Listening to Tarleton Gillespie’s recent talk, “Algorithms, and the Production of Calculated Publics” got your humble blogger thinking about alternatives to phone-based opinion polling.
Gillespie had a valuable point to make about the ill-placed faith we have in information created by proprietary algorithms and derived from proprietary internet use information. He rattled off all sorts of ways the providers of algorithmically produced information routinely manipulate inputs and formulas to influence the results, especially to suppress pornographic and suggestive material.
Gillespie was speaking an MIT Comparative Media Studies/Writing session. He is an, Associate Professor, Department of Communication, Cornell University, and currently a visiting researcher with Microsoft Research, New England. He is the co-editor of Media Technologies: Essays on Communication, Materiality, and Society (2014), and the author of Wired Shut: Copyright and the Shape of Digital Culture (2007), He is also the co-founder of the scholarly blog, culturedigitally.org.
Search rankings and keyword autocomplete rankings are probably the most pervasive algorithms in our daily digital experiences. Gillespie focused more on the increasing use of “trending” reports produced by algorithms on platforms such as Twitter, You Tube and Facebook. He likened trending reports to scientific opinion polling data, which popular media started reporting 100 years ago. The science of opinion polling had developed in marketing research before it became a media product and political tool.
Gillespie mused about the aura of algorithmically produced information. He wondered whether these algorithm-driven information products would withstand scrutiny overtime and what associations of people and ideas our digital records might yield in the future.
Algorithm is a fancy word for a set rules of calculation that yield a result. Algorithmically produced information isn’t new. Algorithms provide all sorts of useful distillations of data. Standard algorithms create statistics such as median, average, etc., routinely cited in all kinds of media.
Gillespie did not delve into the ways in which we judge the validity of polling data. Those of us who understand polls and have done polls and other forms of statistical work, look at the reputation of the polling organization, at sample sizes and selection techniques, questions, survey methods, and detailed tabulations before accepting sweeping conclusions from a poll.
An issue in most trending by algorithm is that if the algorithms details are revealed, interested parties may be able to influence the results. There is also a Wizard of Oz factor: behind the screen of the scientific sounding algorithm may be a set of mundane or inane calculations that don’t justify trust.
However, if we look at how market researchers are now using internet-generated data, adjusting for sample bias etc. often in an ad-hoc manner to reach actionable and profitable conclusions, we are probably looking at public opinion tracking and trending of the future. With increasing privacy concerns and decreasing use of land-line phones, representative opinion samples are getting more expensive to assemble by phone. As the science of online opinion research improves, the results will get better and conventions of credibility-supporting disclosure will take root.
Opinion trending seems like an obvious product to generate from a larger news site. Within constituent groups, voluntarily opted into campaign communication, on-line trending tools seem readily applicable and testable against other means of gauging opinion. However, as trending earns respect, it will be important to distinguish online opinion trending from online governance. While constituents subject to online trending may feel like they are participating in governance, online governance is a separate and very promising concept for modern campaign organizations.