Times, Defending Dissent, News Taco, Big
Education Ape). There are a few mid-sized
mainstream outlets (The Nation, Ms. Magazine, Public Radio International), and one
large national outlet [Huffington Post, the
11th largest news site in the world by traffic
according to Alexa (8)]. The authors cannot disclose separate results by outlet for
confidentiality reasons, but they show supporting evidence suggesting that no single
outlet individually drives the results.
That relatively obscure outlets could produce such large effects is surprising, and
it suggests that these results may capture
just a tiny piece of the influence wielded by
media outlets as a whole. One interesting
piece of observational data supports this
view. The authors examine some examples
of stories produced by The New York Times
on previously little-discussed topics, such as
a story about fracking affecting the quality
of drinking water published at a time when
there was little discussion of this issue.
They find resulting spikes in Twitter traffic
an order of magnitude larger than their experimental estimates.
The second important detail is the nature
of the outcomes. As the authors are careful
to point out, posts on Twitter are far from a
representative slice of the “national conversation.” Three-quarters of Americans do not
use Twitter, and only 10% use it on a daily
basis (9). Those who do are more educated
and have higher incomes than the average
American (9). This means that the impact
this study measures could in some ways be
especially important, as Twitter includes a
particularly large share of elite influencers
with the potential to translate their views
into policy. That the impacts are similar for
posts by heavy and light Twitter users also
suggests that the effects may reach beyond
the elites. Nevertheless, it remains unclear
whether this kind of journalism reaches the
conversations of the broad mass of American voters.
In addition, the main focus of the study
is on the volume of discussion about different topics on Twitter, not the quality or
downstream impact of this discussion. It is
possible, for example, that users with different political leanings or genders differed in
the depth of their engagement with the topics or in the impact their comments had on
SETTING THE AGENDA
The results of King et al. speak to a long
body of work on the channels by which
media exert influence. Early media studies,
motivated in part by the seemingly limit-
less power of propaganda during the World
Wars, looked for simple persuasive effects:
for example, exposure to a conservative
message should make a recipient more con-
servative. The results were mostly negative,
leading scholars to question the power of
media and look for other channels of influ-
ence (10–12). One of these was agenda set-
ting: the idea that media might affect what
issues the public and policy-makers focus
on, even if media could not change how
they thought about these issues (13, 14).
It turns out that the early failures to find
persuasive effects were due more to limita-
tions of the research designs than to lim-
its on the power of media. Teasing out the
causal effect of media content from obser-
vational data is difficult, and biases in sim-
ple correlational studies can be extreme.
More recent studies have used carefully
constructed natural experiments to show
large persuasive effects of media in many
contexts [e.g., (15)]. However, the origi-
nal insight that agenda setting provides a
separate, important channel of influence
remains valid. King et al.’s study provides
one of the most rigorous and convincing
data points to date on the agenda-setting
power of media.
More broadly, these results echo a num-
ber of points from prior literature that push
back against the prevailing narrative about
trends in media and politics. We know
that the polarization of voters has in some
ways been overstated. Views on individual
policy issues, for example, have remained
fairly stable over time with most Americans
holding moderate views (16–18). The claim
that American voters have increasingly
self-segregated geographically has been
largely debunked (19, 20). We know that the
extent of ideological segregation in news
consumption—i.e., the extent to which the
sources and conversations conservatives
are exposed to are disjoint from those that
liberals are exposed to—has at least until re-
cently been substantially lower than much
of the popular discussion would suggest (21,
22). Demographic patterns suggest that to
the extent we see evidence of rising polar-
ization, it is concentrated among the groups
least exposed to online news and informa-
tion (23), suggesting that the polarizing ef-
fect of new media may be more limited than
Although King et al. build upon such
prior literature, several features make their
study stand out. The basic design of ran-
domizing media content has never been
tried before at this scale, and implementing
this design was by all accounts a formidable
effort. Studying impacts on social media
discussions distinguishes their study from
others estimating causal effects of media
content, most of which look at other out-
comes like voting [e.g., (15)]. The analysis
combines a sophisticated approach to sta-
tistical inference with cutting-edge text
The methodology of the study by King et
al. may open up new avenues for research.
The same experimental design could in
principle be extended to look at how the
media influence discussion in forums be-
yond Twitter; how they affect downstream
outcomes such as individuals’ information,
beliefs, or votes; and the extent to which
timely release of accurate information can
neutralize misinformation or biased beliefs.
Taken as a whole, the results of King et
al. provide a timely reminder that there are
positive as well as negative aspects of new
technology. At the current moment, it is
easy to forget that a predominant concern
for most of American history has been that
media power would be concentrated in a
few hands, and that only a few outlets would
have the ability to shape the nation’s agenda.
Although social media gives a platform to
many objectionable voices, it also makes it
possible for journalists at innumerable small
outlets to participate in the conversation as
well. The results of King et al. suggest that
when they speak, many are listening. j
1. A.de Tocqueville,Democracyin America, Transl.I.
Kramnick, G. E. Bevan (Penguin Classics, London, 2003).
2. A. Downs, J. Polit. Econ. 65, 135 (1957).
3. G. S. Becker, J.Law Econ. 1, 105 (1958).
4. G. King, B. Schneer, A. White,Science ,358, 776 (2017)
5. H. W. Wat, J. Mass Commun. Q. 73,40(1996).
6. P. M. Napoli. J. Commun. 49, 151 (1999).
7. Associated Press v. United States , 326 U. S. 1 (1945).
8. Alexa Top News Sites: www.alexa.com/topsites/category/
9. S. Green wood, A. Perrin, M. Duggan, Social Media Update
2016 (Pew Research Center Internet and Technology
Report, 2016); www.pewinternet.org/2016/11/11/
10. P. F. Lazarsfeld, B. Berelson, H. Gaudet, The People’s
Choice: How the Voter Makes Up His Mind in a Presidential
Campaign (Columbia Univ. Press, New York, 1944).
11. B. R. Berelson, P. F. Lazarsfeld, W. N. McPhee, Voting: A
Study of Opinion Formation in a Presidential Campaign
(Univ. Chicago Press, IL, 1954)
12. J. T. Klapper, The Effects of Mass Communication (Free
Press, Glencoe, IL, 1960).
13. M. E. McCombs, D. L. Shaw, Public Opin. Q. 36, 176 (1972).
14. M. McCombs, Setting the Agenda: Mass Media and Public
Opinion (Polity Press, Cambridge, UK, ed. 2, 2014).
15. S.DellaVigna,E.Kaplan, Q.J.Econ. 122,1187(2007).
16. M. P. Fiorina, S. J. Abrams, J. C. Pope, Culture War? The
Myth of a Polarized America (Longman, New York, 2006).
17. S. J. Abrams, M. P. Fiorina, “The big sort” that wasn’t:
A skeptical reexamination. PS Polit. Sci. Polit. 45, 203
18. M. P. Fiorina, S. J. Abrams, Annu. Rev. Polit. Sci. 11, 563
19. E. L. Glaeser, B. A. Ward, J. Econ. Perspect. 20, 119 (2006).
20. M. P. Fiorina, “Has the American public polarized?” A
Hoover Institution Essay on Contemporary American
Politics, Ser. 2 (2016).
21. M. Gentzko w, J. M. Shapiro, Q. J. Econ. 126, 1799 (2011).
22. S. R. Flaxman, S. Goel, J. M. Rao, Public Opin. Q. 80, 298
23. L. Boxell, M. Gentzko w, J. M. Shapiro, Proc.Natl.Acad.Sci.
U.S.A. 114, 10612 (2017).
10 NOVEMBER 2017 • VOL 358 ISSUE 6364 727