News, Copyright, and Online Aggregators

Preliminary
Lesley Chiou and Catherine E. Tucker

October 1, 2010

Abstract This paper examines how the practices of online aggregators of news content aect consumers search of news online. The aggregation of content by third-party websites has become a controversial digital copyright issue. On the one hand, aggregators argue that they provide a useful service to consumers and that the amount and substantiality of the news article used in their links is small enough to be considered fair use. On the other hand, content providers argue that such practices represent copyright infringement and that third-party aggregators prots from advertising associated with content reduces the potential market for or value of copyrighted material. We empirically examine how the severing of a relationship between a major content provider and a major news aggregator aected consumers search for online news. Specically, we investigate how the removal of all hosted articles by The Associated Press from Google News at the end of 2009 (due to a dispute in licensing negotiations) aected which news sites consumers visited. Our empirical analysis suggests that the removal of The Associated Presss content was correlated with a decline in subsequent visits to traditional news sites immediately after visiting Google News compared to other news aggregators that continued to host Associated Press content.

Economics Department, Occidental College, CA MIT Sloan School of Management, MIT, Cambridge, MA. We thank Christopher Hafer of Experian Hitwise.

Introduction

The Internet has reshaped the news and media industry by providing a wealth of information and easy access to news, stories, and events that occur locally and globally. Over 74 million users visit newspapers sites each month, accounting for more than a third of all web users (Advertising Age, 2009). With the proliferation of search engines and news sites, consumers face an array of options and sources for information. Search engines have responded to consumers demand for information by creating online news aggregators, such as Google News and Yahoo! News. These news aggregators feature a collection of stories and headlines from various online news sources, which are collated into a single site. As such, aggregators oer a convenient place for users to consolidate their news reading. We study empirically whether using a news aggregator shifts consumers consumption of news online and whether news aggregators are substitutes or complements for the primary sources they feature. Surprisingly, there has been no empirical work that quanties the eect of aggregators on primary news sources, even though the US Copyright Oce requires proof of fair use and asks users to determine whether their use of the copyrighted material reduces its potential market or value. On the one hand, news aggregators may steal trac from media sites if users rely solely on the aggregators abbreviated descriptions of the article and do not visit the source site. In fact, this accusation has been levied by several major media organizations, including The Associated Press and News Corporation. Rupert Murdoch, chairman of News Corp., has accused news aggregators of stealing content and violating copyrights (Sandoval, 2009). Mark Cuban, chairman of HDNet, even referred to such practices as vampiric, saying that Newspapers are getting their blood sucked by Google and content aggregators (Kaplan, 2010). Moreover, some anecdotal evidence exists that a signicant fraction of readers scan only the headlines from Google News and do not visit the source site (Sullivan, 2010). On the other hand, news aggregators may expose readers

to news sources and sites that they might otherwise not visit, and therefore generate trac for the primary news source. As argued by Arrington (2010), When an aggregator puts up a link to your site, they are doing you a favor by sending you trac. Therefore, the overall eect of news aggregators on consumers news-seeking habits is an empirical question. Our paper focuses on how the practices of news aggregators aect consumers search for news online. We use a rich dataset of consumers online search behavior and site visits to examine the eect of a policy change in content displayed by a major online news aggregator in displaying content from a primary news. The decision to host news sources may be correlated with other factors that inuence a consumers consumption of news, so we use a discontinuous event that altered the set of news sources provided by a major news aggregator. In January 2010, after a breakdown in licensing negotiations, Google removed all news articles by The Associated Press from its news aggregator (Haddad, 2010). We compare consumers site visits before and after this policy relative to trac from Yahoo! News, which continued to provide Associated Press content in this period. Yahoo! News and Google News play a large role in the online news market and are among the top 5 news sites visited by readers. We nd that after Associated Press content was removed from Google News, fewer consumers subsequently visited traditional news sites (the sources for much of Associated Press content) relative to consumers using Yahoo! News. We checked the robustness of the result in a variety of ways. Our nding suggests that the aggregation of news content actually complements the original content. In other words, users are more likely to be provoked to seek the original source and read further when they come across a story summarized by an aggregator, rather than being merely content with the summary. Our paper builds on a growing literature that documents how the Internet has aected the consumption of traditional news media. Gentzkow (2007) investigated the relationship between oine and online newspapers. He found some evidence of complementarity but ulti3

mately concluded that this was an artifact of customer heterogeneity and that the provision of online news was costly to newspaper print edition revenues. This nding was supported in separate research by George (2008) and Filistrucchi (2005). Kaiser and Kongsted (2005) nd evidence of complementarity for online magazines, which suggests that substitution is particularly important for news content rather than more magazine-type content. To our knowledge, however, we are the rst to study how the practice of online aggregation aects online news consumption. Our distinction between traditional media, which is primarily local, and news aggregators that have national reach builds on earlier research that has documented the importance of the interaction between national and local news distribution practices for understanding consumption of news (George and Waldfogel, 2006; OberholzerGee and Waldfogel, 2009). Our work also relates to research that has evaluated the conict between digitization and copyright. These studies have focused predominantly on the issues relating to the piracy of lm and musical content (Rob and Waldfogel, 2006; Oberholzer-Gee and Strumpf, 2007; Danaher et al., 2010) by unauthorized distribution channels. Our paper has several implications for media markets and public policy. First, a erce debate exists over intellectual property and copyrights for content posted online. Search engines and aggregators accumulate information from primary sources, and controversy exists over whether news aggregators violate existing copyrights and whether content can be provided freely by a third party. Secondly, policymakers have long stressed the importance of the diversity of news consumption and how consumption of local news in particular encourages civic engagement. Our results suggest that news aggregators actually provoke readers to seek further news. Given that local news sites comprise a substantial fraction of online news, aggregators may promote more diversity in consumption patterns.

2
2.1

Data and Institutional Setting

Contractual Dispute between Google and Associated Press

The Associated Press, founded in 1846, is one of the most powerful news agencies in the world. Since the demise of United Press International, it is the only national news service in the US, and its major competitors are now the United Kingdom-based Reuters and the France-based Agence-France Presse. It is a cooperative that is owned by various newspapers and radio and television stations in the United States. These stakeholders both contribute stories to the Associated Press and use material which are written by its sta journalists. During the past decade, The Associated Press has been at the forefront of eorts by copyright holders to circumscribe fair use for digital content and protect copyholders rights. For example, in June 2008, Associated Press has invoked the Digital Millennium Copyright Act and insisted that various bloggers remove Associated Press content (Ardia, 2008). Google News is ranked as the fth most visited news website by Hitwise. Receiving 2.90% of all news site visits, it is the second most popular news aggregator service after Yahoo! News, which received 7.09% of all news site visits. Founded on April 2002, Google News electronically aggregates dierent news sources based upon a proprietary algorithm. As of December 2009, Google News claimed that it received news content from 25,000 publishers across the world and that it sent 1 billion clicks to these publishers every month (Cohen, 2009). Google News has been supported by advertising revenues in the US since February 2009. Figure 1 provides a screenshot of Google News. Google News has two noticeable features that distinguishes it from traditional news sites. First, a variety of sources are listed for each story. Second, the order of news is electronically determined based on users preferences, the recency of the story, and the interest it has received from other users. Since both The Associated Press and Google News are key players in the distribution of news, it is not surprising that they have forged a partnership. Table 1 summarizes the 5

Figure 1: Screen shot of Google News screen

Note: On June, 30 2010, the formatting of Google News changed somewhat and reduced the ability of users to customize the placement of the columns containing news. Therefore the screenshot above, which was produced after this formatting change, may be slightly dierent from what users viewed during the period that we study.

major events of their relationship. We study a discontinuity in this relationship, which was engendered by negotiations surrounding the contract renewal at the end of January 2010. As part of their existing contract, Google and The Associated Press agreed that Associated Press content could be hosted by Google for a period of 30 days. Therefore, if the contract ended in January 2010 and was not renewed, Google would have to stop posting new Associated Press content 30 days prior to the end of the contract. Presumably, to make this clean break a credible outside option, Google did indeed stop posting content for seven weeks during these contract negotiations. We should emphasize that this is necessarily based on the observations of industry outsiders, since both Google and the Associated Press signed binding non-disclosure agreements, which prevented them from ever commenting on the course or outcome of negotiations (Sullivan, 2010). This removal of Associated Press content represents a useful natural experiment for empirical researchers. Since the removal of content was provoked by the intricacies of contract negotiation, its timing can be thought of reasonably exogenous, as it was determined by the expiration of the contract rather than any considerations of the popularity (or lack thereof) for Associated Press content at that time. As detailed in Table 1, the dispute with the Associated Press led Google to remove content by the Associated Press from December 23, 2009 to February 9, 2010. Fortunately for our purposes, Yahoo! News continued to host Associated Press content without interruption during this time, which enables us to use its web users behavior as a control in our regressions. We compare which websites consumers navigated to after visiting a news aggregator before and after the removal of content on Google for both visitors to Google News and Yahoo! News. Critics and supporters alike of news aggregators have proposed numerous arguments for whether the removal of Associated Press content may either benet or hurt news websites. On the one hand, if consumers are no longer able to obtain Associated Press news content, they may be more likely to seek the news directly from the Associated Press member orga7

Table 1: Timeline of negotations between Google and Associated Press Date Event March, 2005 Google is sued by Agence France Presse for copyright infringement after AFP content appeared on Google News. August, 2006 Google and Associated Press rst sign contract to enable Associated Press content to appear on Google News for 30 day window. Associated Press content no longer appears on Google. Industry press speculates that this is in preparation for the expiration of contract between Associated Press and Google in a months time. Associated Press and Google contract set to expire Associated Press Content returns to Google News.

December 24, 2009

End January 2009 February 2010

nizations and newspapers. On the other hand, consumers may simply be less likely to seek further information about news. In essence, this distinction can be boiled down to whether consumers view news aggregators as a complement or substitute to original news sources. Do they use news aggregators to identify news stories that they then pursue in greater depth, or do they simply stop after reading the rst news item? For instance, the Associated Press ran a news story about economic depression in Michigan in August 2010. The screenshot of how the story appeared on Google News is depicted in Figure 2. The links relating to the Associated Press story that appear at the bottom of a typical story are also depicted in Figure 2. After reading the Associated Press summary of the story, readers are free to explore the issue further in local newspapers such as the Detroit News and Lansing State Journal. The question we ask is whether the presence of the Associated Press content on Google News makes it more or less likely that a news consumer would then trouble to visit Detroit News or the Lansing State Journal, both of which are members of the Associated Press Network. Our analysis is focused on the period immediately prior to and during the removal of 8

Figure 2: Example screenshot of Associated Press article hosted on Google News

Note: Google News, August 1st 2010. Text of article has been slightly edited to t on page.

Associated Press articles from Google News for two reasons. First, it is not immediately clear at which point in February that Google News and Associated Press resumed their relationship and reached a new agreement. Second, it is not apparent whether the reinstatement during this time consisted of the older, missing content or new content or whether Google changed the presentation of AP articles afterwards. For example, it would be problematic if Google decided to highlight Associated Press content after the contract negotiations were concluded, perhaps as a sweetener to the deal. For these reasons, we focus on visits to news sites during the months of December 2009 and January 2010. 2.2 Data Description

Our data derive from Experian Hitwise. Hitwise develops proprietary software that Internet Service Providers (ISPs) use to analyze website logs created on their network. Once the ISP aggregates the anonymous data, the data are provided to Hitwise. According to their website, Hitwise collects these usage data from a geographically diverse range of ISP networks and opt-in panels, representing all types of Internet usage, including home, work, education and public access. Currently, Hitwise has usage data from a sample of 25 million people worldwide. We collected information on the sites that users visit immediately after navigating to Google News or Yahoo! News. We use weekly data from the week ending December 5, 2009 to the week ending February 27, 2010 for the top 1500 sites navigated after Google News or Yahoo! News. Hitwise reports the fraction of total trac that arrives at each downstream site immediately after a visit to Google News and Yahoo! News. We constructed a 2month panel where the unit of observation is the percentage of weekly clicks a downstream website received from either Google News or Yahoo! News. Twenty-six percent of websites received incoming trac from both Google and Yahoo! News. The remainder of websites were only visited after navigating to one particular aggregator. This may reect internal

10

complementarities for these companies. For instance, someone using Google News is unlikely to navigate to Yahoo! Mail, and similarly someone using Yahoo! News is unlikely to navigate to Gmail. We categorized the websites into two main classes: non-news (e.g., Yahoo! Mail, myspace.com) and traditional news (e.g., newyorktimes.com, bostonherald.com). We applied Hitwises own categorization of news websites to identify traditional news media, but we excluded weather sites and news aggregators from the 5 major search engines (such as Yahoo! News, Google News, Hungton Post) from the category.1 We identied a site as an aggregator based upon whether or not they produced their own original content. We also constructed a separate category for international news (e.g., bbc.com/news, hindustantimes.com), which we use in our robustness checks. We would expect the removal of Associated Press content to aect traditional news media sites, but the removal should not aect visits to international sites that tend to either generate their own content or rely on non-American news agencies for their content. Table 2 reports the summary statistics for our data. It is striking that 20 percent of the time, consumers navigate to a traditional news media website from the news aggregator. Traditional news sites captured most trac. International news received less trac (5.5 percent of sites visited) than traditional news sites.

Hitwise reports the top 10,000 ranked news and media sites in November 2009.

11

Table 2: Summary statistics for downstream websites from Google News and Yahoo! News Mean Std Dev Min Max Observations % clicks 0.016 0.19 0 18.3 100503 Google News 0.50 0.50 0 1 100503 PeriodDispute 0.67 0.47 0 1 100503 Traditional News Site 0.20 0.40 0 1 100503 News Aggregator Site 0.0011 0.033 0 1 100503 International News Site 0.055 0.23 0 1 100503 Observations 100503
Notes: This table reports statistics for websites visited immediately after Google News and Yahoo! News during December 2009 and January 2010. The period during which the dispute occurred between Associated Press and Google News was after December 23, 2009. Traditional news sites refer to news and media sites as dened by Hitwise, excluding weather sites, international news sites, and news aggregators from the top 5 search engines.

12

Table 3 displays the top 50 (traditional) news websites in our dataset and the average percentage of downstream clicks they receive. Table 4 displays the top 50 non-news websites in our dataset and the average percentage of downstream clicks they receive. As shown in Table 4, the top non-news websites reect the top website brands on the Internet. This is suggestive evidence that users of news aggregator sites have both mainstream Internet tastes and regard the sites as part of their normal Internet consumption.

13

Table 3: Top 50 news websites visited after Google News and Yahoo! News
abcnews.com associatedcontent.com bleacherreport.com bloomberg.com boston.com bostonherald.com businessweek.com cbsnews.com chron.com cnn.com csmonitor.com dallasnews.com drudgereport.com edition.cnn.com examiner.com foxnews.com foxnews.com/entertainment foxnews.com/politics freep.com gather.com latimes.com mcclatchydc.com mercurynews.com miamiherald.com msnbc.com news.com nj.com npr.org nydailynews.com nypost.com nytimes.com people.com philly.com politico.com radaronline.com reuters.com seattlep-i.nwsource.com seattletimes.nwsource.com sfgate.com sportsillustrated.cnn.com startribune.com thedailybeast.com theweek.com time.com upi.com usatoday.com usmagazine.com usnews.com voanews.com washingtonpost.com wsj.com 2.11 0.11 0.17 0.51 0.24 0.19 0.15 0.19 0.13 1.85 0.15 0.11 0.64 0.20 0.65 1.13 0.082 0.20 0.13 0.34 0.48 0.095 0.44 0.15 0.83 0.12 0.11 0.16 1.59 0.26 2.88 0.39 0.15 0.53 0.060 0.69 0.11 0.11 0.17 0.10 0.084 0.17 0.14 1.16 0.093 0.72 0.23 0.082 0.13 1.74 0.86

Table 4: Top 50 Non-news websites visited after Google News and Yahoo! Avg Visit Pct News
address.yahoo.com amazon.com aol.com aralifestyle.com ask.com bankofamerica.com bing.com blogsearch.google.com buzz.yahoo.com chase.com cosmos.bcst.yahoo.com ebay.com education.yahoo.net espn.com facebook.com fastip.googlelabs.com nance.google.com nance.yahoo.com games.yahoo.com gmail.com google.com howlifeworks.com hungtonpost.com images.google.com latimesblogs.latimes.com livescience.com mail.live.com mail.yahoo.com maps.google.com members.yahoo.com movies.yahoo.com msn.com my.yahoo.com myspace.com news.google.com omg.yahoo.com rivals.com search.yahoo.com shine.yahoo.com space.com sports.yahoo.com sports.yahoo.com/n tmz.aol.com tv.yahoo.com video.google.com weather.com weather.yahoo.com wikipedia.org yahoo.com 14 youtube.com Avg Visit Pct 0.12 0.59 0.46 0.14 0.19 0.18 0.62 0.77 0.21 0.14 0.95 1.00 0.34 0.56 6.23 3.60 0.36 0.60 0.099 1.55 11.6 1.04 0.96 0.50 0.16 0.38 1.28 9.94 0.23 0.29 0.13 1.03 0.67 1.54 0.24 0.32 0.10 2.20 0.13 0.15 0.26 0.13 0.20 0.12 0.27 0.67 0.39 0.50 7.20 2.47

Table 5: Demographic description of users Measure Yahoo! News Google News New York Times Male 59.95 63.8 61.21 Age 18-24 12.12 13.89 6.17 Age 25-34 18.05 14.72 13.93 Age 35-44 19.03 17.08 12.98 Age 45-54 21.41 22.24 19.45 Age 55+ 29.38 32.06 47.47 Income <30k 22.33 20.77 20.76 Income 30-60k 28.82 27.53 26.36 Income 60-100k 24.95 24.6 24.82 Income 100-150k 14.61 17.5 17.29 Income >150k 9.29 9.6 10.77 Source: Hitwise
Notes: This table reports the fraction of users of a particular website within each demographic category. Statistics are reported for users of Yahoo! News, Google News, and the New York Times website.

To verify that Yahoo! News could be considered an appropriate control group for Google News, we checked that the users shared similar observable demographics. Table 5 reports the fraction of users within each demographic category for a particular site. The users of Yahoo! News and Google News do indeed look reasonably similar; they are skewed towards being older, predominantly male, and wealthier than the general U.S. population. For comparison, we also report demographics for users of the New York Times website. The users of the New York Times site are similar, though signicantly older, than the average users of a news aggregator. Table 5 also provides suggestive evidence of why the debate over ad revenues from news content is so contentious. Users such as these are a remarkably attractive demographic group from an advertisers perspective.

15

Analysis

Figure 3 summarizes our main analysis. Figure 3 illustrates the mean percentage of downstream trac for users that visited Google News and Yahoo! News during our period. As seen in the graph, little change occurs in downstream site navigation for Yahoo!. However, news sites experience a decline in visits from Google News after the removal of Associated Press relative to the change in trac from Yahoo! News. Figure 4 extends this analysis to show how visit behavior varies for international news sites as well. Once again, little change exists in user behavior for these additional types of websites on either Yahoo! News or Google News, suggesting that these sites were not aected by the removal of Associated Press content. As expected, these international websites are unlikely to be aected by the removal of AP content due to the nature of their content. As seen in Figure 5, no such change in clicks occurred in the prior year during the same calendar months of December 2008 and January 2009. Figure 3: Downstream sites visited after Google News and Yahoo! News

Notes: This gure shows the average percentage of clicks for news and non-news sites navigated to after visiting from Google News and Yahoo! News before and after the removal of The Associated Press from Google News.

16

Figure 4: Downstream sites visited after Google News and Yahoo! News

Notes: This gure shows the average percentage of clicks for a variety of website types navigated to after visiting from Google News and Yahoo! News before and after the removal of The Associated Press from Google News.

To formalize the insights provided by Figures 3 and 4, we run a dierence-in-dierences regression for the policy change and estimate the following regression for the percentage of clicks to website i after visiting news aggregator j in month t:

%clicksijt = 0 + 1 N ewsi Googlej P eriodDisputet + 2 N ewsi P eriodDisputet + 3 N ewsi Googlej + 4 Googlej + i + weekt +
ijt

where N ews is an indicator variable equal to 1 if the website is a traditional news source, Google is an indicator variable equal to 1 if the trac originated after viewing Google News, and P eriodDispute is an indicator variable equal to 1 for the weeks after the removal of Associated Press from Google News. The controls are downstream-website xed eects. The vector weekt contains weekly xed eects to capture national variation in the volume

17

Figure 5: Downstream sites visited after Google News and Yahoo! News in prior year (December 2008 and January 2009)

Notes: This gure shows the average percentage of clicks for news and non-news sites navigated to after visiting from Google News and Yahoo! News in December 2008 and January 2009 for the year prior to the removal of The Associated Press from Google News.

and interest generated by news stories in that week. The coecient on the interaction term N ews Google P eriodDispute captures the eect of the Associated Press removal on visits to traditional news sites compared to non-news sites from Google News with the corresponding change in traditional news and non-traditional news sites on Yahoo! as a control. We estimate this specication using ordinary least squares and cluster our standard errors at the website level to avoid the downward bias reported by Bertrand et al. (2004). Table 6 reports the results for various regression specications, incrementally building up to our full specication described by equation (1). Very little variation exists in the size or precision of our coecient of interest in each of the columns. The negative coecient on N ewsGoogleP eriodDispute implies that during the dispute with Associated Press, Google News users were less likely to visit traditional news websites after visiting Google News. This suggests that the presence of Associated Press articles in Google News prompted users to seek further information at traditional news sites and thereby encouraged more diversity in 18

Table 6: Downstream trac Google and Yahoo! News before and after the policy change (1) (2) (3) % clicks % clicks % clicks PeriodDispute X Google X News -0.00583 -0.00583 -0.00583 (0.00271) (0.00284) (0.00284) PeriodDispute X Google 0.00152 0.00152 0.00152 (0.00219) (0.00229) (0.00229) PeriodDispute -0.000514 -0.000514 -0.000560 (0.000655) (0.000686) (0.00105) Google -0.00372 -0.0115 -0.0115 (0.00367) (0.00617) (0.00617) News -0.000393 (0.00356) PeriodDispute X News 0.00143 0.00143 0.00143 (0.000963) (0.00101) (0.00101) News X Google 0.0184 0.0324 0.0324 (0.00600) (0.00753) (0.00753) Website Fixed Eects Yes Yes Yes Week Fixed Eects No No Yes Observations 100503 100503 100503 R-Squared 0.000543 0.581 0.581 Robust standard errors clustered at website level. *p < 0.1, **p < 0.05, ***p < 0.01. The dependent variable is the fraction of trac to websites after visiting Google News or Yahoo! News. The policy change is the removal of hosted articles by The Associated Press from Google News. news consumption. News sites on Google experience a 6 percentage point decrease in clicks after the removal of Associated Press articles. Compared to the mean percentage share of 2.9 percent before the policy change, this drop represents an approximately 20 percent decrease in trac to news sites after the removal of Associated Press articles from Google. If the claim in Cohen (2009) is true that Google sends a billion clicks each month to its partner news providers, then this percentage translates into a very large change in the number of clicks that traditional news websites receive. While we do not know precisely the international breakdown, our data from Hitwise suggest that 40 percent of all clicks before the policy change went to 19

traditional news media websites hosted in the US. Therefore, this 20 percent decrease could imply a 80 million decrease in visits each month from Google News users each month to traditional news media websites hosted in the US. Our results suggest that news aggregators complement the news sources that they feature by directing trac to these news sites. The provision of content on news aggregators encourages readers to seek further information from other news sources.

20

Robustness Checks

We conducted various robustness checks as reported in Table 7. Columns (1) and (2) check the robustness of our results to alternative specications. We apply a Tobit regression to account for sites that receive zero clicks in a given week and also a semi-log regression.2 Both regressions have similar signs for the coecients of interest; news sites receive less trac from Google after the policy change. Columns (3)-(5) check robustness of the results to alternative denitions of the control group. As described previously, users navigated to a variety of non-traditional news sites after visiting a news aggregator. These sites included both non-traditional and nonAssociated Press sources of news. In columns (3) and (4), our robustness checks omit the top news aggregators and international websites as part of the control group. These alternative denitions of the control group could be warranted if the removal of Associated Press content also aected navigation to these sites directly (e.g., if Associated Press content had previously encouraged people to visit international websites, or if the removal of Associated Press content on Google altered peoples perceptions of news aggregators.) In column (5), we check robustness to removing both news aggregators and international sites from our data. Generally, the results are robust in sign.

For the semi-log regression, we use log(%clicks+0.01) as the dependent variable.

21

Table 7: Robustness checks: Downstream trac to local news sites from Google News and Yahoo! News before and after the policy change

PeriodDispute X Google X All News

PeriodDispute X Google

PeriodDispute

Google

(2) Semi-log -0.0216 (0.0127) -0.00783 (0.00572) -0.00685 (0.00527) -0.0207 (0.0123)

(3) No Aggregators -0.00583 (0.00284) 0.00152 (0.00230) -0.000464 (0.00108) -0.0114 (0.00619)

(4) No International -0.00620 (0.00303) 0.00178 (0.00249) -0.000762 (0.00111) -0.0153 (0.00674)

(5) News vs Non-News -0.00632 (0.00306) 0.00179 (0.00251) -0.000763 (0.00112) -0.0155 (0.00680)

All News

22
0.0155 (0.00876) 0.0785 (0.0232) Yes Yes 100503 0.684 0.00142 (0.00101) 0.0323 (0.00754) Yes Yes 100395 0.580

PeriodDispute X All News

All News X Google

Website Fixed Eects Week Fixed Eects Observations R-Squared

(1) Tobit -0.0229 (0.00924) 0.00394 (0.00557) -0.00556 (0.00446) 0.0248 (0.00874) 0.112 (0.0208) 0.0152 (0.00564) -0.0126 (0.0142) No Yes 100503 0.00148 (0.00104) 0.0364 (0.00804) Yes Yes 94959 0.581

0.00154 (0.00105) 0.0362 (0.00812) Yes Yes 94203 0.581

Robust standard errors clustered at website level. *p < 0.1, **p < 0.05, ***p < 0.01. The dependent variable is the fraction of trac to websites after visiting Google News or Yahoo! News. The policy change is the removal of articles by The Associated Press from Google News.

We also veried that no global changes occurred in the usage of Yahoo! News and Google Newss news aggregation service during the period we study. Of particular concern is that the omission of Associated Press content led people to perhaps leave Google News and explore alternative news aggregators. When we checked the Hitwise data, we saw no evidence of such changes in behavior. Indeed, throughout the period we study, Google News remained solidly at its # 5 ranking for unique visits among news websites while Yahoo! remained at #1. Moreover, no change occurred in alternative metrics such as average visit time or the number of pages navigated within a website.

Conclusion

This paper studies empirically whether content providers benet from the collaging of their content by electronic news aggregators. In particular, we examine whether news aggregators are substitutes or complements with the primary sources they feature. This question is much debated in the media industry. On the one hand, news aggregators may steal trac from media sites if users rely solely on the aggregators abbreviated descriptions of the article and do not visit the source site. On the other hand, news aggregators may expose readers to news sources and sites that they might otherwise not visit and therefore generate new trac for the primary news source. We use data from Internet users collected by Hitwise, which documents the sites that users visit after navigating to a news aggregator. We exploit an unusual natural experiment in the form of a breakdown in contract negotiations between the Associated Press and Google, which led Google to stop host ing Associated Press content for 7 weeks. We nd evidence that during this period where no Associated Press content was available on Google, Google News users were less likely to visit traditional news websites after visiting Google News, compared to before the content was removed and relative to Yahoo! News users who experienced no such removal of Associated Press content.

23

There are obvious limitations to how well our ndings generalize. We believe that the removal of Associated Press content from Google News was indeed an exogenous and unexpected event related to contract negotiations rather than underlying changes in how consumers navigated newsites, but there is always the possibility that Google felt enabled by anticipated changes in consumers navigation of the web to take such a strident position. Perhaps more crucially, there is a concern that Associated Press content is itself slightly atypical, in that it represents an attempt (prior to the rise of the Internet) at the aggregation of news content. Since The Associated Press is an umbrella organization, it is somewhat unusual among content providers. These limitations notwithstanding, the ndings of this paper do illuminate a growing debate around the conict between copyright and electronic content aggregators. Our results highlight that while content providers may feel that online aggregators prot unfairly from their content, these aggregators do provide a useful function by increasing visits to the content providers proprietary website. In that sense, our results bolster Googles argument that their display of the headlines from Associated Press content does constitute fair use by not harming the value of or market for the copyrighted material.

24

References
Ardia, D. (2008, June 16). Associated Press Sends DMCA Takedown to Drudge Retort, Backpedals, and Now Seeks to Dene Fair Use for Bloggers. Citizen Media Law Project . Arrington, M. (2010, Feb 2). Everybody forgets the readers when they bash news aggregators. Techcrunch . Bertrand, M., E. Duo, and S. Mullainathan (2004, February). How much should we trust dierences-in-dierences estimates? The Quarterly Journal of Economics 119 (1), 249 275. Cohen, J. (2009, December 2). Same protocol, more options for news publishers. Posting on Google Newss Blog . Danaher, B., S. Dhanasobhon, M. D. Smith, and R. Telang (2010). Converting Pirates Without Cannibalizing Purchasers: The Impact of Digital Distribution on Physical Sales and Internet Piracy. forthcoming, Marketing Science Science . Filistrucchi, L. (2005). The Impact of Internet on the Market for Daily Newspapers in Italy. Technical report. Gentzkow, M. (2007, June). Valuing New Goods in a Model with Complementarity: Online Newspapers. American Economic Review 97 (3), 713744. George, L. and J. Waldfogel (2006, March). The New York Times and the Market for Local Newspapers. The American Economic Review 9 (1), 435447. George, L. M. (2008). The Internet and the Market for Daily Newspapers. The B.E. Journal of Economic Analysis & Policy 8 (1).

25

Haddad, M. (2010). Associated Press Google News Partnership Ends. Business 2.0 Press, January 12. Kaiser, U. and H. C. Kongsted (2005). Do Magazines Companion Websites Cannibalize the Demand for the Print Version? Discussion Paper No. 05-049 . Kaplan, D. (2010, Feb 2). Mark Cuban: Google, Content Aggregators Are Vampires; Newspapers Are Zombies. PaidContent.org. Oberholzer-Gee, F. and K. S. Strumpf (2007). The Eect of File Sharing on Record Sales: An Empirical Analysis. Journal of Political Economy, Vol. 115, pp. 1-42, February 2007 . Oberholzer-Gee, F. and J. Waldfogel (2009, December). Media Markets and Localism: Does Local News en Espanol Boost Hispanic Voter Turnout? view 99 (5), 21202128. Rob, R. and J. Waldfogel (2006, April). Piracy on the high cs: Music downloading, sales displacement, and social welfare in a sample of college students. Journal of Law & Economics 49 (1), 2962. Sandoval, G. (2009). Google May Lose WSJ, Other News Corp. Sites. CNET News, November 9. Sullivan, D. (2010, January 8). Wheres AP In Google News? Apparently In Limbo, As Contract Running Out. Search Engine Land . The American Economic ReZEW - Centre for European Economic Research

26