies for recording, connecting, networking, and creating information . Human interactions through phone calls, e -mail, texts, tweets, social media posts, and other technological methods are now digitally recorded, time and location -stamped , and attributable to nodes in networks in ways that go far beyond the much more ephemeral media of the past. Many business, governmental, social, and scientific tasks now have digital trails such as Fed -Ex tracking services, Web searches and purchases, parking meter payments and automobile trips, tax payments, photographs of social gatherings, weather and environmental measurements, digital images from microscopes and telescopes, and much more. When combined with the fact that the World Wide Web is an e xcellent site for social networks and accessing information and that computers can now author information and interact with us Šperhaps even producing artificial intelligence and autonomous robot -like entities and virtual realities Š a picture emerges less o f fibig datafl than of fiimmersive datafl that surrounds us and affects our lives on a daily basis. The fidecentralization of datafl identified by NIST may also be more than just a set of techniques for dealing with large computing problems, but the future shap e of computing and the internet is still not clear. Consequently, the real impact of the big data revolution is not so much the amount of data as a change in our cognitive environment (Neumann 2016; Lugmayr et al 2017 ; Schroeder 2018 ) that requires new pe rspectives to deal with datification, connectedness, networking, and computer authoring. These phenomena stem from the invention of new technologies including innovative methods in data science . fiData Sciencefl Œ The companion idea of fidata sciencefl relies less on the scale of the data than on a definition of a way to discover new knowledge in an age when data have proliferated and cry out for analysis. In 2001, the statistician William S. Cleveland put forth a pl an to fienlarge the major areas of technical work in the field of statisticsfl (Cleveland 2001, 21) by providing more resources for ficomputing with datafl (22) and to call the new field fidata science.fl In an address to the Computer Science and Telecommunicat ions Board of the National Research Council in 2007 (Gray 2007 in Hey et al. 2009) , computer scientist Jim Gray advocated for fidata -driven sciencefl as a new scientific paradigm
PAGE – 8 ============
8 5.5 Data archiving, indexing, and search and data governance such as standards for open data and reproducibility ; determining rules for access and privacy protection where necessary. fi6. Science about Data Science fl [e.g., ways that people do data science and the impacts of data science and big data on society ].fl (Donoho 2 017, 755). Judging from this list , data science borrows methods and techniques that go beyond the traditional core of statistics which is largel y encompassed in fi 4. D ata Modeling.fl Techniques of data gathering and preparation are typically taught in subject matter disciplines even though statistics started as an endeavor to collect data on the state and its people through censuses and surveys. Computer science and other academic departments deal with data representation and transformation and with computin g with data. Data visualization and presentation often involve media laboratories and psychology departments. Data archiving, indexing, and availability form the core of work in s chools of library science an d their modern incarnations as schools of information. In one subject matter area, bioinformatics, more than 100 colleges and universities now offer programs that focus on these tasks, and there are a few digital humanities, social sciences, and environmental science programs. But at the moment it seems that the most popular way to move forward in this area is to create fidata sciencefl programs including computer science, information, and statistics which allow for relationships with subject matter disciplines. The unsolved problem is the exact way that the applied data science being done in these disciplines can be incorporated into these programs. For example, in addition to benefittin g from using data science and big data, t he social sciences can provide fundamental help in understanding the social construction and meaning of data, the causal impact of new information technologies, the ethical issues of privacy and data ownership, and the best ways for social institutions to use cyberinfrastructure (Berman & Brady 2005). Data science must encompass these issues . However universities organize themselves to deal with these seven tasks, the following seems clear to me. The explosion in the number of methods and techniques for undertaking the tasks listed above means that there must be some way for universities to bring together the people working on them to learn from one an other and to be able to teach the next generation of students and scholars what they need to know to use them. There must also be some way to help scholars, either through collaboration with other scholars or by having specialists akin to collections spec ialists in libraries or museums, to use the many kinds of data, software, and techniques that are now available. Gone are the days where someone could learn , as I did, about a few kinds of data collection (e.g., surveys , content analysis, and administrati ve data ), some FORTRAN and subroutine libraries such as NAG and IMSL , a bit about dBase and SQL, some statistics through econometrics and psychometrics and some statistics packages such as BMD P-SPSS -SAS-STATA -GAUSS , and a few other things and be at the for efront of data science in their discipline. There is just too much to be learned. Real Phenomena, Inadequate Language Œ Many of the developments related to big data and data science are not new, but they have achieved a scale and level of impact that require new ways of describing them. The right language, however, is hard to find. J ust as the transportation revolution was not just about the steam engine Šit also involved the discovery of new forms of energy (oil and electricity) , the invention of new k inds of motors (internal combustion and electrical) , the creation of networks of rails, roads, and rivers, and even the development of new social norms such as standard time zones Šthe information revolution is more than just computer s or any other single t hing. It also involves sensors, data bases, programming languages, artificial intelligence, telecommunications, machine learning, social media, the Internet, and many other inventions. Neither fibig datafl nor fidata
PAGE – 9 ============
9 sciencefl nor any other words or phrases encompass all these in novations. The term cyberinfrastructure might have been a useful one, but it has not caught on. fiArtificial Intelligencefl is too limited. One leading data science scholar (Jordan 2018) argues for the use of the term fiIntelligent infrastructure,fl but it also has its limitations. We are left with real phenomen a but in adequate language. Societal and Political Change from Big Data and Data Science Many authors have provided overviews of areas that are being affected by fibig datafl (Chen at al. 2012; Cukier & Mayer -Schoenberger 2013, 2014; Evans 2018; Mosco 2014). We cannot provide an exhaustive review of the possible societal impacts of big data and data science, but it is worth listing a few prominent examples to show how they deserve more scrutiny b y political scientists. I have chosen cyberwarfare and homeland security, smart cities, medicine , and the media . Several recent books have proposed that ficyber war farefl is here and a threat to internation al security (e.g., Clarke & Kanke 2011; Kaplan 2017), but skeptics (Rid 2012 , Libicki 2014 ) have argued that while cyber disruptions may be a problem, they do not constitute classical warfare like the Japanese attack on Pearl Harbor on December 7, 1941 which involved a purposeful and publicly claimed act of violence for political advantage . Some of the leading examples of fiCyber Warfl such as the Stuxnet virus™s introduction i nto Iranian centri fuges (an essential part of Iran™s nuclear fuels enrichment program) that led to their destruction or the massive denial of service attack (presumably by Russian hackers) on Estonia in April 2007 were almost surely purposeful but at most they caused lost productivity and perhaps p roperty damage . Most importantly, no state claimed responsibility in order to achieve direct political advantage . Although the case for cyberwarfare may be weak , the web has certainly been used for fi sabotage, espionage, and subversionfl (Rid 2012, 5) as recent events involving Russia and the 2016 American election make clear (Sanger 2018) . Moreover, the American military is collecting and processing a flood of sensor and digital information (Porche et al. 2014) which could change the face of conflict (Dunlap 2014). Obviously, these developments get at the heart of political science studies of international relations and security. fiSmart Citiesfl is a popular book title with subtitles such as fiBig Data, Civic Hackers, and the Quest for a New Utopia,fl fiA Spatialized Intelligencefl and fiThe Internet of Things, People, and Systems fi (Townsend 2013, Picon 2015, Dustdar et al. 2017) . Three streams of big data work come together in this area. First, there are large digitized administrative datasets on people and their relationship to schools, social welfare agencies (Brady et al. 2001), medical care, and police , and there are similar datasets on physical structures and their relationship to streets, services, land -use and zoning. Second , the reduced costs of sensors , wireless networks, video cameras and the ability to connect them with a n fiInternet of Thingsfl makes it possible to monitor and sometimes remotely control air pollution, traffic, e lectricity usage, utilities, parking, safety, water usage, police and fire deployments, and many other aspects of a modern city. Third, Internet dat a such as Google Street View, Zillow, Airbnb, or Yelp can provide information about businesses, real -estate, and the physical condition of the city (Glaeser et al. 2018). These data can be linked by geocoding the location of each person™s house (or place of work), each structure or business , and each sensor. Increasing ly, we can go farther and link data through recognition of vehicle s, face s, or RFID tags which makes it possible to track m ovements throughout the city (Hashem et al. 2016) Using these data, the city and its operations can be descr ibed, managed, and evalu ated. Traffic, air pollution, or poverty map s can provide useful descriptions for those trying to understand where to live,
PAGE – 10 ============
10 where to travel, or what to do. Improve d conditions on a real -time basis can be managed by involving citizens in constant feedback on services, changing the timing of traffic lights, deploying police to areas with disturbances, asking industries to fispare -the -airfl by reducing some activities , and so forth. Finally, evaluation results can indicate what is working and what is not so t hat processes can be improved. Because the decision s about what data are collected, how they are processed, and how they are used all involve choices, often influenced by who has power and who does not, these systems are inherently political, and they can easily become technocratic, overly influenced by cor porate interests, and perhaps most alarmingly, the basis for the fipanopticfl city Œ the urban counterpart of Jeremy Bentham™s circular Panopticon , a prison in which all inmates were constantly visible to a centrally located guard station (Kitchin 2014). fiToward Precision Medicine,fl a 2011 report of t he National Research Council of the National Academy of Sciences defined precision medicine as fithe tailoring of medical treatment to the individual characteristics of each patient fl (125) . To practice precisio n medicine, information about the individual must be combined with medical knowledge about how people vary in their response to illnesses and treatments (Dzau et al. 2016) . Individual information would come from electronic medical records and genomic dat a. The 2011 report suggested creating a new taxonomy of human disease based upon molecular biology that would serve as the starting place for classifying diseases and people™s reaction s to them . To do this, an fiinformation commonsfl would be created tha t linked molecular data, medical histories, and health outcomes (Beachy, Olson, Berger 2015), and these data would be used to explore clinical associations (Hanauer et al. 2009; Miller 2011 -12). These data could be a great boon to medical researchers, bu t they raise significant questions about privacy , ownership of data, and their relationship to issues such as race in America (Hochschild & Sen 2015) that could become high -profile political issues. Changes in the media from the rise of the Internet are now manifestly important for politics, but political scientists have lagged in their awareness of them . In the first examination of the mass media in the Annual Review of Political Science in 2002 , Michael Schudson quite properly t akes political science to task because it fihas never extended to the news media the lovingly detailed attention it has lavished on legislatures, parties, presidents, and prime ministers.fl (Schudson 2002, 249). Yet he does not even mention the Internet or World Wide Web , and he focuse s on the relative merits of state versus commercial controlled media, journalism as fithe story of the interaction of reporters and government officialsfl ( 255), and the cultural norms that shape coverage of topics such as homosexuality and crime. Schudson concludes that fiThe news media have always been a more important forum for communication among elites (and some elites more than others) than with the general populationfl (263) . Not even a hint comes through about the possible anarchy of uncontroll ed news fisourcesfl and direct leader -follower communications now bedeviling a world with Facebook, Google, and Twitter . By 2012, Farrell™s Annual Review article recognizes the potential importance of the Internet for exacerbating political polarization or facilitating the Arab Spring, and he argues that the Internet could sort citizens into homogeneous groups seeking information to confirm their ideological biases, discourag e preference falsification in authoritarian regimes by making available a broader a rray of opinions, and overcom e the costs of collective action by allowing like -minded and intense people to find one another . Although Prior still concludes in h is 2013 Annual Review article on fiMedia and Political Polarizationfl that fiInternet use shows few signs of ideological segregation (Prior 2013, 122), fl he takes the Internet seriously. And communications theorists such as Bennett and Segerberg (2012), Neumann (2016) , and Schroeder (2018) argue for developing new models to understand the new media on the
22 KB – 31 Pages