CHAPTER 1. RETHINKING THE ATTENTION ECONOMY
1. For a full discussion of Google’s early experiments, see Mayer, 2007. Mayer has used slightly different figures in describing the experiments’ impact in other speeches.
2. Subsequent experiments have shown that it can take weeks or months for slowed-down users to return to their previous level of usage; see Hölzle, 2012.
3. Berners-Lee, 2000, p. 23.
4. Ingram, 2017.
5. Simon, 1971.
6. Goldhaber, 1997.
7. Ibid.
8. J. Webster, 2014, p. 1.
9. See discussion in Krugman, 1997, pp. 52–55.
10. Ibid., p. 54.
11. von Thunen [1826], 1966.
12. Blumler and Kavanagh, 1999.
13. Negroponte, 1995, p. 58.
14. Reynolds, 2006.
15. Mele, 2013.
16. Sifry, 2009.
17. Bai, 2009.
18. Morton, 2011.
19. Benkler, 2006, pp. 3–4.
20. Shirky, 2009, pp. 59–60.
21. Rosen, 2011.
22. Hindman, 2009, pp. 86–87.
CHAPTER 2. A TILTED PLAYING FIELD
1. Schwartz, 1998.
2. See discussion in chapter 1, especially Benkler, 2006.
3. AT&T, 1908, p. 21.
4. See, for example, O’Reilly, 2005; Wu, 2011.
5. The major exception, the Chinese site Weibo, illustrates the same logic: the fact that social networks in China and the United States barely overlap has allowed a different firm to seize this niche in China. Other internet services, from instant messaging to Skype, show the same pattern.
6. E.g. Pariser, 2011, p. 41.
7. Hundt, 1996.
8. See, for example, Briscoe, Odlyzko, and Tilly, 2006.
9. Cialdini and Goldstein, 2004; Asch, 1955.
10. Kirkpatrick, 2011, p. 101.
11. Ksiazek, Peer, and Lessard, 2014.
12. Sonderman, 2012.
13. Toth, 2014.
14. Sonderman, 2011.
15. Bain, 1954, 1956; Haldi and Whitcomb, 1967.
16. A. D. Chandler, 1977.
17. Pearn, 2012; see also Munroe, 2013.
18. Ghemawat, Gobioff, and Leung, 2003.
19. Burrows, 2006; Lo et al., 2015.
20. J. Dean and Ghemawat, 2008; Chang et al., 2008.
21. Bhatotia et al., 2011.
22. Shute et al., 2012; Corbett et al., 2012.
23. Verma et al., 2015, p. 1.
24. On Google’s acquisition of UK-based machine learning startup DeepMind, see “What DeepMind brings to Alphabet,” 2016. Access to Google’s computing power was reportedly a key factor in why DeepMind agreed to be acquired by Google. On TensorFlow, see Abadi et al., 2016.
25. Jouppi et al., 2017.
26. S. Levy, 2012; McMillan, 2012.
27. TeleGeography, 2012.
28. Labovitz et al., 2009.
29. Google, 2013.
30. DeepMind, 2016.
31. McKusick and Quinlan, 2009.
32. Mayer, 2007.
33. Hölzle, 2012.
34. Schurman and Brutlag, 2009.
35. Artz, 2009.
36. Hölzle, 2012.
37. Singhal and Cutts, 2010.
38. Bowman, 2009.
39. On the relation between site design, traffic, and revenue, see Flavián et al., 2006; Cyr, 2008.
40. Kohavi et al., 2013.
41. Tang et al., 2010.
42. Kohavi et al., 2013.
43. McKusick and Quinlan, 2009.
44. Pike et al., 2005; Chattopadhyay et al., 2011; Melink et al., 2010.
45. Palmer, 2002.
46. Bohn and Hamburger, 2013; see also Buchanan, 2013.
47. Brian, 2014.
48. Kohavi et al., 2013.
49. Ibid., p. 4.
50. Davis, 1921, p. 169.
51. Nelson, 1970; Shapiro, 1983.
52. Hoch and Deighton, 1989.
53. Shaprio and Varian, 1998, pp. 113–14, emphasis added.
54. Wells, Valacich, and Hess, 2011; McKnight, Choudhury, and Kacmar, 2002.
55. H. E. Krugman, 1972; Tellis, 2003, 1988.
56. Lambrecht and Tucker, 2013.
57. Semel, 2006.
58. Yarow, 2013.
59. Hargittai et al., 2010.
60. Jansen, Zhang, and Mattila, 2012, p. 445.
61. Jansen, Zhang, and Schultz, 2009.
62. Pan et al., 2007. Another interpretation of these results might be that users have entrenched habits and site-specific skills, as we will discuss later.
63. Microsoft has removed the initial study from the web. Note that even if Microsoft consistently performed worse on this challenge, it might benefit from this sort of direct comparison. If Microsoft wins 40 percent of the time (for example), but has only 30 percent market share, more Google users will be told they chose Bing than the other way around. Note that this assumes that an individual’s choice is not related to his or her current browser usage, which is likely false in real life, as browsers increasingly learn from users’ past behavior.
64. Ataullah and Lank, 2010, p. 337.
65. Ayres et al., 2013.
66. Zara, 2012.
67. Ataullah and Lank, 2010, p. 337.
68. Iyengar and Hahn, 2009.
69. Stroud, 2011.
70. Hargittai and Shaw, 2015.
71. Hargittai, 2010.
72. Hargittai et al., 2010.
73. Stigler and Becker, 1977; Wernerfelt, 1985, 1991.
74. On this point see, for example, Wernerfelt, 1991, p. 232.
75. Shapiro and Varian, 1998.
76. Ajax is an acronym for Asynchronous JavaScript and XML.
77. J. J. Garrett, 2005.
78. E. Johnson, Bellman, and Lohse, 2003; see also Murrary and Häubl, 2007, p. 62.
79. See discussion in Brynjolfsson and Smith, 2000.
80. Shankar, Smith, and Rangaswamy, 2003; Ha and Perks, 2005; Aksoy et al., 2013.
81. Murray and Häubl, 2007.
82. Quoted in Beam, 2010.
83. Ibid.
84. Schmidt, 2014.
85. Wheeler, 2013.
86. Arthur, 1989; David, 1985.
CHAPTER 3. THE POLITICAL ECONOMY OF PERSONALIZATION
1. Negroponte, 1995, p. 153.
2. See, for example, Kennard Gates, 2000; Kennard, 1999.
3. Sunstein, 2001, 2009.
4. Schafer, Konstan, and Riedl, 2001.
5. Zuckerberg quote from Pariser, 2011.
6. Zelizer, 2009.
7. Bucy, 2004; Deuze, 2003; but see Stromer-Galley, 2004.
8. Thurman and Schifferes, 2012; Thurman, 2011.
9. See, for example, Haim, Graefe, and Brosius, 2017, and Möller, Trilling, Helberger, and van Es, 2018.
10. Hindman, 2018.
11. Stigler, 1961, p. 216.
12. Ibid., p. 220.
13. Mayer-Schoenberger and Cukier, 2013.
14. Netflix, 2007.
15. This chapter’s overview of AT&T’s participation in the contest draws heavily from from the team’s official history (AT&T 2009, 2010), and from Yehuda Koren’s (2009) recounting of the contest.
16. Funk, 2006.
17. Ibid.
18. Ibid.; see also Gorrell, 2006.
19. AT&T, 2009.
20. AT&T, 2010.
21. AT&T, 2009, emphasis in original.
22. AT&T, 2009.
23. Hunt, 2010.
24. AT&T, 2009.
25. Netflix, 2007.
26. Amatriain and Basilico, 2012.
27. Ibid.
28. Koren, 2009.
29. Banko and Brill, 2001, p. 28.
30. Funk, 2006.
31. Koren, 2009.
32. Pariser, 2011.
33. Amatriain and Basilico, 2012.
34. Ibid.
35. Das et al., 2007.
36. Ibid., p. 271.
37. Ibid., p. 279.
38. Liu, Dolan, and Pedersen, 2010.
39. Ibid., p. 32.
40. Kirshenbaum, Forman, and Dugan, 2012, p. 11.
41. Boyd, 2011.
42. Pandey et al., 2011.
43. Ibid.
44. Ibid., p. 3.
45. Hindman, 2018.
46. Cadwalladr and Graham-Harrison, 2018.
47. Kosinski, Stillwell, and Graepel, 2013.
48. Frier, 2018.
49. Negroponte, 1995, pp. 57–58.
50. See discussion in Neuman, 1991.
CHAPTER 4. THE ECONOMIC GEOGRAPHY OF CYBERSPACE
1. Ohlin, 1935.
2. Krugman, 1979, 1980.
3. E.g., Dixit and Stiglitz, 1977.
4. Pai, 2017; see also Faulhaber, Singer, and Urschel, 2017.
5. Pooley and Winseck, 2017; see also chapter 6.
6. Box, 1979.
7. For example, see Steiner, 1952; Negroponte, 1995; see also the discussion later in this chapter.
8. For an overview of research on bundling see Adams and Yellen, 1976; Shapiro and Varian, 1998; Bakos and Brynjolfsson, 1999.
9. Adams and Yellen, 1976
10. This example loosely adapted from Shapiro and Varian, 1998; see also a similar example of bundling in Hamilton, 2004.
11. Hamilton, 2004.
12. Carroll, 2008.
13. Bakos and Brynjolfsson, 1999.
14. On how bundling digital products is different, see Bakos and Brynjolfsson, 2000.
15. Nalebuff, 2004.
16. Chandler, 1964; Flink, 1990, but see Raff, 1991.
17. Peles, 1971, p. 32.
18. Ingram, 2017.
19. Meyer, 2004, p. 45.
20. Lewis and Rao, 2015; Johnson et al., 2016.
21. Brodersen et al., 2015.
22. Pandey et al., 2011; see discussion in chapter 3.
23. Mutter, 2012.
24. Last, 2002.
25. See, for example, Bagdikian, 1985.
26. C. Anderson, 2004.
27. Steiner, 1952; Beebe, 1977—though see J. G. Webster and Wakshlag, 1983 for an influential critique. For an older but still excellent overview of economic models of program choice, see Owen and Wildman, 1992.
28. Ehrenberg, 1968; Kirsch and Banks, 1962.
29. Goodhardt and Ehrenberg, 1969.
30. J. Webster, 2014, p. 30.
31. Koren, 2009.
32. Prior, 2006.
33. Boczkowski and Mitchelstein, 2013.
34. Stroud, 2011; Iyengar and Hahn, 2009; R. K. Garrett, 2009.
35. Levendusky, 2013; see also Gentzkow and Shapiro, 2011.
36. J. Webster, 2014; Ariely and Norton, 2008.
37. Monsell, 2003; Kahneman, 2011.
38. Krug, 2013.
39. Somaiya, 2014.
40. Hotelling, 1929; see also Downs, 1957.
41. Unequal browsing time also limits portal sites’ production. Additional content no longer automatically increases profitability, because at high levels of production fewer and fewer users have the time budget needed to consume additional content.
Alternatively, we might reintroduce narrow preference windows, making it so that some users get no utility from the middle-of-the-road content produced by portal sites. However—as we have already seen in this chapter—this change arguably makes the model less rather than more realistic.
42. Tankersley, 2015. Note also Steve Wildman’s (1994) work on “one-way flows” in news production.
43. Benton, 2016.
44. Cairncross, 2001.
45. Hindman, 2009.
46. Athey, Mobius, and Pál, 2017.
47. P. Krugman, 2009.
48. D. Dean et al., 2012.
CHAPTER 5. THE DYNAMICS OF WEB TRAFFIC
1. Glaeser, 2005.
2. On this point, see Hindman, 2009.
3. Dahlgren, 2005; Benkler, 2006; Hindman, 2009; Meraz, 2009; Caldas et al., 2008.
4. Barabási and Albert, 1999.
5. Newman, 2005; Clauset, Shalizi, and Newman, 2009.
6. For early examples of this, see Goel and Richter-Dyn, 1974.
7. Rioul and Vetterli, 2002.
8. Anderson and Mattingly, 2007.
9. Caldentey and Stacchetti, 2010.
10. Volz and Meyers, 2009.
11. Redner, 1998.
12. Small and Singer, 1982.
13. Levy and Solomon, 1997.
14. Gabaix, 1999.
15. Fernholz (2002) notes that the stock market deviates slightly from a power law at both the extreme head and at the tail of the distribution. The very largest firms are slightly smaller than mathematical models predict, which Fernholz attributes to antitrust laws and real-world limits. Concentration at the very top of the market has been increasing, though—perhaps because of weakened antitrust enforcement (more on that in chapter 8). At the other end of the market, many small firms stay privately owned instead of becoming public companies, shortening what would otherwise be a longer tail.
16. Fernholz, 2002, p. 95.
17. PriceWaterhouseCoopers, 2008.
18. Interactive Advertising Bureau [IAB], 2010.
19. Companies such as Chartbeat can provide reasonable time measures across different sites that use their instrumentation, but can measure only behavior on their partners’ sites. This makes their data useful for analysis within the network of affiliates, but still limited in key ways (more on the time-measurement approach and comScore’s data in the next chapter).
20. See Meiss et al., 2008; E. Johnson, Lewis, and Reiley, 2003.
21. The key difference between power law and lognormal distributions, interestingly enough, has to do with what happens to observations that go to zero. If observations that go to zero (or the lower bound) are replaced, then the distribution becomes a power law; if they are not replaced the distribution becomes lognormal. See Mitzenmacher, 2004; Gabaix, 1999.
22. Clauset et al., 2009.
23. Fernholz, 2002.
CHAPTER 6. LESS OF THE SAME: ONLINE LOCAL NEWS
1. This account of the founding of Patch.com is taken from Carlson, 2013.
2. Carr, 2013.
3. Romenesko, 2013.
4. Jack Marshall, 2016.
5. Carr, 2013.
6. Jarvis, 2013.
7. Mitchell, Gottfried, Barthel, and Shearer, 2016.
8. See, for example, Noam, 2009.
9. Prometheus v. FCC, 2004, is one prominent example.
10. Kirchoff, 2010.
11. Olmstead, Mitchell, and Rosenstiel, 2011.
12. Graves, 2010.
13. Cook and Pettit, 2009.
14. Boczkowski, 2010.
15. Olmstead, Mitchell, and Rosenstiel, 2011.
16. See, for example, Hindman, 2009.
17. Department of Justice and Federal Trade Commission, 2010.
18. Because HHI attempts to assess firms’ market power, I combine the market shares of multiple outlets in the same market owned by the same firm. For example, page views on the Atlanta Journal-Constitution site, the WSB-TV site, and the WSB-radio site are all summed together.
19. Kopytoff, 2011.
20. Schaffer, 2010.
21. Pew, 2010.
22. On this point, see Gelman, 2010.
23. Negative binomial models are closely related to poisson regression models. Whereas poisson models have only a single parameter λ, which governs both the mean and the variance of the distribution, negative binomial models add the parameter α to capture overdispersion. While overdispersed count data are often the norm in the social sciences, in this case both models produce estimates of α very close to zero. When α = 0, as it does here, poisson and negative binomial models are identical.
24. Greenslade, 2012.
25. Morton, 2011.
26. Waldfogel, 2002.
27. Chyi and Tenenboim, 2017.
28. FCC 2017, p. 87.
CHAPTER 7. MAKING NEWS STICKIER
1. Pew Internet and American Life Project, 2017b.
2. Graves, Kelly, and Gluck, 2010.
3. Kanagal et al., 2013; Pandey et al., 2011.
4. Pew, 2010.
5. Usher, 2014a.
6. Usher, 2014b.
7. Mutter, 2009.
8. New York Times Company, 2013. Note, though, that by the end of 2016 digital subscriptions had crept up to 15 percent of total revenue—though this shift reflects decline in print as much as digital growth. For more on this topic, see the discussion in New York Times Company, 2016.
9. Ellis, 2014.
10. New York Times, 2014; see also Usher, 2014c.
11. Lee and Molla, 2018.
12. Chittum, 2014.
13. Gannett Co., 2018.
14. The “burn the boats” quote is from entrepreneur and investor Marc Andreessen, referenced in Schonfeld, 2010.
15. Lee and Molla, 2018.
16. See, for example, McClatchy Company, 2013.
17. Holcomb and Mitchell, 2014.
18. A. Newman and Leland, 2017.
19. McChesney and Nichols, 2011.
20. Usher and Layser, 2010.
21. Mutter, 2014.
22. Pew Internet and American Life Project, 2017a.
23. Barthel and Mitchell, 2017.
24. Mitchell, Rosenstiel, Santhanam, and Christian, 2012.
25. Knight Foundation, 2016; Nielsen, 2014.
26. eMarketer, 2014.
27. E.g., Pontin, 2012.
28. Mitchell et al., 2012.
29. Boczkowski, 2010.
30. Pontin, 2012.
31. Kalogeropoulos and Newman, 2017.
32. As of this writing, the pivot to video has now received an outpouring of commentary; noteworthy essays on the downturn include Moore, 2017, Josh Marshall, 2017, and Thompson, 2017.
33. Cohen, 2017.
34. Shoenfeld, 2017.
35. Benes, 2017.
36. Josh Marshall, 2017; see also Thompson, 2017.
37. E.g., Schurman and Brutlag, 2009.
38. Castillo, 2014.
39. Moos, 2009.
40. Author conversations with multiple Post staffers.
41. Konigsburg, 2014.
42. Hamann, 2014.
43. Bart et al., 2005; Wells, Valacich, and Hess, 2011.
44. Das et al., 2007; Liu, Dalan, and Pedersen, 2010.
45. Starkman, 2010.
46. Manjoo, 2013.
47. Wemple, 2014b, 2014a.
48. Upworthy, 2013.
49. Somaiya, 2014.
50. Mitchell, Jurgowitz, and Olmstead, 2014.
51. Ellis, 2012.
52. Bell, 2018.
53. Moses, 2018.
54. Kohavi et al., 2013; see discussion in chapter 2.
55. Bakshy, Eckles, and Bernstein, 2014
56. Author conversation with senior executives at the Washington Post.
57. Kovach and Rosenstiel, 2007.
CHAPTER 8. THE “NATURE” OF THE INTERNET
1. Barlow, 1996. Capitalization in original text.
2. Turner, 2006; see also Helmreich, 1998.
3. Doherty, 2004.
4. On the manufactured underdog origins of many tech firms, see Heath and Heath, 2011.
5. Wheeler, 2013.
6. Turner, 2006; Schumpeter, 1942.
7. Dimmick, 2002.
8. Napoli, 2011; see also Stober, 2004.
9. This account of Darwin is drawn especially from Ernst Mayr’s (1982) digest of Darwin in his classic book The Growth of Biological Thought. See also Bowler, 1989; Gould, 2002.
10. Shirky, 2010.
11. Hobbes, 1996.
12. O’Hara et al., 2013; for a more general discussion of the micro-macro problem, see Watts, 2011, pp. 61–64.
13. Abbatte, 1998.
14. Earl and Kimport, 2011.
15. Schlozman, Verba, and Brady, 2010; see also Schradie, 2012.
16. Thrall, Stecula, and Sweet, 2014.
17. Karpf, 2016.
18. Benkler, Shaw, and Hill, 2015.
19. MacArthur and Wilson, 1976.
20. One this point see Vaidhyanathan, 2012; Mueller, 2010.
21. Picard, 2014.
22. DeNardis, 2014.
23. The phrase “content is king” seems to have been coined by Bill Gates in a 1996 essay published on the Microsoft website that has long since been taken offline.
24. Bagdikian, 2004.
25. E. M. Noam, 2015; see also Odlyzko, 2001.
26. Wu, 2003.
27. FTC, n.d.
28. Amazon is something of a special case, with its core business long optimized for explosive growth rather than for profit. Their AWS cloud computing business, though, is exceptionally profitable and responsible for a large chunk of the firms’ market valuation.
29. Mullins, Winkler, and Kendall, 2015.
30. Federal Trade Commission, 2012, p. 112.
31. Shapiro and Varian, 1998; see chapter 2.
32. Federal Trade Commission, 2012, p. 112.
33. Scott, 2017.
34. Wu, 2011; though for a critique, see Starr, 2011.
35. Yarow, 2013.
36. For a good discussion of this point, see Orbach, 2013.
37. Easterbrook, 2008.
38. Zuboff, 2015.
39. Bradshaw and Howard, 2017.
40. King, Pan, and Roberts, 2013, 2017.
41. Silverman, 2017.
42. Howard et al. 2017.
43. Calabresi, 2017.
44. Petrova, 2011.
45. Quoted in Starr, 2004, p. 257.
46. Chyi and Tenenboim, 2017.
APPENDIX ON DATA, METHODOLOGY, AND MODELS
1. Quoted in Dyson, 2004.
2. Noam, 2004 and 2009.
3. Noam, 2004.
4. Hindman, 2009; Hamilton, 2004; Napoli, 2012.
5. Shiman, 2007; Crawford, 2007.
6. Smith, 2010.