ArticlePDF Available

Media Ownership and Bias: Evidence by the Washington Post’s Reporting after Amazon’s Acquisition in 2013

Authors:

Abstract and Figures

This study examines the impact of ownership change on media coverage by investigating the impact of Jeff Bezos's (Amazon's owner) purchase of the Washington Post (WP) in October 2013. We collect all articles that have Amazon mentions published by WP and the New York Times (NYT) for 12 months before and 12 months after the acquisition. Then, we use the difference-in-differences method to compare changes in sentiment, length, and time of publication of the articles reported by WP compared to NYT. From the comparison, we show that Amazon's acquisition has posed a certain extent of influence on the publication time of the articles. Our findings highlight that ownership change limits the media's ability to fulfill its role as a watchdog without interference.
Content may be subject to copyright.
Modern Economics & Management Forum 86 | Tongyi Jin, et al.
Media Ownership and Bias: Evidence by the Washington Post’s
Reporting after Amazon’s Acquisition in 2013
Tongyi Jin1*, Altynay Abilgaziyeva2, Tsz Ting Lam3
1 Department of Applied Social Sciences, The Hong Kong Polytechnic University
2 Ernst & Young
3 RSM Hong Kong
Email: serena-tongyi.jin@polyu.edu.hk
Abstract: This study examines the impact of ownership change on media coverage by investigating the impact of Jeff
Bezos's (Amazon's owner) purchase of the Washington Post (WP) in October 2013. We collect all articles that have Ama-
zon mentions published by WP and the New York Times (NYT) for 12 months before and 12 months after the acquisition.
Then, we use the difference-in-differences method to compare changes in sentiment, length, and time of publication of the
articles reported by WP compared to NYT. From the comparison, we show that Amazon's acquisition has posed a certain
extent of inuence on the publication time of the articles. Our ndings highlight that ownership change limits the media's
ability to fulll its role as a watchdog without interference.
Keywords: media ownership, bias, acquisitions
1. Introduction
"Most of what you have read or watched in the media is true and 100% accurate. But HOW you are told the stories,
and When, there lies the manipulation!" (Kanjo, 2016).[1] We often receive different information from the media, but
how these media deliver the story could preemptively affect how we comprehend the event. Media might attempt
to predetermine our perception of a subject with labels, omissions and frame of the news by packaging them with a
tone, exposure and other effects. "The press itself is a facility and falsely labeled "the liberal media" by the hundred of
conservative columnists, commentators, and talk show hosts who crowd the communication universe while claiming to be
shut out of it." (Parenti, 1997).[2]
In our research, we focus on a particular event — Amazon's purchase of the Washington Post in October 2013
to identify how the change in ownership shapes the media coverage. Right after acquisition, new owner, Jeff Bezos,
encouraged WP's shift of focus from "For and about Washington" to becoming a "global" company and encouraged WP
to cover both him and Amazon aggressively (Meyer, 2014).[3] Bezos intended to expand the national and worldwide
audience, and he started a massive digital transformation at WP and aggressively hired new people to spread them across
new projects. Although it is stated that Bezos has focused on technology, not journalism ("Exploring the Amazon", 2015)[4],
from the following example, it is clear that on average the coverage of WP owned by Bezos has changed to be more
benecial towards Amazon.
The recent Amazon-related news that caught a lot of media attention was the shocking divorce of Jeff Bezos with
his former wife (Day & Metcalf, 2019).[5] The news articles about the divorce extracted from Factiva showed that there is
an evident variation of tone between different media. For instance, the number of articles issued by the New York Times
is nine times more than that by WP with the average length of articles being 243 words longer than WP. It may represent
WP's desire not to emphasize the news because a divorce is a negative event for Amazon's stock price, i.e. it might worry
investors about the control stability. Meanwhile, NYT framed shareholders' concerns in one of the headlines (Stewart,
2019).[6] After the divorce was nalized, WP published an article that emphasized Bezos retaining 75% of the couple's
stock and ALL voting power in order to stabilize the market ("Jeff Bezos, in divorce", 2019).[7] When the National Enquirer
published text messages between Bezos and a woman he was having an affair with, WP produced a rather lengthy article
(2782 words), accusing National Enquirer in blackmail and defending Bezos ("Inquiring minds", 2019).[8] From these,
we assume that WP has been acting as Bezos' mouthpiece after the acquisition. Hence, we nd it interesting to study how
biased WP became in reporting Amazon related news after the acquisition.
In this paper, we contributed to the existing literature by assessing the media's difference across three important
metrics — length, sentiment, and publication time of articles before and after the ownership change compared to its
competitors. We introduced a new tool, a natural language processing technique, to measure the sentiment score of the
Volume 1 Issue 3 | 2020 | 87 Modern Economics & Management Forum
articles. By analyzing the differences in sentiment, length, and publication time of the articles, we could identify the bias
in media coverage. Hence, it is evident that the new owner could achieve various goals by using his power to control news
published.
2. Literature Review
Media's essential role in a democratic society is to act as a watchdog of the public and hold the actions of the
government, political leaders, and business elites accountable. The free flow of ideas is critical to ensure the well-
functioning of society (Archer & Clinton, 2018).[9] However, nowadays, with six corporations controlling almost 90% of
the media in the USA (Lutz, 2012)[10], there is increasing pressure on the media’s true motivations. That is, owners of the
large conglomerates, taking more control over mass media in the USA, have incentives to report prot-maximizing news.
The importance of media as an information intermediary that enables citizens to make well-thought choices and critical
evaluations has shifted towards turning to a platform for owners to exercise their business interests and influence the
public.
Plenty of prior literature explores the consequences of ownership change to media tycoons, moguls, or public
organizations such as political parties, the state, and multinational corporations. For instance, a study conducted by
Chomsky (2006) illustrated how media ownership limits the media’s capacity to perform its so-called watchdog role in
society without any interference.[11] His study is based on the memoirs written by Turner Catledge in 1971, the top editor at
the NYT. In his memoirs, Turner Catledge revealed that Arthus Hays, NYT owner, has written more than 400 memos with
instructions, suggestions, complaints, and orders regarding the news published. Such ownership intervention was regarded
as usual practice, and Catledge would remove the owner’s name from the memo and suggest his ideas as his own, or as
ideas of an “interested reader”. Almost in all cases, the owner’s suggestions were met, and even in cases when the owner’s
ideas were not newsworthy, the stories would still appear in the newspaper. Careful observations of internal interactions
showed that the owner presented no self-consciousness about interference and the reporter never questioned the owner’s
right to interfere while deciding the news content.
We can see further evidence on how ownership frames the coverage from research done by Archer & Clinton (2018).[9]
On August 1, 2007, Rupert Murdoch purchased the Wall Street Journal (WSJ), being drawn to its status of the top-selling
US daily newspaper. With the incentive to further increase the profitability, Murdoch changed the WSJ’s coverage to
include more political stories to increase media protability. Archer and Clinton studied the impact of Murdoch’s purchase
on the WSJ’s coverage. They examined the percentage change in political stories coverage across different newspapers,
such as NYT, WP, and USA Today. As a result, they found that the post-acquisition change in the WSJ’s political coverage
is remarkable, especially in the front-page stories. Furthermore, Archer and Clinton identied that the WSJ’s coverage
became more conservative, which reects the political views of Murdoch, who is republican himself. In the past, Murdoch
used other media outlets in his control to restrict publishing positive articles on the UK entering the Eurozone (Harding,
2002)[12], and to inuence public opinion on the Iraq War (Szalai, 2007)[13].
In addition, anecdotal evidence concerning the signicance of the media owners controlling the accent or coverage of
events in the news is bountiful. One of the specic examples of media coverage being inuenced by the owner's interest
happened in 1980 when the Los Angeles Times did not report US$ 2 billion tax-paid water projects in California, benecial
to the Times-Mirror Company (the owner of LAT). In contrast, NYT and the Bakerseld Californian chose the event as
appropriate enough to cover (Bagdikian, 1992).[14] Another example includes the Telecommunications Act of 1996 when
the newspapers that beneted from the new act, in their coverage reported more positively on the proposed changes. In
contrast, newspapers that did not gain anything reported more negatively with negative consequences appearing three times
more (Gilens and Hertzman, 2000).[15]
Hanretty (2014) suggested that different types of media ownership have different consequences on media coverage
as particular owners such as political parties, media tycoons, moguls, etc. have distinct goals to achieve.[16] The ownership
by conglomerates implies active self-promotion in the media. For instance, when Warner Communications acquired Time
magazine in 1989, the magazine which was in recession before, was transformed into an exclusive platform to exercise
the conglomerate’s promotion and entertainment possibilities (Turow, 1994).[17] In our research, we assume that Jeff Bezos
acquired the Washington Post, which also was in recession before the acquisition, to turn it to the platform to exercise
promotion opportunities. Thus, we expect aggressive coverage of Amazon-related events by the Washington Post after the
acquisition.
The media outlets utilize different types of media bias to achieve their goals. For instance, when Butler & Gurun
(2012) studied the bias of the local newspapers in reporting local rms compared to non-local companies, they identied
Modern Economics & Management Forum 88 | Tongyi Jin, et al.
that reports on local companies contain fewer negative words compared to non-local companies.[18] The reason behind
such favoritism was due to the local companies being the source of advertising income for the local newspapers. Media
bias is shown to affect the public opinion even when the rational reader understands the bias of the media outlet. Because
the public never knows how much information the media has, the media may present the contrary information more
powerfully, tilt the selection of words, and use other ways to inuence the public’s perception (Anderson & McLaren,
2012).[19]
The research conducted by Hamborg, Donnay, and Gipp (2019) provides insights on how to measure different types of
media bias.[20] This study combines computer science and social science approaches to study media bias and identies the
systematic way of testing it. They dened media bias as three types: selection, coverage, and presentation. The selection
bias, concerning stories media selected for reporting. Coverage bias is related to the transparency of entities or topics
in media coverage. Presentation bias is related to the way that news is reported. In our research, we aim to explore the
impact of the Amazon acquisition on WP’s reporting of Amazon-related news. Our study aims to contribute to the existing
literature by providing insights on how the change of media ownership impacts the media bias in terms of the sentiment,
length, and publication date of the articles.
3. Hypotheses Development
3.1 Length
The size allocation of an article is an indication of the value that a news outlet attributes to the story (Hamborg,
Donnay & Gipp, 2019).[20] As long articles are expected to draw more attention than a short one (Bhowmick, Basu & Mitra,
2009)[21], the media frequently applies favoritism of specic topics and avoidance of others (Druckman & Patrick, 2005)[22].
Consequently, by providing more information in the article, the media gets to raise the awareness of the readers and affect
their perception of the story. Although measuring the length of the articles is a straightforward task, only a few studies
measure the length bias of the media. Such includes research by Saez-Trumper et al. (2013), where a slight correlation
between the importance of a story and length was found.[23]
Prior literature shows that the media has incentives to report profit-maximizing news and thus would dive more
attention to negative events (Niessner & So, 2017)[24] and sensationalism that focus on critical corporate names, well-
recognized brands (Doyle, 2006)[25]. In our research, we assume that in reporting Amazon-related news, WP would have
the incentive to please the owner. As such, we expect that WP would have a lengthier presentation of Good-for-Amazon
articles compared to NYT. Consequently, our rst hypothesis is as follows.
H1: The Washington Post reports lengthier articles on good news for Amazon, and shorter articles about bad news
compared to the New York Times.
3.2 Sentiment
The connotation of an article is a subtle but impactful method of manipulating people's perception. From Bhowmick,
Basu & Mitra (2009), writing style could be affected by "labelling" and "word choice".[21] Readers' reactions to each event
or move highly dependent on the word choice by the author. Wordings used could pose a relatively positive (negative)
signal for readers. By evaluating the emotional signals, one could interpret the bias on Amazon by the Washington Post in
the post-acquisition period.
Thibodeau & Boroditsky (2011) researched to prove that people are more inclined to perceive differently with
different word choices, in their case, "metaphor" in news articles to ignite different sentiments in the reader.[26] Their
experiments showed that when people encounter "metaphor", they are not aware of how these words of metaphor
have exerted influences on them. However, the metaphor often imposes a frame of how people interpret an issue, "by
instantiating frame-consistent knowledge structures, and inviting structurally-consistent inferences... People like to think
they are objective and making decisions based on numbers," Boroditsky said. "They want to believe they are logical.
Nevertheless, they are being swayed by metaphors." (Gorlick, 2011).[27] From the metaphor study on news articles and
readers' reactions, it is understood the importance of word selection and how subtly it could affect readers' cognitive and
sentiments without knowing.
Based on these studies above, we assume that the overall polarity of WP news will be more positive by stressing the
positive news and down-playing on negative news, and our second hypothesis is expressed as follows:
H2: After the sale, the Washington Post reports good news about Amazon using more positive connotation and reports
bad news about Amazon using less negative connotation compared to the New York Times.
Volume 1 Issue 3 | 2020 | 89 Modern Economics & Management Forum
3.3 Publication time
The publication time is a signicant dimension of news reporting that could inuence the materiality or usefulness of
the information provided to investors. Talking about the control of publication time of news, based on previous studies, we
suppose that there are two main reasons. On the one hand, investors are probably interested in collecting timely information
from companies they are involved in. Firms, therefore, react to pressures yielded from investors to disclose earnings more
rapidly (Sengupta, 2014).[28] On the other hand, organizations prefer news concerning earnings information to be released
as rapidly as possible. Sengupta's research suggested that managers have the purpose of delaying the disclosure of negative
news based on their examination of potential benets as well as the costs of releasing earning information. By facilitating a
favourable publication time of news, companies may better manage rm performance.
Qin, Strömberg, & Wu (2018) examine the impact of market competition on political media bias through investigating
the difference between media coverage of Chinese government-owned media and other local and worldwide public
media.[29] The general-interest newspapers were commonly named as Daily, Evening, and Metro according to editorial
policies. Those newspapers vary in different publication time, i.e. "Metros" in the morning, whereas "Evening" in the
afternoon. Despite that researchers did not distinctly illustrate whether different publication time of different newspapers
directly affects political media bias, "Daily" is ofcially monitored by a government-regulated system, while "Metro" and
"Evening" offer subsidiary newspapers that are more associated with the politics-oriented content and less controlled by the
central departments. Accordingly, news publication time is highly possible to be an essential factor in dominating political
media bias.
Interestingly, there is no study examining the reasons why the media differs publication time of news toward certain
critical events in the eld of nancial journalism. However, based on our observation, newspapers tend to tilt news with
controlling news publication time of nancial news. It is worthy of researching whether the difference in news publication
time is related to management's control of media coverage.
In the view of relatively limited previous work providing evidence on the link between publication time of news and
control of media bias, we assume that the earlier news reporting one news is, the more useful information that is likely to
boost the owner's protability and earnings. Regarding the fact that Amazon directly owns WP, we thus assume that WP
prefers to report good news such as disclosure of favourable information about owners quickly for improving nancial
results. Accordingly, we formulate the third hypothesis of our study as follows:
H3: The Washington Post reports bad Amazon-related news later than the New York Times, and good Amazon-related
news earlier than the New York Times.
4. Research design and data
To measure the impact of Amazon ownership on WP coverage, we collect articles that have Amazon mentioning
published by WP and NYT from the Factiva database. Our sample period begins on October 1, 2012, and ends on
September 30, 2014, altogether 24 months, including 12 months before and 12 months after the buyout on October 1, 2013
(Fahri, 2013).[30] By applying tools developed by Google Cloud AI & Machine Learning, we extract all needed information
in the table format, containing the headline, author, publication date, length (words/sentences), rst sentence, sentiment
score, and magnitude.
The sentiment score and magnitude are the numerical values representing the sentiment of the article (positive,
neutral, or negative), which were tested using the sentiment analysis tool developed by Google. According to the research
done by Google Research Team (2008)[31], sentiment score (x) uses the following function:
Sx st
t
n
core =
=
1
The score represents the individual score for a word within a text. The sentiment score ranges from -1.0 (negative)
to +1.0 (positive), showing the overall emotional leaning of the text. Magnitude is the sum of without considering signs,
ranging from 0.0 to +innity. Thus, if the score is "0.0" it can mean two things. (1) If the magnitude is close to 0, the text
is neutral; (2) If the magnitude is large, then the text is mixed (positive and negative scores cancelled each other out). As
such, a score closer to "+1.0" with high magnitude means that text is "clearly positive" and a score closer to "-1.0" with
high magnitude means that text is "clearly negative".
For robustness, each article was then encrypted for the content of the news coverage, i.e. we read each article's
headline and first sentence to identify the positivity of WP and NYT's reporting. The main event of the article can be
described by ve journalistic questions (who did what, when, where, and why), and can be answered by reading the rst
Modern Economics & Management Forum 90 | Tongyi Jin, et al.
few sentences (Hamborg et al., 2018).[32] Then, we lter out events that are reported by both newspapers. Afterwards, we
form two groups of the articles - Good-for-Amazon articles and Bad-for-Amazon articles.
In our primary analysis, we use a difference-in-differences (DID) method to compare how WP and NYT reacted to
identical circumstances over time. It is suitable because both WP and NYT compete for similar audiences on local and
international levels, and have similar economic coverage.
This paper applies the DID method for two datasets (i.e. Good-for-Amazon articles and Bad-for-Amazon articles)
separately. Each dataset includes one treatment group (i.e. WP's articles) and one control group (i.e. NYT's articles). We
compare changes in NYT's reporting style unaffected by the pressure from Amazon's acquisition to the changes over time
and WP's reporting style inuenced by the acquisition intervention and attribute the "difference-in-differences" to the effect
of the acquisition on newspapers' coverage. It provides us unbiased effect estimates if the trend over time would have been
the same between the treatment and control groups in the absence of the intervention.
In our DID model, we assess the impact of the acquisition on WP's coverage by using our extracted data to compare
how signicant is the change in the way newspapers reported on Amazon along with several metrics: sentiment, length,
and publication time. Any signicant changes that we identify can be interpreted as due to the change in ownership. This
means that both papers in the comparison are likely to change their coverage in response to the purchase of WP -- probably
in response to the changing competitive landscape -- but the effect we identify is whether WP experienced an even more
signicant change by comparing with NYT.
5. Empirical analysis
Tables 1 & 2 show means of length, sentiment, and publication of Good-for-Amazon and Bad-for-Amazon articles
published by WP and NYT across time. While Table 1 represents the aggregate mean differences of both papers before and
after the acquisition, Table 2 contrasts the separate mean differences between two papers for the whole time period.
Table 1. Differences before and after acquisition
Good-for-Amazon articles by WP & NYT Bad-for-Amazon articles by WP & NYT
Variables Pre-Acq Post-Acq Diff Pre-Acq Post-Acq Diff
Length 735.9 556.2 179.7*** 666.9 652 14.86
Sentiment score 0.0230 0.0410 -0.0170 -0.0440 -0.0700 0.0260
Publication time 2.8962 3.7079 -0.8117 2.7983 7.7959 -4.9976
t statistics in parentheses
* p < 0.1, ** p < 0.05, *** p < 0.01
From Table 1, it is evident that the length difference of the good-for-Amazon articles published by both papers has
substantial change after the acquisition. The respective p-value is less than 0.01, which indicates the signicance of the
difference at 1%. While it is possible that some of the changes are due to the acquisition, it may not fully explain the
variation identified. By way of comparison, if we look at Table 2, it is clear that the length difference of the good-for
Amazon articles between individual papers has substantial change as well. The corresponding p-value is less than 0.05,
which indicates the significance of the difference at 5%. Therefore, it is reasonable to assume that the acquisition has
impacted WP's coverage in terms of length.
Table 2. Differences of good-for and bad-for Amazon articles
Good-for-Amazon articles Bad-for-Amazon articles
Variables WP NYT Diff WP NYT Diff
Length 588.8 735.3 -146.517** 670.6 628.7 41.95
Sentiment score 0.0410 0.0200 0.0210 -0.0610 -0.0520 -0.0090
Publication time 2.7251 4.1931 -1.4679 4.1578 4.2577 -0.0998
t statistics in parentheses
* p < 0.1, ** p < 0.05, *** p < 0.01
On the other hand, the sentiment difference revealed different patterns — only minor changes took place after
the acquisition and between individual newspapers, with corresponding p-value in both cases being greater than 0.1.
However, from Table 2, we can see that WP reports good-for Amazon news more positively than NYT (0.0410 and
0.0200, respectively), and reports bad-for Amazon news more negatively than NYT (-0.0610 and -0.0520, respectively).
Volume 1 Issue 3 | 2020 | 91 Modern Economics & Management Forum
Apparently, given the nature of sentiment (sum of single words' scores ranging from -1.0 to +1.0), WP seems to be more
emotional in reporting Amazon related news compared to NYT.
As Table 1 shows, the publication time difference of the Bad-for-Amazon articles has a signicant change (-4.9976)
with p-value < 0.05. This could possibly imply that WP published Bad-for-Amazon articles later than NYT. However, the
publication time difference has no signicant changes for Good-for-Amazon articles (-0.8117, p value > 0.1).
To characterise the change more thoroughly and better control for the possible unobservable factors, we apply double
difference regression models to model the differences in length, sentiment, and publication time between the NYT and
WP, using an indicator for the time after the acquisition Bezos and an indicator of the possible difference between the WP
and NYT before the acquisition Graham. Specically, our regression form is:
WP NYTBezos Grahm
tt t
()
=+∗+
∗+
ββ
βδ
01 2
where WPt-NYTt are the mean differences in length or sentiment or publication time in time period t, β0 is the baseline
average difference between WP and NYT, β1 is change in that difference after acquisition, β2 is the difference between the
NYT and WP pre-intervention, and δt is the event matching effects to control for the disparity in Amazon-related articles
and remove confounders. The positive β value indicates an increase in length or sentiment of the Amazon-related articles
and negative β value means the vice versa.
For reference, Table 3 represents the differences in length, sentiment, and publication time of Good-for-Amazon and
Bad-for-Amazon articles before and after the acquisition.
Table 3. Differences in length, publication time and sentiment
Good-for-Amazon articles
Length Sentiment Publication time
(1) (2) (3)
WP - NYT -193.6 -0.00888 6.2920
(Std. Err.) (129.82) (-0.029) (26.75)
Bezos -76.09 0.0211 -0.9009
(-0.87) (0.92) (1.47)
Graham 217.1** -0.0165 -1.0358
(2.34) (-0.85) (1.31)
Intercept 620.8*** 0.0323* 3.2786***
(9.89) (1.93) (0.95)
N 220 220 220
0.0638 0.0159 0.0361
Bad-for-Amazon Articles Length Sentiment Publication Time
(1) (2) (3)
WP-NYT 166.904 -0.0879** -5.7070
(Std. Err.) (143.71) (0.038) (25.46)
Bezos -64.03 -0.0003 -2.6534
(-0.78) (-0.02) (2.54)
Graham -133.6* 0.0568** -0.4781
(-1.76) (2.46) (3.55)
Intercept 706.8*** -0.0611*** 4.2962**
(11.62) (-4.25) (1.91)
N 174 174 174
R20.0102 0.0445 0.0492
t statistics in parentheses
* p < 0.1, ** p < 0.05, *** p < 0.01
In contrast with the patterns evident in Table 1 & Table 2, the rst column of Table 3 reveals that the length difference
of the Amazon related articles did not signicantly change following the acquisition. The results, contrary to our hypothesis
(H1), suggest that length of Good-for-Amazon articles became shorter and length of Bad-for-Amazon articles became
Modern Economics & Management Forum 92 | Tongyi Jin, et al.
longer post-acquisition.
Following the trend, the results reported in the second column of Table 3 indicated that the sentiment difference of
Good-for-Amazon articles has no significant change as well. In contrast, the sentiment difference of Bad-for-Amazon
articles has significant change with p-value significant at 5%. The sentiment difference derived from Bad-for-Amazon
articles seems to follow the patterns revealed in Table 2, whereas the sentiment difference derived from Good-for-Amazon
articles suggest the opposite.
The results in the third column reveal that the publication time difference has insignicant changes for both Good-
for-Amazon and Bad-for-Amazon articles, p value > 0.1. However, the difference value of publication time itself does not
provide any insights. Because publication time difference is the mean of (reporting date - acquisition date), positive value
of the WP-NYT for articles before acquisition means WP reports earlier than NYT, and for articles after acquisition means
WP reports later than NYT. Thus, results obtained conrm our hypothesis H3 because the value of WP-NYT is negative
for both before & after acquisition for Good-for-Amazon articles, -47.064 and -40.772, respectively (Appendix B).
6. Discussion
Our empirical test results show that ownership change at WP has significant influence on publication date of
Amazon-related articles compared to NYT. For length and sentiment of the articles, the existing results do not show much
correlation with the ownership change. We revised literature review and proposed following concerns in enhancing the
research outcome:
6.1 Length
The empirical test results show that the acquisition has no signicant impact on the length difference between the
NYT and WP. In contrast, the existing literature has shown that the size allocation of the article indicates the value media
assigns to the story (Bhowmick, Basu & Mitra, 2009)[21], and accordingly, a signicant change in the length difference of
the Amazon-related articles was expected as Bezos encouraged WP to cover him and Amazon aggressively (Meyer, 2014)[3].
Both the nature of the event (good/bad) and scale (big/small) impact the attention given to the article, but we did
not include the scale of the event in our regression model. To address this limitation we test matching articles written on
the same events. In the double-difference regression, we need to test what is the pattern of changes in length difference
between Good-for-Amazon and Bad-for-Amazon articles. Druckman & Parkin (2005) have discovered that news outlets
report disproportionally much on one topic.[22] Bhowmick, Basu & Mitra (2009)[21] also claimed that longer stories grab
more attention. Thus, to accept our hypothesis, the length difference should be positive for Good-for-Amazon articles and
negative for Bad-for-Amazon articles. Since the empirical analysis results suggest that after the acquisition the length
difference of Good-for-Amazon is negative (-193.6) and positive (166.9) for Bad-for-Amazon articles, and the signicance
level is low, the hypothesis is rejected. WP does not report longer articles on news that are good for Amazon or shorter
articles on news that are bad for Amazon.
One major concern with the results is the potential discrepancy of the data. Given that we identify and match articles
based on headline & rst sentence, there is a possibility of overlooking important articles and/or including articles that
contain the digest of different news that adversely affected the robustness of the results. Since our research focuses on
articles published online, future study could incorporate the placement of article and size of preview on main and other
web-pages (Hamborg, Donnay & Gipp, 2019).[20] In that way, more systematic and comprehensive analysis can be
produced.
6.2 Sentiment
The result has demonstrated media ownership poses no signicant impact on the reporting sentiment of the owner.
The results of DID analysis on sentiment score (See Appendix A) has shown that WP reported Good-for-Amazon articles
less positively than NYT in both the pre-acquisition (-0.017) and the post-acquisition (-0.025) periods. For Bad-for-
Amazon articles, WP has reported them less negatively than NYT in the pre-acquisition (0.057) period and more negatively
in the post-acquisition (-0.031) period. With DID eliminating the unrelated factors, the effect caused by the acquisition is
deduced to be -0.00888 and -0.0879 respectively representing a low signicance level for Good-for-Amazon and Bad-for-
Amazon articles. Meanwhile, literature review suggested that the media frequently frames the subjects/events with various
predetermined words to manipulate our perceptions (Parenti, 1997).[2] Thibodeau & Boroditsky (2011) even showed
that bias would be pervasive in the media as different word sets are adopted by various media.[26] This idea was further
evidenced by Arthur Hays, owner of NYT, who would send notes with instructions, suggestions, complaints and orders
to NYT editor Turner Catledge (Chomsky, 2006) regarding news that ranges from being directly related to the owner to
Volume 1 Issue 3 | 2020 | 93 Modern Economics & Management Forum
personal preferences.[11] The empirical results oppose our assumptions and suggest that the sentiment difference is negative
and minor, and thus, the hypothesis H2 is refuted.
Nonetheless, we consider that results are inconsistent with literature review due to the faults in the sentiment analysis
tool. Since the machine assigns sentiment score to every single word and then sums them together to get the mean, the
negative words dedicated to other companies or people could possibly cancel out the effect of positive words used to
describe Amazon. Thereby, showing a higher negativity in the articles which have mentioned Amazon positively. Another
limitation of the sentiment score is the inability to distinguish emotions. For example, when the article described Amazon
to be angry or sad, the machine would have recognized the article negatively, yet, it does not mean the article is reporting
negatively on Amazon, the post might still be supportive towards Amazon. It is the same for sarcastic sentences as well.
Therefore, we recommend future study to improvise on the sentimental data collection by applying more comprehensive
AI word processing tools.
6.3 Publication time
Based on the empirical results, we can identify the acquisition has significant influence on the publication time
difference between NYT and WP. Prior studies have indicated that the publication time could inuence materiality of the
information provided to investors. Kothari, Li, & Short (2009) claimed that management might delay the publication of
bad-for-company news for uncertainty-increasing content.[33] We have provided a directional solution to predict timeliness
of favorable and unfavorable news.
According to the empirical analysis, the publication time difference of Good-for-Amazon articles is -47.064 before
acquisition and -40.772 after the acquisition, both signicant at 5% (Appendix B). This implies that WP reported Good-
for-Amazon articles later than NYT before the acquisition, and earlier after the acquisition, which is consistent with our
hypothesis H3. Regarding Bad-for-Amazon articles, the empirical analysis revealed that the publication time difference
of articles is -11.873 before acquisition and -5.707 after the acquisition, but results are not significant (Appendix B).
It indicates that WP and NYT publish news around the same time. This result is reasonable as the media bias is often
expressed through the omission of the bad news and/or facts (Anderson & McLaren, 2012).[19] Thus, the hypothesis H3 is
accepted.
One limitation of our study is that we only use 24 months' data. The study of a more extended period could provide
more robust results, which may be predicted to generate a wider range of situational and organizational variation. We also
recommend using higher-quality quantitative methods of data analysis to calculate the publication time difference. Besides,
we have marked 66 signicant events through the matching of all articles. Future research can perform sub-analysis to
excavate more profoundly into question whether the acquisition had a different or similar effect on the publication time
between papers. This would be helpful for future studies to focus more on how different papers respond to relevant events
with the impact of media manipulation.
6.4 Further research on media ownership
Future research on media ownership and bias may explore how media manipulation impacts the company's stock
market performance because this will help us better understand the effects of such coverage on the public. Plenty of
exemplary investigations have shown the signicant impact of media on capital markets. In terms of the general correlation
between media reporting and firm's performance, Tetlock (2007) found that great amounts of media pessimism cause
descending pressure on market prices.[34] His study provided evidence on the actual effect of the WSJ publicizing U.S. stock
market returns. Another study by Huberman & Regev (2001) showed that when the old article on ENMD's cancer research
breakthrough was published by NYT, it caused a dramatic increase in stock prices of ENMD and other biotechnological
companies in the same industry, which was more signicant than the effect of this article when it was reported ve months
ago in Times and Nature.[35]
There are more questions remaining in the eld of media ownership and bias such as what are the consequences faced
by those journalists that report biased news, e.g. career promotion, higher pay, dismissal, prosecution, and etc. As such,
measuring the journalist's incentive could provide insights on hidden conicts within the organization (Turow, 1994) and
extend the ndings on media ownership and bias.[17]
7. Conclusion
This research aims at investigating the relationship between changing media ownership and the bias in media
coverage. We found that change in the ownership at WP correlates with the changes in news reporting, namely, in the
length, sentiment, and publication time of the articles. Based on the results of the empirical study, we conclude that
Modern Economics & Management Forum 94 | Tongyi Jin, et al.
Amazon's acquisition has a favourable impact on the publication time of articles and a relatively small impact on sentiment
and length, however, the results are not robust.
This paper was an attempt to deal with the area of research that had little attention previously. Although our results
are modest, they do guide further study with constructive headers and a deeper understanding of the role of the media for
entrepreneurs. We conclude by remarking that inuence by media ownership is present and media, as an essential outlet
for the information dissemination, will remain to be a potential tool for companies or the government, as an owner, to
manipulate the public's perception.
References
[1] Kanjo W. (2017). What the media won't tell you about the war in Syria: Essays. Place of publication: not identied;
Publisher: not identied.
[2] Parenti M. Methods of media manipulation. The Humanist. 1997; 57(4): 5.
[3] Meyer M. (2014, July-August). Brick by brick. Columbia Journalism Review. Available from: https://archives.cjr.org/
cover_story/washington_post_jeff_bezos.php.
[4] Exploring the Amazon. The Economist. 2015; 415: 64.
[5] Day M., Metcalf T. (2019, July 6). MacKenzie Bezos Gets Ready for a $38 Billion Stake in Amazon. Bloomberg. Re-
trieved from: https://www.bloomberg.com/news/articles/2019-07-05/bezos-split-nalizes-as-38-billion-amazon-stake-
transfer-looms.
[6] Stewart J. (2019, January 21).Why Jeff Bezos’ Divorce Should Worry Amazon Investors. The New York Times. Re-
trieved from Factiva database.
[7] Jeff Bezos, in divorce settlement, retains 75 percent of the couple's Amazon stock and all voting power (2019, April 4).
The Washington Post. Retrieved from Factiva database.
[8] Inquiring minds: How did a tabloid get Jeff Bezos's texts? (2019, February 6). The Washington Post. Retrieved from
Factiva database.
[9] Archer A. M., Clinton J. Changing Owners, Changing Content: Does Who Owns the News Matter for the News? Po-
litical Communication. 2018; 35(3): 353-370.
[10] Lutz A. (2012) These 6 Corporations Control 90% of The Media In America. Business Insider. Available from: https://
www.businessinsider.com/these-6-corporations-control-90-of-the-media-in-america-2012-6.
[11] Chomsky D. “An Interested Reader”: Measuring Ownership Control at the New York Times. Critical Studies in Me-
dia Communication. 2006; 23(1): 1-18.
[12] Harding J. (2002, June 11). Media king warms to his subjects. Financial Times. p. 30.
[13] Szalai G. (2007, January 27). Murdoch: Big media has less sway on Internet. The Hollywood Reporter.
[14] Bagdikian B. H. The Media Monopoly (4th ed.). Boston: Beacon Press; 1992.
[15] Gilens M., Hertzman C. Corporate Ownership and News Bias: Newspaper Coverage of the 1996 Telecommunications
Act. The Journal of Politics. 2000; 62(2): 369-386.
[16] Hanretty C. Media outlets and their moguls: Why concentrated individual or family ownership is bad for editorial in-
dependence. European Journal of Communication. 2014; 29(3): 335-350.
[17] Turow, J. Hidden Conicts and Journalistic Norms: The Case of Self-Coverage. Journal of Communication. 1994;
44(2): 29-46.
[18] Butler A. W., Gurun U. G. Don’t Believe the Hype: Local Media Slant, Local Advertising, and Firm Value. The Jour-
nal of Finance. 2012: 67(2): 561-598.
[19] Anderson S. P., McLaren J. Media Mergers and Media Bias with Rational Consumers. Journal of the European Eco-
nomic Association. 2012; 10(4): 831-859.
[20] Hamborg F., Donnay K., Gipp B. Automated identication of media bias in news articles: an interdisciplinary litera-
ture review. International Journal on Digital Libraries. 2019; (20): 391-415.
[21] Bhowmick P. K., Basu A., Mitra P. Reader Perspective Emotion Analysis in Text through Ensemble-based Multi-La-
bel Classication Framework. Computer and Information Science. 2009; 2(4): 64-74.
[22] Druckman J. N., Parkin M. The impact of media bias: how editorial slant affects voters. J. Polit. 2005; 67(4): 1030-
1049.
[23] Saez-Trumper D., Castillo C., Lalmas M. (2013, October). Social media news communities: gatekeeping, coverage,
and statement bias. In Proceedings of the 22nd ACM International Conference on Information & Knowledge Manage-
ment (pp. 1679-1684).
[24] Niessner M., So E. (2017) Bad News Bearers: The Negative Tilt of Financial Press. Social Science Research Network.
[25] Doyle G. Financial news journalism: A post-Enron analysis of approaches towards economic and nancial news pro-
Volume 1 Issue 3 | 2020 | 95 Modern Economics & Management Forum
duction in the UK. SAGE Publications. 2006; 7(4): 433-452.
[26] Thibodeau P. H., Boroditsky L. Metaphors we think with: The role of metaphor in reasoning. PloS one. 2011; 6(2):
e16782.
[27] Gorlick A. (2011, Feb 23) Is crime a virus or a beast? When describing crime, Stanford study shows the word you
pick can frame the debate on how to ght it. Stanford News. Available from: https://news.stanford.edu/news/2011/feb-
ruary/metaphors-crime-study-022311.html.
[28] Sengupta P. Disclosure timing: Determinants of quarterly earnings release dates. Journal of Accounting and Public
Policy. 2004; 23(6): 457-482.
[29] Qin B., Strömberg D., Wu Y. Media bias in China. American Economic Review. 2018; 108(9): 2442-76.
[30] Fahri P. (2013, Oct 1) The Washington Post Closes Sale to Amazon Founder Jeff Bezos. The Washington Post. Avail-
able from: https://www.washingtonpost.com/business/economy/washington-post-closes-sale-to-amazon-founder-jeff-
bezos/2013/10/01/fca3b16a-2acf-11e3-97a3-ff2758228523_story.html.
[31] Google Research Team: Blair-Goldensohn S., Neylon T., Hannan K., Reis G. A., McDonald R., Reynar J. (2008)
Building a Sentiment Summarizer for Local Service Reviews. Google. Available from: https://research.google/pubs/
pub34368/.
[32] Hamborg F., Lachnit S., Schubotz M., Hepp T., Gipp B. (2018) Giveme5W: Main Event Retrieval from News Articles
by Extraction of the Five Journalistic W Questions. In: Proceedings of the iConference.
[33] Kothari S. P., Li X., Short J. E. The effect of disclosures by management, analysts, and business press on cost of capi-
tal, return volatility, and analyst forecasts: A study using content analysis. The Accounting Review. 2009; 84(5): 1639-
1670.
[34] Tetlock P. C. Giving Content to Investor Sentiment: The Role of Media in Stock Market. The Journal of Finance.
2007; 62(3): 1139-1168.
[35] Huberman G., Regev T. Contagious Speculation and a Cure for Cancer: A Nonevent that Made Stock Prices Soar. The
Journal of Finance. 2001; 56(1): 387-396.
Modern Economics & Management Forum 96 | Tongyi Jin, et al.
Appendix A
DID of sentiment — Good-for-Amazon articles
DID of sentiment — Bad-for-Amazon articles
Volume 1 Issue 3 | 2020 | 97 Modern Economics & Management Forum
Appendix B
DID of publication time — Good-for-Amazon articles
DID of publication time — Bad-for-Amazon articles
Remarks: Control — NYT; Treated — WP
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Media bias, i.e., slanted news coverage, can strongly impact the public perception of the reported topics. In the social sciences, research over the past decades has developed comprehensive models to describe media bias and effective, yet often manual and thus cumbersome, methods for analysis. In contrast, in computer science fast, automated, and scalable methods are available, but few approaches systematically analyze media bias. The models used to analyze media bias in computer science tend to be simpler compared to models established in the social sciences, and do not necessarily address the most pressing substantial questions, despite technically superior approaches. Computer science research on media bias thus stands to profit from a closer integration of models for the study of media bias developed in the social sciences with automated methods from computer science. This article first establishes a shared conceptual understanding by mapping the state of the art from the social sciences to a framework, which can be targeted by approaches from computer science. Next, we investigate different forms of media bias and review how each form is analyzed in the social sciences. For each form, we then discuss methods from computer science suitable to (semi-)automate the corresponding analysis. Our review suggests that suitable, automated methods from computer science, primarily in the realm of natural language processing, are already available for each of the discussed forms of media bias, opening multiple directions for promising further research in computer science in this area.
Conference Paper
Full-text available
Extraction of event descriptors from news articles is a commonly required task for various tasks, such as clustering related articles, summarization, and news aggregation. Due to the lack of generally usable and publicly available methods optimized for news, many researchers must redundantly implement such methods for their project. Answers to the five journalistic W questions (5Ws) describe the main event of a news article, i.e., who did what, when, where, and why. The main contribution of this paper is Giveme5W, the first open-source, syntax-based 5W extraction system for news articles. The system retrieves an article's main event by extracting phrases that answer the journalistic 5Ws. In an evaluation with three assessors and 60 articles, we find that the extraction precision of 5W phrases is p = 0.7.
Conference Paper
Full-text available
We examine biases in online news sources and social media communities around them. To that end, we introduce unsupervised methods considering three types of biases: selection or ``gatekeeping'' bias, coverage bias, and statement bias, characterizing each one through a series of metrics. Our results, obtained by analyzing 80 international news sources during a two-week period, show that biases are subtle but observable, and follow geographical boundaries more closely than political ones. We also demonstrate how these biases are to some extent amplified by social media.
Article
The press is essential for creating an informed citizenry, but its existence depends on attracting and maintaining an audience. It is unclear whether supply-side effects—including those dictated by the owners of the media—influence how the media cover politics, yet this question is essential given their abilities to set the agenda and frame issues that are covered. We examine how ownership influences media behavior by investigating the impact of Rupert Murdoch’s purchase of the Wall Street Journal (WSJ) in August 2007. We collect data on every front-page story and editorial for 27 months, and we compare the difference in political coverage between the New York Times (NYT) and WSJ using a difference-in-differences design. We show that the amount of political content in the opinion pages of the two papers were unchanged by the sale, but the WSJ’s front-page coverage of politics increased markedly relative to the NYT. Similar patterns emerge when comparing the WSJ’s content to USA Today and the Washington Post. Our finding highlights potential limits to journalists’ ability to fulfill their supposed watchdog role in democracies without interference from owners in the boardroom.
Article
Existing research on discretionary disclosures provides valuable insights on the potentials causes and consequences of alternative forms of disclosure. However, relatively little is known about how managers choose to time the release of financial information. This paper focuses on the quarterly earnings release dates and investigates why some choose to release earnings information relatively early, compared to others. The results indicate that the reporting lag (days between fiscal period end and quarterly earnings release date) is shorter for firms facing greater demand for information from investors and greater litigation costs. The reporting lag, however, is longer for firms with greater block ownership and those whose operations are somewhat more complex.
Article
We document systematic evidence of risk effects of disclosures culled from a virtually exhaustive set of sources from the print medium. We content analyze more than 100,000 disclosure reports by management, analysts, and news reporters (i.e., business press) in constructing firm-specific disclosure measures that are quantitative and amenable to replication. We expect credibility and timeliness differences in the disclosures by source, which would translate into differential cost of capital effects. We find that when content analysis indicates favorable disclosures, the firm's risk, as proxied by the cost of capital, stock return volatility, and analyst forecast dispersion, declines significantly. In contrast, unfavorable disclosures are accompanied by significant increases in risk measures. Analysis of disclosures by source-corporations, analysts, and the business press-reveals that negative disclosures from business press sources result in increased cost of capital and return volatility, and favorable reports from business press reduce the cost of capital and return volatility.
Article
This article investigates the levels of owner influence in 211 different print and broadcast outlets in 32 different European media markets. Drawing on the literature from industrial organization, it sets out reasons why we should expect greater levels of influence where ownership of individual outlets is concentrated, where it is concentrated in the hands of individuals or families and where ownership groups own multiple outlets in the same media market. Conversely, we should expect lower levels of influence where ownership is dispersed between transnational companies. The article uses original data on the ownership structures of these outlets and combines it with reliable expert judgements as to the level of owner influence in each of the outlets. These hypotheses are tested and confirmed in a multilevel regression model of owner influence. The findings are relevant for policy on ownership limits in the media and for the debate over transnational versus local control of media.
Article
The collapse of Enron and other corporate scandals have raised concerns about the efficacy of financial journalism. Based on research on where reporters get their ideas for stories and how they approach their work, this article explores the particular circumstances in which production of financial and economic news takes place. The author argues that, while reporters are generally highly sceptical about ‘spin’ and strongly inclined towards highlighting instances of corporate underperformance and mismanagement, the circumstances and constraints they work within nonetheless make it unlikely that financial irregularities obscured within company accounts will be detected on a routine or consistent basis. Moreover, the way in which the commercial sector is organized (with in-depth analysis generally confined to specialist media whose audiences are already financially literate) means that the task of facilitating a sound public grasp over the significance of financial and economic news developments is largely being neglected.