movieratings14-10-27.dvi

NBER WORKING PAPER SERIES

DOES CONFLICT OF INTEREST LEAD TO BIASED COVERAGE? EVIDENCE

FROM MOVIE REVIEWS

Stefano DellaVigna

Johannes Hermle

Working Paper 20661

http://www.nber.org/papers/w20661

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue

Cambridge, MA 02138

November 2014

A previous version of this paper circulated in 2011 with the title `Does Media Concentration Lead

to Biased Coverage? Evidence from Movie Reviews' with Alec Kennedy as collaborator. Ivan Balbuzanov,

Natalie Cox, Tristan Gagnon-Bartsch, Jordan Ou, and Xiaoyu Xia provided excellent research assistance.

We thank Marianne Bertrand, Saurabh Bhargava, Fanny Camara, Lucas Davis, Casey Dougal, David

Dranove, Matthew Ellman, Ignacio Franceschelli, Matthew Gentzkow, Fabrizio Germano, Austan

Goolsbee, Emir Kamenica, Brian Knight, Jesse Shapiro, Noam Yuchtman, Joel Waldfogel and audiences

at Brown University, Boston University, Chicago Booth, the Paris School of Economics, UC Berkeley,

and at the 2011 Media Conference in Moscow for very helpful comments. We also thank Bruce Nash

for access to data from the-numbers, as well as helpful clarifications about the industry. The views

expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau

of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been peer-

reviewed or been subject to the review by the NBER Board of Directors that accompanies official

NBER publications.

to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including

Does Conflict of Interest Lead to Biased Coverage? Evidence from Movie Reviews

Stefano DellaVigna and Johannes Hermle

NBER Working Paper No. 20661

November 2014

JEL No. D72,L41

ABSTRACT

Media outlets are increasingly owned by conglomerates, inducing a conflict of interest: a media outlet

can bias its coverage to benefit companies in the same group. We test for bias by examining movie

reviews by media outlets owned by News Corp.—such as the Wall Street Journal—and by Time Warner—

such as Time. We use a matching procedure based on reported preferences to disentangle bias due

to conflict of interest from correlated tastes. We find no evidence of bias in the reviews for 20th Century

Fox movies in the News Corp. outlets, nor for the reviews of Warner Bros. movies in the Time Warner

outlets. We can reject even small effects, such as biasing the review by one extra star (out of four)

every 13 movies. We test for differential bias when the return to bias is plausibly higher, examine

bias by media outlet and by journalist, as well as editorial bias. We also consider bias by omission:

whether the media at conflict of interest are more likely to review highly-rated movies by affiliated

studios. In none of these dimensions do we find systematic evidence of bias. Lastly, we document

that conflict of interest within a movie aggregator does not lead to bias either. We conclude that media

reputation in this competitive industry acts as a powerful disciplining force.

Stefano DellaVigna

University of California, Berkeley

Department of Economics

549 Evans Hall #3880

Berkeley, CA 94720-3880

and NBER

[email protected]

Johannes Hermle

University of Bonn

Bonn, Germany

[email protected]

1Introduction

On Dec. 13, 2007, News Corp. oﬃcially acquired Dow Jones & Company, and hence the Wall

Street Journal, from the Bancroft family. The acquisition w as controversial in part because of

concerns about a conﬂict of interest. Unlike the Bancroft family whose holdings were limited

to Dow Jones & Company, Murdoch’s business holdings through News Corp. include a movie

production studio (20th Century Fox), cable channels such as Fox Sports and Fox News, and

satellite televisions in the Sky group, among others. The Wall Street Journal coverage of

businesses in these sectors may be biased to beneﬁt the parent company, News Corp.

The Wall Street Journal case is hardly unique. Media outlets are increasingly controlled by

large corporations, such as Comcast, which owns NBC and Telem undo, the Hearst Corporation,

which o wns a network of newspapers and ESPN, and Time Warner, whic h owns HBO, CNN,

and other media holdings. Indeed, in the highly competitive media industry, consolidation

with the ensuing economies of scale is widely seen as a necessary condition for survival.

But is this consolidation without cost for the quality of coverage given the induced conﬂict

of interest? Addressing this question is important, since potential biases in co verage can

translate into a policy concern in the presence of sizeable persuasion eﬀects from the media

(e.g., DellaVigna and Kaplan, 2007; Enikolopov, Petrova, and Zhuravskaya, 2011).

Yet should we expect coverage to be biased due to consolidation? If consumers can detect

the bias in coverage due to cross-holdings and if media reputation is paramount, no bias should

occur. If consumers, instead, do not detect the bias perhaps because they are unaware of the

cross-holding, coverage in the conglomerate is likely to be biased.

Despite the importance of this question, there is little systematic evidence on distortions in

cov e rage induced by cross-holdings. In this paper, we study two conglomerates–News Corp.

and Time-Warner–and measure how media outlets in these groups review movies distributed

by an aﬃliate in the group–such as 20th Century Fox and Warner Bros. Pictures, respectively.

The advantage of focusing on movie reviews is that they are frequent, quantiﬁable, and are

believed to inﬂuence ticket sales (Reinstein and Snyder, 2005), with monetary beneﬁts to the

studio distributing the movie. As such, they are a potential target of distortion.

To iden tify the bias, we adopt a diﬀerence-in-diﬀerence strategy. We compare the review

of movies distributed by 20th Century Fox by, say, the Wall Street Journal to the reviews by

outlets not o wned by News Corp. Since the Wall Street Journal may have a diﬀerent rating

scale, we use as a further control group the reviews of movies distributed by diﬀerent studios,

such as Paramount. If the Wall Street Journal provides systematically more positiv e reviews

for 20th Century Fox movies, but not for Paramount movies, we infer that conﬂict of interest

induces bias.

Still, a legitimate worry is that this comparison ma y capture correlation in taste, rather

than bias. The Wall Street Journal may provide more positive reviews to, say, action movies

of the type distributed b y 20th Century Fox because this reﬂects the tastes of its audience (or

of its journalists), not because of conﬂict of interest. When estimating the impact of conﬂict

of interest, one would like to compare the reviews of a given 20th Century Fox movie only to

reviews of movies by other studios which share the same st yle. In principle, one could attempt

to ﬁnd movies with similar features such as genre. This approach however is infeasible because

users may like (or dislike) movies based on ﬁne features or details which are not observable.

Instead, we propose a reported preference approach. We say that movies A and B are

comparable if viewers that like movie A also like movie B, and vice versa users who dislike movie

A also dislike movie B. This approach does not require any information on mo vie characteristics,

since preferences already distill the relevant features. It does however require a rich individual-

level data set of movie ratings by poten tial audience members.

We take advantage of three rich such data sets by Netﬂix, Flixster,andMovieLens.Using

the individual user ratings, we ﬁnd, for each 20th Cen tury Fox and Warner Bros. movie, the ten

movies distributed b y other studios which are most comparable by the above reported prefer-

ence criterion. In a validation test, we show that matching movies are likely to share the genre,

the MPAA rating, and the average rating by movie review ers, among other characteristics. We

thus use these matching movies as comparison group.

We start from a data set of half a million professional reviews of movies released from 1985

(the y ear in which News Corp. acquired 20th Century Fox) until 2010 (the year in which the

user ratings data ends). The data sources are the online aggregators Metacritic and Rotten

Tomatoes. We compare the reviews by 324 outlets with no conﬂict of interest (known to us)

to the reviews issued by 12 media outlets with cross-holdings. Eight media outlets are owned

by News Corp. during at least part of the sample–the U.S. newspapers Chicago Sun-Times

(owned until 1986), New York Post (owned until 1988 and after 1993), and Wall Str eet Journal

(owned from 2008), the U.K. newspapers News of the World, Times and Sunday Times,the

weekly TV Guide (o wned from 1988 until 1999) and the website Beliefnet (owned from 2007

to 2010). Four media outlets are owned by Time Warner–the weeklies Entertainment Weekly

and Time as well as CNN and the online service Cinematical (owned from 2004 until 2009).

We provide six pieces of evidence on the extent, type, and channel of bias. In the ﬁrst

test, we compare the reviews of movies distributed by the studios at conﬂict of interest to the

reviews of the ten matching movies. We test for average bias in outlets at conﬂict of interest,

such as the Wall Street Journal and Time magazine.

We ﬁnd no evidence of bias for either the News Corp. or Time Warner outlets. In the

benchmark speciﬁcation we estimate an average bias of -0.2 points out of 100 for News Corp.

and of 0 points for Time Warner. The richness of the data ensures tight conﬁdence intervals for

the ﬁnding of no bias. We can reject at the 95% level a bias of 1.9 points for News Corp. and of

1.7 points for Time Warner, corresponding to a one-star higher review score (on a zero-to-four

scale)foroneoutof13moviesatconﬂict of interest. We ﬁnd similar results on the binary

‘freshness’ indicator emplo yed by Rotten Tomatoes.

We underscore the importance of the matching procedure for the estimates of bias: cross-

sectional regressions on the whole sample of movies yield statistical evidence of bias for News

Corp. This seeming bias depends on the inclusion in the control group of movies that are

not comparable to the movies at conﬂict of interest, thus biasing the estimates. Further, we

provide direct evidence on correlated tastes: News Corp. outlets provide more positiv e reviews

to movies distributed by other studios when they are similar to the 20th Century Fox movies.

Second, while there appears to be no bias ov erall, a bias may be detectable for movies where

the return to bias is plausibly larger, holding constant the reputational cost of bias to the media

outlets. While we do not have measures of the return to bias, we consider dimensions which

are lik ely to correlate with it. We expect that movies with generally higher review scores are

likely to have higher return to bias, as an extra star is likely to matter more if it is the 4th star

out of 4, as compared to the second star. Also, movies distributed by the mainstream studios,

movies with larger budgets or larger box oﬃce sales are likely to have higher returns to bias.

We ﬁnd no systematic pattern of diﬀerential bias in this respect.

Third, the overall result of no bias ma y mask heterogeneity in bias by the individual outlets.

We ﬁnd no overall statistical evidence in the t welve outlets, with more precise null eﬀects for

the New York Post and TV Guide (News Corp.) as w ell as for Entertainment Weekly and

Time (Time Warner). Given that each outlet employs a small number of reviewers, we go

further and test for bias by journalist, and again do not ﬁnd any systematic evidence of bias.

Fourth, we test for bias at the editorial level by examining the assignment of movies to

review ers. Since reviewers diﬀer in the average generosit y of their reviews, even in the absence

of bias at the journalist level, assignment of mov ies to more generous review ers would generate

some bias. We ﬁnd no evidence that aﬃliated movies are more likely to be assigned to reviewers

who are on average more positive, conﬁrming the previous results.

So far we tested for bias by commission: writing more positive reviews for movies at conﬂict

of interest. In our ﬁfth piece of evidence, we examine bias by omission. A reviewer that intends

to beneﬁtanaﬃliated studio may selectively review only above-average movies by this studio,

while not granting the same beneﬁt to movies by other studios. This type of bias would not

appear in the previous analysis, which examines bias conditional on review. Bias by omission

is generally hard to test for, since one needs to know the universe of potential news items.

Movie reviews is a rare setting where this is the case, and thus allows us to test for this form

of bias which plays a role in models of media bias (e.g., Anderson and McLaren, 2012).

We thus examine the probability of reviewing a movie as a function of the average review the

movie obtained in control outlets. The reviewing patterns do not diﬀer for movies at conﬂict

of in terest versus the matching mo vies of other studios, thus providing no evidence of omission

bias. We show how apparent evidence of omission bias for Time magazine reﬂects a spurious

pattern, since it appears also before Time became part of the Time Warner conglomerate.

The sixth and ﬁnal piece of evidence on bias examines conﬂictofinterestforthemovie

aggregator itself. Rotten Tomatoes was independent when launched in 1998, was then acquired

by News Corp. in September 2005, only to be divested in January of 2010. This ownership

structure generates an incentive for Rotten Tomatoes to assign more positive reviews (its

‘freshness’ indicator) to 20th Century Fox movies during the period of News Corp. ownership.

This test of bias is particularly powerful: bias is identiﬁed within a media outlet and by

comparison of the Rotten Tomatoes review versus the Metacritic score for the same movie

review. Further, the public is lik ely to be largely unaware of the cross-ownership, making bias

harder to detect by the audience. Once again, we ﬁnd no evidence of bias, even where bias

would be hardest to detect (and hence presumably most likely), for unscored reviews which

are evaluated qualitatively by the Rotten Tomatoes staﬀ.

Overall, reputation-based incentives appear to be eﬀective at limiting the occurrence of

bias: we ﬁnd no evidence of bias by commission, no evidence of editorial bias, no systematic

evidence of bias by omission, and no evidence of bias among the aggregators.

Using these results, we compute a back-of-the-envelope bound for the value of reputation.

Assume that an extra star (out of 4) persuades 1 percent of readers to watch a movie, an

eﬀect in the lower range of estimates of persuasion rates (DellaVigna and Gentzkow, 2010) and

smaller than the estimated impact of media reviews of Reinstein and Snyder (2005), though

admittedly we have no direct evidence.

Under this assumption, an extra star in a single movie

review for a 20th Century Fox mov ie in a newspaper like the New York Post with a circulation

of about 500,000 readers would add approx imately $40,000 in proﬁts for News Corp.

If the

New York Post had biased by one star all reviews for the 448 20th Century Fox movies released

since 1993, the proﬁt could have been nearly $20m. The fact that such systematic bias did

not take place indicates a higher value of reputation. The potential returns are about an order

of magnitude larger for a periodical like Time magazine, with an average of 4 million readers,

and so is the value of the reputation.

This paper relates to a vast literature on conﬂict of interest and most speciﬁcally in the

media (e.g., Hamilton, 2003; Ellman and Germano, 2009). Reuter and Zitzewitz (2006) and Di

Tella and Franceschelli (2011) ﬁnd that media outlets bias their co verage to earn advert ising

revenue. We conjecture that the diﬀerence from our result of no bias is lik ely due to two

factors. First, the revenue from the ads directly beneﬁts the media outlet, while the revenue

from increased movie attendance w ould accrue to the shareholders of the conglomerate, a

more indirect link. Second, movie reviews are easy to compare across a number of outlets

while coverage of mutual fund performance and of scandals are less so.

The average 20th Century Fox and Warner Bros. movie has a total domestic box oﬃce audience of around

7millionviewers,thatis,about3percentoftherelevant US population. Newspapers and magazine readers

are likely to ha ve higher rates of mo vie attendance, which could be 5 percent. The 1 percent persuasion rate

thus implies that, say, a very positiv e review (4 stars out of 4) compared to a positiv e review (3 stars) increases

the share of readers who decide to watch the movie from, say, 5 to 6 percent, a magnitude we ﬁnd plausible.

Of course, the potential eﬀect of reviews on attendance is larger for high-proﬁle movies, and smaller for more

indie-type movies, a diﬀerence that we consider in the paper.

Thestudiosreceiveabouthalfoftheboxoﬃce sales (at an average price of $8 per ticket), and about another

half from higher DVD and TV royalties. P ersonal communication with Bruce Nash, founder of the-numbers.com.

A small number of papers considers media bias due to consolidation, as we do. Gilens and

Hertzman (2000) provide some evidence that the coverage of the debate on TV deregulation

is biased by conﬂict of interest. Goolsbee (2007) and Chipty (2001) examine the extent to

which v ertical integration in the entertainment industry aﬀect network programming and cable

oﬀering. Dobrescu, Luca, and Motta (2013) estimate the bias in 1,400 book rev iews due to

aﬃliation with the outlet reviewing the book; consistent with our ﬁndings, their evidence of

apparent bias is most consistent with correlated tastes, not conﬂict of interest. Rossman (2011)

and Ravid, Wald, and Basuroy (2006) examine the exten t of bias in movie reviews, including

due to conﬂict of interest. Both papers use a small sample of reviews–about 1,000 reviews

for Rossman (2011) and about 5,000 reviews for Ravid et al. (2006). Relative to these papers,

the granularity of information embedded in half a million reviews and the matching procedure

allow us to obtain more precise measures and study the bias in a nu mber of novel directions,

such as editorial bias and bias b y omission. Camara and Dupuis (2014) estimate a cheap talk

game using movie reviews, including in the estimates a parameter for conﬂict of int erest.

This paper also relates to the economics of the media (Str¨omberg 2004; George and Wald-

fogel, 2006; DellaVigna and Kaplan, 2007; Mullainathan, Schwartzstein, and Shleifer, 2008;

Knight and Chiang 2011; Enikolopov, Petrova, and Zhuravskaya 2011; Dougal et al., 2012),

and in particular to papers on media bias (Groseclose and Milyo, 2005; Gentzkow and Shapiro,

2010; Larcinese, Puglisi and Snyder, 2011; Durante and Knight 2012). Within the context of

movie reviews we address questions that have arisen in this literature–such as whether bias

occurs by omission or commission and the role of journalists versus that of editors–about

which there is little evidence.

Finally, the paper relates to the literature on disclosure, such as reviewed in Dranove and

Jin (2010). In our settings, media outlets do not withhold reviews for low-quality aﬃliated

movies, consistent with the Milgrom and Roberts (1986) unraveling result. Brown, Camerer,

and Lovallo (2012) instead provide evidence of strategic movie releases by studios with cold

openings for low-quality movies.

The remainder of the paper is as follows. In Section 2 we in troduce the data and explain

the movie matching procedure, in Section 3 we present the results of the tests of bias due to

conﬂict of interest, and in Section 4 we conclude.

2Data

2.1 Movie Reviews

Media Review Aggregators. The data used in this paper comes from two aggregators,

metacritic.com and rottentomatoes.com. Both sites collect reviews from a variety of media and

publish snippets of those reviews, but they diﬀer in their scoring rules. M etacritic assigns a

score from 0 to 100 for each review, and then averagessuchscoresacrossallreviewsofamovie

to generate an o verall score. For reviews with a numeric evaluation, such as for the New York

Post (0-4 stars), the score is a straightforward normalization on a 0-100 scale. For reviews

without a numerical score, such as primarily for Time magazine, Metacritic staﬀers evaluate

the review and assign a score on the same 0-100 scale (typically in increments of 10).

Rotten Tomatoes does not use a 0-100 score, though it reports the underlying rating for

reviews with a score. It instead classiﬁes each movie as ‘fresh’ or ‘rotten’, and then computes

ascoreforeachmovie—thetomatometer — as the percent of reviews which are ‘fresh’. For

quantitative reviews, the ‘freshness’ indicator is a straightforward function of the rating: for

example, movies with 2 stars or fewer (out of 4) are ‘rotten’, movies with 3 or more stars are

‘fresh’, and movies with 2.5 stars are split based on a subjective judgment. For reviews with

no quantit ative score, the movie is rated as ‘fresh’ or ‘rotten’ by the staﬀ.

The two data sets have diﬀerent advantages for our purposes. Metacritic contains more

information per review, since a review is coded on a 0-100 scale, rather than with a 0 or 1

score. Rotten Tomatoes, however, contains about ﬁve times as many reviews as Me tacritic,

due to coverage of more media (over 500 compared to less than 100) and a longer time span.

We take advantage of both data sets and combine all reviews in the two data sets for movies

released since 1985 and reviewed up until July 2011 in the Metacritic website and until March

2011 on the Rotten Tomatoes website. We eliminate earlier reviews because the review data

for earlier years is sparse, and before 1985 there is no conﬂict of interest: Newscorp. acquired

20th Cen tury Fox in 1985 and the conglomerate Time Warner was created in 1989.

We merge the reviews in the two data sets in two steps. First, we match the mo vies by

title, year and studio with an approximate string matching procedure, checking manually the

imperfect matches. Then, we match reviews of a given movie by media and name of the

review er.

We then exclude movies with fewer than 5 reviews and media with few er than 400

reviews, for a ﬁnal sample of 540,799 movie reviews.

To make the two data sets compatible, we then apply the Metacritic conv ersion into a 0-100

scale to the Rotten Tomatoes reviews whic h report an underlying quantitative score. To do

so, we use the reviews present in both data sets and assign to each Rotten Tomatoes score the

corresponding median 0-100 score in the Metacritic data, provided that there are at least 10

reviews present in both samples with that score. For a small number of other scores which are

common in Rotten Tomatoes but not in Metacritic, we assign the score ourselves following the

procedure of the Metacritic scoring rules (e.g., a score of 25 to a movie rated ‘2/8’).

Media Outlets. The data set includes eight media outlets within the News Corp. con-

glomerate: the American newspapers Chicago Sun-Times (owned by News Corp. only up un til

1986), New York Post (o wned until 1988 and after 1992), and Wall Street Journal (owned

from 2008), the British newspapers News of the World, Times and Sunday Times (all owned

throughout the period), the magazine TV Guide (owned from 1988 until 1999) and the website

We allow for the year of the movies in the two data sets to diﬀer by one year.

Beliefnet (owned from 2007 to 2010). The number of reviews and the data source diﬀer across

these outlets. The British newspapers are represented only in Rotten Tomatoes and have less

than 1,000 reviews each. The New York Post is represented in both data sets and has the most

reviews (5,657). TV Guide and Wall Street Journal have a relatively high number of reviews,

but only a minority while owned b y News Corp. All but one of these eight media (the Wall

Street Journal) have quantitative scores in the reviews. These media employ as reviewers a

small n umber of journalists who stay on for several years, and often for the whole time period.

Therefore, within each media the two most common reviewers (three for the New York Post)

cover the large majority of the reviews, with two media using essentially only one reviewer:

Chicago Sun-Times (Roger Ebert) and the Wall Street Journal (Joe Morgenstern).

The second media conglomerate, Time Warner, includes four media: the w eekly magazines

Time and Entertainment Weekly (both owned by Time Warner from 1990 on), CNN (owned

from 1996) and the web service Cinematical (owned between 2007 and 2010). The reviews in

these media are at conﬂict of in terest with Warner Bros. movies, since the studio was acquired

in 1989 by Time, Inc. Two of the four outlets — CNN and Time — use only qualitative reviews;

since the reviews from CNN are only in the Rotten Tomatoes data set, there is almost no

0-100 score for these reviews, but only a freshness rating. Most of the observations are from

Entertainment Weekly, with more than 4,000 reviews. These outlets, like the News Corp.

outlets, employ only one or t wo major reviewers.

Studios. Dozens of diﬀerent studios distribute the 11,832 movies reviewed in our data

set, including the 6 majors 20th Century Fox, Columbia, Disney, Paramount, Universal, and

Warner Bros. Among the distributors owned by News Corp., 20th Century Fox movies are the

largest group (426 movies), follo wed by Fox Searchlight which distributes movies in the ‘indie’

category. Among the studios owned by Time Warner, the largest distributor is Warner Bros.,

followed by a number of distributors of ‘indie’ movies: Fine Line, New Line, Picturehouse, and

Warner Independent. In most of the following analysis, w e group all the studios in to those

owned by News Corp., which we call for brevity 20th Century Fox, and those owned by Time

Warner, which we call Warner Bros.

Additional Movie Information. We also merge this data set to additional information

on movies from the-numbers.com, including the genre and the MPAA rating.

2.2 Matching Procedure

User Ratings. We employ user-generated movie ratings from Netﬂix, Flixster,andMovieLens

to ﬁnd the movies most similar to a 20th Century Fox or Warner Bros. movie.

Netﬂix is an online movie streaming service. Users rate movies on a scale from 1 to 5 with

1-point increments, typically right after watching a movie. Netﬂix made public a large data set

of (anonymized) reviews as part of its Netﬂix prize competition. This dataset contains roughly

100 million ratings by 480,000 users of 17,700 movies released up to 2005.

Flixster is a social network for users interested in the ﬁlm industry. Besides other services,

Flixster oﬀers movie recommendations based on user ratings. We use a subset of this data

which is available at http://www.cs.ubc.ca/˜jamalim/datasets/. The rating scale ranges from

.5 to 5 in .5 steps. The dataset contains about 8 million ratings by 150,000 users on 48,000

movies released up to 2010.

MovieLens is an online movie recommendation service launc hed by GroupLens Researc h at

the Univ ersity of Minnesota. The service provides users with recommendations once a suﬃcient

number of ratings has been entered (using the same .5 to 5 scale as in Flixster). The dataset,

which can be downloaded at http://www.grouplens.org/datasets/movielens/, was designed for

research purposes. It provides 7 million ratings by roughly 70,000 users on more than 5,000

movies released up to 2004.

Online Appendix Table 1 summarizes the key features of the three samples. Netﬂix has

the most comprehensive data set of reviews but, like MovieLens, it does not cover more recent

movies. Flixster covers the most recent y ears but it is a smaller data set and has a small

number of ratings per user. We use all three data sets, and perform the matches separately

before aggregating the results.

To determine the movie matc hes for a particular 20th Century Fox or Time Warner movie

based on the user-generated reviews, we use the following procedure. Given movie  by 20th

Century Fox, we narrow down the set of potential matching movies 



according to four criteria:

(i) the distributing studio of a mo vie  ∈ 



is not part of the same conglomerate as  in order

to provide a conﬂict-of-interest-free comparison; (ii) at least 40 users reviewed both movie 

and movie  so as to guarantee enough precision in the similarity measure; (iii) movie  is

represented in either the Metacritic or Rotten Tomatoes data set; (iv) movies  and  are close

on t wo variables: the diﬀerence in release years does not exceed 3 years, and the absolute

log-diﬀerence of the number of individual user ratings is not larger than .5.

Amongtheremainingpotentialmatches ∈ 



for movie , we compute the mean absolute

diﬀerence in individual ratings between movie  and a movie  as 





|



− 



|,where

we aggregate over all users  who reviewed both movies (hence the requirement  ≥ 40). We

then keep the 10 movies with the lowest distance measure 



To determine the overall best ten matc hes for movie , we pool the matching mo vies across

the three data sets. If movie  is presen t in only one data set, say because it was released after

2006 and th us is only in Flixster, w e take the ten matches from that data set. If mo vie  is

present in multiple data sets, we take the top match in each data set, then move to the second

best match in each data set, and so on until reaching ten unique matches.

We denote as a

Within each of the three data sets, w e match the movies to the movies in the Metacritic/Rotten Tomatoes

data set using a parallel procedure to the one used when merging the Metacritic and Rotten Tomatoes data.

This allows us also to import the information on the year of release of the movie, used below.

We take matches from Netﬂix ﬁrst, then MovieLens,thenFlixster. Notice that to identify the top 10

matchesoverall,onemayneedtogodownto,say,thetop5matchesorlowerevenwiththreedatasets,given

that the diﬀeren t data sets ma y yield the same matching movie .

movie group the set of 11 movies consisting of movie  and its ten closest matches. Later, we

examine the robustness of the results to alternative matching procedures.

Main Sample. We illustrate the sample construction with an example in Table 1. For

the 20th Century Fox movie Black Knight, the movie group includes movies of similar genre

like Down To Earth and Snow Dogs. We combine the movie-group information with the

review information from MetaCritic and Rotten Tomatoes. We th us form movie-me dia groups

consisting of reviews in a given media outlet of an y of the 11 movies in the mov ie group. The

ﬁrst movie-media group in Table 1 consists of reviews by the New York Post of Black Knight

and its 10 matches. The diﬀerence within this group between the review of Black Knight and

the review of the matc hing movies contributes to identify the eﬀect of conﬂict of interest. The

next movie-media group consists of reviews by Entertainment Weekly magazine of the same 11

movies. These reviews by a ‘control’ media outlet cont ribute to identify the average diﬀerential

quality of a 20th Century Fox mo vie. In the speciﬁcations w e include movie-media group ﬁxed

eﬀects, thus making comparisons within a movie group for a particular media outlet.

Note two features of the procedure. First, each media typically reviews only a subsample

of the 11 movies and thus a movie-media group can consist of fewer than 11 observations.

Second, a movie can be a match to multiple 20th Century Fox or Warner Bros. movies and as

such will appear in the data set multiple times. In Table 1, this is the case for 102 Dalmatians

which is a match for both Black Knight and Scooby-Doo. In the empirical speciﬁcations, we

address this repetition b y clustering the standard errors at the movie level.

The initial sample for the test of conﬂict of interest in the News Corp. conglomerate includes

all mo vie-media groups co vering movies distributed by 20th Century Fox and all media outlets

in the sample. We then drop matching movies which were not reviewed by at least one News

Corp. media outlet. A movie group has to fulﬁll two conditions to remain in the ﬁnal sample:

(i)therehastobeatleastonereviewwithconﬂict of interest (e.g., one review of the 20th

Century Fox movie by an outlet owned by News Corp.) and (ii) the movie group has to contain

at least one movie match (which was reviewed by a News Corp. outlet).

Appendix Table 1, Panel A reports summary statistics on the sample for the News Corp.

conglomerate (top panel) and for the Time Warner conglomerate (bottom panel). The data

set covers reviews from 335 diﬀerent media outlets. Appendix Table 1, Panel B presents

information on the studios belonging to News Corp. and to Time Warner.

3BiasinMovieReviews

3.1 Validation of Matching Procedure

In the analysis of potential bias due to conﬂict of interest, we compare the reviews of movies

by 20th Century Fox (and by Warner Bros.) to the reviews of the ten most comparable movies

distributed by other studios, according to the matching procedure discussed above. This

procedure is designed to address the concern that movies distributed by, sa y, 20th Cen tury

Fox may hav e special features that not all movies have.

Does the matching procedure work? Figures 1a-f document the similarities between the

movies distributed by 20th Century Fox and Warner Bros. and the ten matching movies by

other studios with respect to three characteristics: movie genre, MPAA rating, and number

of theaters at opening. Since the match procedure did not use any of these features, we can

interpret the similarit y, or lack thereof, as a validation of the approach. Furthermore, we

choose features that are pre-assigned and thus cannot be aﬀected by the reviews themselves.

Figure 1a shows that 20th Century Fox mo vies are most likely to be action, comedy, and

drama, and unlikely to be documentaries. The matc hing movies displa y a similar pattern,

while the movies that are never a match to a 20th Century Fox movie are more likely to be

documentaries. Figure 1b displays parallel ﬁndings for the Warner Bros. movies. Turning to

the MPAA rating (Figures 1c-d), the 20th Century Fox and Warner Bros. movies and their

respective matches are very similar, while the non-matching movies are more likely to be rated

R and less likely to be rated PG-13. Figures 1e-f display a third feature, the number of theaters

on the opening weekend, sorted in quintiles in 5-year bins. The 20th Century Fox and Warner

Bros. movies are close to their matches, while the non-matching movies follow an opposite

pattern.

In the online appendix we provide further evidence on the quality of the match. We

document that the same match pattern holds for a micro level comparison, namely match ing

movies are disproportionately likely to share the genre, rating, and theaters at opening of the

movie they are matched to, compared to the general population of movies (Online Appendix

Figure 1a-f). We also document that matching movies resemble the movies at conﬂict of

interest with respect to two key outcomes of interest, the average 0-100 review score (Online

Appendix Figure 2a-d) and the probability of review (Online Appendix Figures 3a-d). Hence,

the matching procedure appears to be successful in identifying broadly comparable movies.

Building on this evidence, we now turn to the comparison outlined above.

3.2 Overall Bias

Graphical Evidence. We examine whether conﬂict of interest induces a bias on average,

starting with the 20th Century Fox movies. The bars on the right of Figure 2a indicate the

average review score for media not o w ned by News Corp. (the ‘placebo’ group). In this group,

the average review score for the 20th Century Fox movies (dark blue bar) and for the matching

movies distributed b y other studios (light blue bar) is indistinguishable. The matching movies

appear to provide a good counterfactual: in the absence of conﬂict of interest, their average

score is essentially identical to the one of the 20th Century Fox movies.

The left bars in Figure 2a present the average score for reviews in News Corp. media outlets,

like the Wall Street Journal. The score for the matching movies (light blue bar) is somewhat

lowe r than in the non-News Corp. media, indicating that the News Corp. media outlets are

on average somewhat harsher in their reviews. The key question is whether this pattern is

the same for the mo v ies distributed by 20th Century Fox, or whether those movies receive a

more generous treatment. The leftmost bar provides no evidence of bias: the average score

for the 20th Century Fox movies is essentially identical to the one for the matching movies by

other studios, with tight conﬁdence intervals. A diﬀerence-in-diﬀerence estimator indicates a

diﬀerence of -0.12 points (out of 100, with p-value of .908 of the test of equality to zero).

Figure 2b presents the evidence for Time Warner. Once again, the reviews in non-Time

Warner media outlets are scored in about the same way for Warner Bros. movies and for

matching movies (right panel). Turning to the reviews in the Time Warner outlets (left panel),

the score is also essentially the same for the Warner Bros. movies and for the matching movies.

In this second conglomerate we also ﬁnd no evidence of bias due to conﬂict of interest.

Regressions. To present a formal test and to examine the impact of control variables, we

estimate a regression-based speciﬁcation. For the 20th Century Fox movies we estimate the

diﬀerence-in-diﬀerence OLS regression:





=  + 









+ 









+ 















+ 

()

+ 



(1)

Each observation is a review for movie  by outlet . The dependent variable 



is a 0 to

100 score, or an indicator for ‘freshness’. The coeﬃcient 



captures the average diﬀerence

in reviews for movies distributed by 20th Century Fox, compared to the matching movies

distributed b y other studios. The coeﬃcient 



captures the average diﬀerence in reviews for

outlets owned by News Corp. at the time of the movie release, compared to the other outlets.

The key coeﬃcient, 



, indicates the estimated impact of the conﬂict of interest, that is, the

av erage rating diﬀerence for a 20th Century Fox movie when reviewed by a media owned by

a News Corp. outlet, compared to the counterfactual. We include a ﬁxed eﬀect  for each

movie-media group, where we denote with  () the 11 movies in the group for movie 

The standard errors are clustered at the movie level to allow for correlation of errors across

multiple reviews of a movie.

We run a parallel speciﬁcation for the Time Warner group.

Panel A of Table 2 reports the results for the score variable. Considering ﬁrst the 20th

Century Fox mo vies (Columns 1-2), we present the results ﬁrst with no controls and then with

ﬁxed eﬀects for each movie-media group (see Table 1). In this second speciﬁcation, regression

(1) is a matching estimator, comparing reviews for a 20th Century Fox movie to reviews for

the 10 matched movies. The result is similar in the two speciﬁcations, so we mostly discuss the

estimates with ﬁxed eﬀects (Column 2). The estimated coeﬃcient on 20th Century Fox movies

in Column 2,





= −075, is close to zero indicating, consistent with Figure 2 a, that the

20th Century Fox movies and the matched movies are comparable in quality. The estimated

coeﬃcient on the News Corp. outlets in Column 1,





= −434 is negative, again consistent

InTable3weshowthatalternativeformsofclustering lead to comparable or lower standard errors.

with Figure 2a.

The key coeﬃcient, ˆ



= −019 indicates a null eﬀect of the conﬂict of

interest for News Corp. outlets: 20th Century Fox movies receive slightly less positiv e reviews

by News Corp. outlets by 0.2 points out of 100. The small standard errors imply that we can

reject at the 95% conﬁdence level an eﬀect of bias of 1.92 points out of 100, equivalent to an

increase of one star (on a zero-to-four scale) for one out of 13 movies reviewed.

In Columns 3 and 4 we estimate the impact of conﬂict of interest on the Warner Bros.

movies. The results are parallel to the ones for the News Corp. conglomerate: w e ﬁnd no

evidence of an impact of conﬂict of interest: ˆ



= −002. Given the larger sample of Warner

Bros. movies, we can reject even smaller eﬀects, corresponding to 1.72 points out of 100,

equivalent to an increase of one star (out of 4) for one out of 14.5 reviews.

In Panel B of Table 2 we present parallel speciﬁcations with the ‘freshness’ indicator as

dependent variable. The results for the ‘freshness’ variable are parallel to the results for the

score variable: we ﬁnd no evidence of bias for either of the t wo conglomerates. For the rest

of the paper we focus on the 0-100 score variable given the higher statistical power given by a

contin uous variable, the results are parallel with the freshness indicator.

Robustness. In Table 3 we present alternative speciﬁcations of the benchmark results

(Columns 2 and 4 of Table 2), reporting only the conﬂict-of-interest coeﬃcient. We examine

alternatives for: (i) standard errors, (ii) additional controls, (iii) the underlying data source,

(iv) the matching procedure. Clustering the standard errors b y studio and by media outlet lead

to lower standard errors (Columns 2 and 3, compared to the bench mark clustering reproduced

in Column 1). Adding movie ﬁxed eﬀects has a small impact on the estimates (Column 4).

Estimating the eﬀect separately for the Metacritic database (Column 5) and in the Rotten

Tomatoes database (Column 6) yields similar results. (Movie reviews which are in both data

sets are present in both samples).

We also investigate the robustness of the matching procedure. Restricting the match to

only the best 3 mo vie matches (rather than 10) does not change the estimate appreciably

but, predictably, lowers the precision somewhat (Column 7). Changing the closeness measure

to maximizing the correlation in reviews yields similar results (Column 8). Not using any

observable variable (ye ar of release and number of reviews) in the match procedure also has

little impact (Column 9). In Online Appendix Table 2 we show that the results are robust to

computing matches using only one of the user reviews data sets, and using as a criterion for

closeness a lik elihood ratio measure of the probability of rating a movie.

Cross-Sectional Estimates. In Table 4 we compare the matching estimates to cross-

sectional estimates for speciﬁcation (1) using all reviews in the MetaCritic and Rotten Toma-

toes data, making no use of the matching procedure. The estimates of the eﬀect of conﬂict

of interest for Time Warner are negative but close to zero, similar to the main speciﬁcation

(Table 2). Instead, the estimate of the conﬂict of interest eﬀect for the News Corp. outlets

In Column 2 this coeﬃcien t is identiﬁed oﬀ of media outlets that change ownership within our sample.

indicate a statistically signiﬁcan t, if quite small, 2.04 points of bias in the speciﬁcation with

movie and media ﬁxed eﬀects (Column 2).

The next columns reconcile this estimate with the

matc hing estimate. The diﬀerence is not due to excluding movies that are not reviewed by

outlets at conﬂict of interest (Column 3) or to the fact that some of the 20th Century Fox and

Warner Bros. mo v ies are not present in the Flixster/Netﬂix/MovieLens data set (Column 4),

and thus dropped from our matching analysis. Instead, the main diﬀerence is the inclusion of

control movies that are not matches to a 20th Century Fox or Warner Bros. movie. When we

drop these movies (Column 5), the estimate is very similar to the benchmark estimate.

Given the evidence that non-matching mo vies diﬀer from matching movies in several di-

mensions like genre, rating, and opening weekend (Figures 1a-f), this suggests that reviewers

in the News Corp. outlets have speciﬁc tastes that it is important to control for in the analysis.

To more directly document this, we show that mo vies that are distributed by other studios

but are matches to 20th Century Fox movies are indeed reviewed more positively by the News

Corp. outlets (Online Appendix Table 3). We ﬁnd similar evidence for correlated tastes when

considering movies similar in genre to the Fox movies, though not for similarity of MPAA

rating and budget. Giv en the evidence of the confound due to correlated tastes, we use the

matching estimator in the rest of the paper.

3.3 Bias b y Movie Quality and Type

So far, w e presented evidence on bias for the average movie. Yet, bias should be larger for

movies with a higher return to bias, holding constant the reputational cost. While we do not

hav e direct measures of return to bias, we consider two dimensions which are likely to correlate

with it. We expect that movies with generally higher review scores are likely to have higher

return to bias, as an extra star is likely to matter more if it is the 4th star out of 4, as compared

to the ﬁrst star. We also assume that high-proﬁle movies are likely to have higher returns given

the larger poten tial audience (holding constant the persuasive impact of a review).

Bias by Movie Quality. In Figure 3a we presen t evidence on potential bias as function

of movie quality for the 20th Century Fox movies. We assign to each movie the average review

score computed excluding the reviews in media at potential conﬂict of interest. We then display

a polynomial plot of the review score in the News Corp.-owned media outlets for the movies

distributed by 20th Century Fox (dark blue line) and for the matc hing movies distributed by

A previous working paper version of this paper estimated a bias of similar magnitudes using this speciﬁcation.

Notice that this last speciﬁcation still diﬀers from the matching one because (i) the set of ﬁxed eﬀect diﬀers

and(ii)inthebenchmarkspeciﬁcation reviews for a matching movie appear multiple times if the movie is a

match to multiple, say, 20th Century Fox movies; instead, in the cross-sectional speciﬁcation each movie review

appears only once. Column (5) shows that this diﬀerence is immaterial to the results. Furthermore, the estimate

in Column (5) is similar to the estimate in Column (4) in Table 3 where movie ﬁxed eﬀects are included as

additional controls.

other studios (light blue line).

The plot for the matching movies indicates that the News

Corp. outlets largely follow the other outlets in their review. The plot for the movies at conﬂict

of interest hovers around the one for the matching movies, with no evidence of deviation for

movies of higher, or lower, quality. For Time Warner as well (Figure 3b), the average score for

aﬃliated movie s tracks v ery closely the score for the non-aﬃliated movies, with no systematic

deviation for higher-qualit y movies. Thus we do not ﬁnd evidence of diﬀerential bias.

Bias by Movie Proﬁle. In addition to the mainstream studios 20th Century Fox and

Warner Bros., the News Corp. and Time Warner conglomerates include indie studios like Fox

Searc hlight, Fine Line, and New Line (see Appendix Table 1B). Figures 4a and 4b plot, for

each studio, the average review score in media outlets at conﬂict of interest (y axis) and in

other outlets (x axis). To make the comparison clear, we plot the same measure for the other 9

major studios.

There is no evidence of diﬀerential bias, whic h consists of points lying above

the 45 degree line, for the mainstream studio compared to the indie studios.

In Online Appendix Table 4, we present additional evidence. We re-estimate speciﬁcation

(1) allowing for a diﬀeren tial eﬀect of conﬂict of interest for four proxies of return to bias: (i)

distribution by a mainstream studio, (ii) production budget, (iii) number of theaters at opening

and (iv) domestic box oﬃce.

We ﬁnd no statistically signiﬁcant evidence of diﬀerential bias

by the four proxies, even though directionally the sign of the eﬀects is as expected for the 20th

Century Fox movies. Overall, there is no clear evidence of diﬀerential bias for mo v ies with

plausibly higher return to bias.

3.4 Bias by Media and Journalist

The previous evidence indicates the apparent lack of bias due to conﬂict of interest, even when

considering separately movies with plausibly higher incentives for bias. These results reject

the scenario of widespread bias across all outlets within a conglomerate. Still, it is possible

that some media outlets, or some journalists, bias their reviews, but this is balanced by the

lack of bias in other outlets in the same conglomerate, or perhaps even by negative bias (to

avoid criticism). We thus examine the occurrence of bias by media and by journalist.

Bias By Media. The scatter plot in Figure 5a reports for each media outlet the average

review for the 20th Century Fox movies and the average review for the matching movies by

other studios. To provide a counterfactual, we also plot these measures for the 200 largest

media outlets not owned by News Corp.

No News Corp. media outlet deviates substantially

We use an Epanechnikov kernel and a 1st degree polynomial, with a kernel of 5 rating points. We truncate

movies with average movie score below 30 or above 80, since such movies are rare.

Dot Sizes are proportional to the square root of the number of reviews by News Corp. or Time Warner

outlets. We do not use the matching procedure in order to ensure a larger sample of movies b y other studios.

Forthelastthreeproxies,weusedeciles,formedwithin5-yearperiods,ofthevariabletoadjustforchanges

over time and skewness.

We only include outlets with at least 15 reviews of 20th Century Fox movies while owned by News Corp.

on the positive side of the trend line, the indication for bias.

We estimate a speciﬁcation like

(1) for each outlet at conﬂict of interest separately, and ﬁnd no signiﬁcant evidence of bias for

any of the outlets (Online Appendix Table 5).

Figure 5b prov ides parallel evidence for the Time Warner conglomerate, with Entertainment

Weekly, Time magazine and Cinematical right on the regression line indicating no bias, a

ﬁnding that is replicated in regression format (Online Appendix Table 5). Thus, the pattern

for the individual outlets is similar to the ov erall pattern of no bias.

Bias By Journalist. We further take advantage of the fact that most media have only

a small number of movie reviewers, and these journalists typically stay on for years, if not

decades. This long tenure allows us to estimate journalist-speciﬁc patterns which, as far as

we know, is a rare feature within the media economics literature (Dougal et al., 2012). In

Appendix Figures 1a-b w e provide parallel plots to Figures 5a-b, but by journalist. In addition

to the journalists working in the two conglomerates, we include the 500 other journalists with

the most reviews. Only one journalist stands out, Maitland McDonagh (at TV Guide), with a

statistically signiﬁcant estimate of bias (Online Appendix Table 6). Yet, given that the pattern

appears for only one out of 12 journalists, it is plausible that this pattern is due to chance.

3.5 Editorial Bias

In the previous section we tested for bias in the presence of conﬂict of interest, focusing

on the role o f journalists. Conversely, we now examine the role of editors. An editor who

intends to bias the review process can do so in at least two ways: by putting pressure on the

journalists, or by assigning the aﬃliated movies to journalists who on average assign higher

review scores.

We examine the latter mechanism , whic h is w ell-suited to test for biased

coverage as a managerial policy of the conglomerate. While journalists could resist managerial

pressure to bias the content of their reviews, this form of bias only requires the assignment of

movies to diﬀerent reviewers.

We provide graphical evidence on this test for the reviewers in News Corp. media outlets

in Figure 6a. We plot for each reviewer the average generosity in review score (relative to

the media outlet a verage) (x axis) and the share of their reviews of 20th Centu ry Fox movies

(y axis).

As the scatter shows, movie reviewers diﬀer sizably in generosity within a given

outlet. Yet, there is no evidence that the more generous reviewers are more lik ely to review

The Sunday Times and Wall Str eet Journal are outliers belo w the line, but the estimate of bias is imprecise

for these outlets given the small n umber of reviews at conﬂict of interest .

A third form of editorial inﬂuence is the hiring of more favorable journalists and ﬁring of less favorable ones.

We observe no evidence of elevated turn-over for the outlets after a change in o wnership.

To compute the average generosity, we only take into account score reviews (on a 0-100 scale) and generate for

eac h review an idiosyncratic review score deﬁned as the score minus the average review score of the corresponding

movie. We then compute the a verage of this variable for all journalists and their aﬃliated outlets. The measure

of the ave rage generosity of a journalist (relative to the aﬃliated outlet) is calculated as the diﬀerence between

the t wo means. Here, we do not use the matching procedure in order to preserve a larger sample of movies.

20th Century Fox movies. Indeed, the regression line points to a slight negative relationship

between generosit y and review probability.

In Figure 6b we report the parallel evidence for the Time Warner outlets. As for the News

Corp. outlets, we ﬁnd no evidence of a systematic pattern of assignment of movies to reviewers

in order to beneﬁttheaﬃliated studio.

3.6 Bias by Omission

The previous evidence rules out sizable bias in the movie quality assessed in reviews, whether

due to editorial or journalistic decisions. But this evidence does not cast light on a potentially

more insidious form of bias: bias by om ission. The media can selectively display items of

information, as in Anderson and McLaren (2012). In our setting, an outlet may decide to not

review a below-average movie by an aﬃliated studio, but make sure to review an above-average

movie by the same studio. A media outlet following this strategy would not display any bias

conditional on review; hence, bias by omission would not be detected by the previous analysis.

In Figure 7a we present evidence on omission bias for the News Corp. media. We test

whether Ne ws Corp. outlets are more likely to review 20th Century Fox movies with high

predicted review (as proxied by high average rating by other reviewers), compared to their

reviewing patterns for non-20th Century Fox mo vies. We display a polynomial smoother o f

the review probability as a function of the average review score of a movie (in the range between

30 and 80).

The average probability of review by News Corp. media outlets of 20th Century

Fox movies is barely increasing in the review score (darker continuous line). By comparison,

the probability of review of the matching movies b y other studios (lighter continuous line) is

more clearly increasing in the movie review, suggesting if anything a negative bias by omission.

To strengthen the inference, we also compare these patterns to the probability of review

by other media outlets not own ed by News Corp. In doing so, we need to take into account

that media outlets diﬀer in their reviewing propensity. Thus, for each media outlet owned by

News Corp. w e choose the ten media outlets which displa y the closest pattern in the review

probability of non-20th Century Fox movies.

The dotted lines in Figure 7a display the

The sample for the omission bias test in this section is determined as follows. For each of the 8 News Corp.

outlets, like the New York Post, we determine all 20th Century Fox movies and their movie matches which

w ere released during News Corp. ownership of the respective outlet. For each movie in this subsample and

outlet-either the News Corp. or one of the control outlets (see below)-we generate a dummy of whether it wa s

reviewed (0-100 score or ‘freshness’ indicator). Thus, there is only one observation per movie and media outlet.

We use this data set when testing for omission bias for that particular outlet. To obtain the sample for the

overalltestpoolingacrossall8outlets,werepeatthisprocedureforall8NewsCorp. outletsandappendthe

data sets. We follow a parallel procedure for the Time Warner test.

The matching outlets are the ten outlets with the smallest distance in the probability of review for the

matching movies. We form bins with a width of 5 points of the average review score and determine the average

distance between two media outlets in the review probabilities within each bin. The overall distance is computed

a veraging the distance across the bins, weighting by the number of mo vies in a bin.

probability of review by these matched media of 20th Century Fox movies (dotted darker line)

and of the matching movies (dotted lighter line). The dotted lines track remarkably we ll the

contin uous lines for the matching movies, suggesting that the matching media provide a good

counterfactual to the News Corp. media. Overall, Figure 7a suggests no evidence of omission

bias. Online Appendix Figures 4a-d show that the same pattern holds when considering the

News Corp. media outlets individually.

The corresponding ﬁgure for the Time Warner outlets (Figure 7b) instead provides some

evidence consistent with omission bias. The probability of review of Warner Bros. movies

in Time Warner outlets is increasing in the measured quality of the movie, more so than in

the matched media. Yet, this increasing pattern is similar for matching movies in the Time

Warner media (lighter continuous line), suggesting that the pattern may be due to a reviewing

strategy in the Time Warner media outlets, rather than to bias.

To provide more evidence, in Online Appendix Figures 5a-d we disaggregate the eﬀect

by the four Time Warner media outlets. The evidence suggestive of omission bias is almost

entirely due to Time magazine. To ascertain whether the pattern in the data is due to intended

omission bias or an idiosyncratic reviewing strategy by Time, w e exploit two placebos. First,

we take advantage of the fact that in years 1985-89 Time magazine was not yet part of the

Time Warner conglomerate. Second, we exploit the fact that 20th Century Fo x movies share

some characteristics with Warner Bros. movies (see Figures 1a-f), but there is no conﬂict of

interest in place with those movies at Time magazine. As Online Appendix Figures 6b and 6c

show, these two placebos show a similar reviewing pattern to the one in the main sample. This

suggests that the pattern at Time magazine should not be interpreted as bias b y omission.

To further put these ﬁndings in context, we compare the extent of selective reviewing in

the media at conﬂict of interest with the same phenomena for the largest 200 other outlets.

Figures 8a-b display for each media outlet the estimated sensitivity of the review probability

to the average score for the movies at conﬂict of interest (y axis) versus the same movies in

the matching outlet (x axis). The two sensitivity coeﬃcien ts are just the slope coeﬃcient

of separate linear regressions of the review probability on the a verage review score. Bias by

omission would manifest itself as an outlier above the regression line: an outlet is more sensitive

to quality when reviewing a movie at conﬂict of interest. The patterns conﬁrm the ﬁndings

above. None of the News Corp. outlets stand out for omission bias, while among the Time

Warner outlets, only Time magazine stands out, a case we discussed above.

To provide a statistical test of omission bias, we estimate a linear probability model in

Table 5, which we illustrate for the case of media owned by News Corp.:





=  + 















+ Γ















¯



+ 









+(2)











¯



+ 









+ 









¯



+ ¯



+ 

()

+ 





An observation is a possible review of a 20th Century Fox movie or of a matching movie by

one of the News Corp. or matching outlets with similar probabilit y of review. The dependent

variable is the indicator 



which equals 1 if media outlet  reviews movie  The key

coeﬃcient is Γ



on the interaction of the conﬂict of interest variable with the mean rating

score ¯



.Thiscoeﬃcient indicates how the probability of a review varies with the average

review score, in the presence versus absence of a conﬂict of interest. The regression includes

movie-media group ﬁxed eﬀects. A key assumption made in equation (2) is that the probability

of movie review is linearly increasing in the average movie score; we adopt this assumption

given the evidence of approximate linearity in Figures 7a-b.

Table 5 provides no evidence of selective review consistent with omission bias for the News

Corp. or for the Warner Bros. media. For News Corp. outlets, we can reject that a one-

standard deviation increase in movie quality (14 points in overall score) for a 20th Century

Fox movie increases the probability of review (diﬀerentially) by more than 1.7 percentage

poin ts. Similarly, for Time Warner we can reject for a similar increase in movie quality an

increase in review probability of more than 2.2 percentage points. In Online Appendix Table

7 we present the results separately for each media outlet. The relevant coeﬃcien t 



on the

interaction between conﬂict of interest and average review score is signiﬁcan tly positiv e only

for Time Magazine, a special case we discussed abo ve. Overall, we conclude that it is unlikely

that any of the outlets is explicitly adopting a strategy of bias by omission.

3.7 Bias in Movie Aggregator

Sofarwehavefocusedontheconﬂict of interest induced by the consolidation of studios like

20th Century Fox and Warner Bros. into media conglomerates which employ movie reviewers.

But consolidation aﬀects the review aggregators themselves. Rotten Tomatoes, independent

when launched in 1998, was acquired by IGN Entertainment in June 2004, and IGN itself was

purchased b y News Corp. in September 2005. IGN, and hence Rotten Tomatoes, was then

sold in Janu ary of 2010 by News Corp. and acquired in April 2011 by Time Warner.

This ownership structure generates an incentive for Rotten Tomatoes to post more positive

reviews of 20th Century Fox movies during the period of News Corp. ownership (2006-2009).

Since the reviews are posted quickly on the Rotten Tomatoes site and then rarely updated

weusetheyearofreleaseofthemovietotestthehypothesisofconﬂict of interest. We estimate





=  + 











2006−09



+ 









+ 



+ 



+ 



 (3)

where 



is the ‘freshness’ indicator on Rotten Tomatoes for movie  in media outlet .The

coeﬃcient of interest, 



 captures how movies distributed by the 20th Century Fox studio

We also examined a form of partial omission: whether media at conﬂict of interest are more likely to

display delayed reviews and shorter reviews for low -quality aﬃliated movies. Using a smaller data set (since

the information on date of review and length of review is not in Metacritic or Rotten Tomatoes) we do not ﬁnd

evidence of such bias.

Consistent with this, tw o separate scrapes of the site at 3 month distance yielded no change in the reviews

for older mo vies.

(





= 1) are characterized in years 2006-2009, compared with the years before and

after. We allow for a baseline diﬀerence in reviews for 20th Century Fo x movies (captured by





)andﬁxed eﬀects for year  and for the movie-media group. Most importantly, we cont rol

for the MetaCritic scoring 



for the same movie review

Column 1 in Table 6 shows that the eﬀect of conﬂict of interest is a precisely estimated zero

(ˆ



=00031), a result that replicates when using all reviews, rather than just the matc hed

sample (Column 2). We can reject as an upper bound that conﬂict of interest increases the

probability of a fresh score by 0.6 percentage points (Column 2), a small eﬀect. In Figure 9a,

using the matched sample, we present graphical evidence using a local polynomial estimator

of the Rotten Tomatoes ‘freshness’ indicator on the 0-100 quantitative score. We run the

non-parametric regressions separately for the 20th Century Fox movies (the continuous lines)

and the matching movies by other studios (dotted lines), split by the period of News Corp.

ownership (dark blue line) and the remaining time period (ligh t blue line). The two continuous

as well as the two dotted lines are very close on the graph, again indicating no bias.

While we detect no bias on average, bias may have been present in some years, for example

when News Corp. just acquired Rotten Tomatoes and awareness of the conﬂict of interest was

presumably lo wer. We estimate an event study speciﬁcation:





=  + 















+ 





(1 − 





)



+ 



+ 





The speciﬁcation is parallel to (3) except that, instead of separating the years into a period of

ownership (2006-09) and all else, w e interact the year ﬁxed eﬀects 



with an indicator for 20th

Century Fox movie and an indicator for the complement. Figure 9b shows that the residual

freshness score for the 20th Century Fox movies (





) trac ks the series for other movies (





)

also during the years of ownership, providing no evidence of bias. Since bias may still be

present in a subset of the data, we analyze separately reviews with a quantitative score (i.e.

stars) and qualitative reviews for which the freshness score is determined by a staﬀ reading.

For the quantitative reviews, we focus on reviews with scores between 50 and 70, for which

Rotten Tomatoes appears to use qualitative information to assign the ‘freshness’ rating. Even

in this sample (Column 3), we detect no bias.

However, bias should be most likely for reviews without a quantitative score since the

probability of detection is particularly low. Yet, w e ﬁnd no evidence of bias in this sample either

(Column 4). We replicate this result on the smaller sample of qualitative reviews stored in both

aggregators, so as to include as a control the score attributed by the Metacritic staﬀ (Column

5), again ﬁnding no eﬀect of the conﬂict of interest on bias, with more precise estimates.

Despite the conﬂict of interest, there is no semblance of bias in the movie aggregator Rotten

Tomatoes, even for the types of reviews for whic h detection of bias would be hardest.

The quantitative scoring is as reported b y Rotten Tomatoes, translated into the 0-100 score. If the Rotten

Tomatoes score is missing, for example for qualitative reviews, we use the score in MetaCritic if a vailable. We

conﬁrm that Rotten Tomatoes does not bias this quantitative score by regressing it on the corresponding score

for the same review in MetaCritic, when both are available.

4Conclusion

Consolidation in the media industry is considered by man y a condition for survival in an

industry hit hard by the loss of advertising. Yet, consolidation does not come without potential

costs. In addition to the potential loss of diversity (George and Waldfogel, 2003), consolidation

increases the incidence of conﬂict of interest due to cross-holdings, and possible ensuing bias.

We focus on conﬂict of interest for movie reviews, such as when the Wall Stre et Journal

reviews a 20th Century Fox movie. The holding company, News Corp., can potentially beneﬁt

ﬁnancially from a more positive review, and hence higher movie attendance, creating a conﬂict

of interest.

Using a data set of over half a million movie reviews from 1985 to 2010, we ﬁnd no statistical

evidence of media bias due to conﬂict of interest in either the News Corp. conglomerate or

the Time Warner conglomerate. The null ﬁnding is not due to imprecision. We can reject

small estimates of bias, such as one extra star (out of 4) in one out of every 13 movies at

conﬂict of interest. Moreover, w e are able to examine bias at a high level of detail, including

bias by media outlet and journalist, comparative statics in the return to bias, bias in editorial

assignment, bias by omission as opposed to commission, and bias b y the aggregator. In the

end, we feel conﬁdent of the overall estimate.

As we discussed in the introduction, these estimates imply a back-of-the-envelope bound for

the value of a reputation for the media outlets. Under the assumption that an extra star (our

of 4) conv inces 1 percent of readers to watc h a movie, an extra star in a single movie review

for a 20th Century Fox in the New York Post would add approximately $40,000 in proﬁts for

News Corp. The correspondent ﬁgure for an extra star in a review of a Warner Bros. movie

in Time magazine is appro ximately $300,000. The fact that such systematic bias did not take

place in either outlet indicates that the value of their respective reputations is larger.

Within the context of movie reviews, we addressed questions that have arisen in the eco-

nomics of the media, such as whether bias occurs by omission or commission, about which we

hav e limited information. We view this contribution as a step forward in better understanding

the functioning of media outlets, which play a key role in the formation of public opinion.

In particular, this paper pro v ides some evidence on how media outlets navigate the trade-

oﬀ between professional journalism and revenue maximization for the owners. In this case,

professionalism and reputation concerns appear to trump possible short-term revenue gains.

As w e discussed above, our results diﬀer from the impact of conﬂict of interests for advertis-

ing. In the latter case, both Reuter and Zitzewitz (2006) and Di Tella and Franceschelli (2011)

document sizable distortions. We conjecture that two diﬀerences likely play a role. First, the

beneﬁts from the ads directly beneﬁt the media outlet, while in the case at hand the beneﬁts

accrue to the shareholders of the conglomerate, a more indirect link. Second, movie reviews

are easy to compare across a number of outlets while coverage of scandals and of mutual fund

performance are less so. More research will hopefully clarify these diﬀerences.

References

[1] Anderson, Simon and John McLaren. 2012. “Media Mergers and Media Bias with Rational

Consumers.” Journal of the European Economic Association, 10(4), 831-859.

[2] Brown, Alexander L., Camerer, Colin F. and Lovallo, Dan. 2012. “To Review or Not

to Review? Limited Strategic Thinking at the Movie Box Oﬃce”, American Economic

Journal: Microeconomics, 4(2), 1-26.

[3] Camara, Fanny and Nicolas Dupuis. 2014. “Structural Estimation of Expert Strategic

Bias: the Case of Movie Reviewers” Working paper.

[4] Chipty, Tasneem. 2001. “Vertical Integration, Market Foreclosure, and Consumer Welfare

in the Cable Television Industry”, The American Economic Review, 91 (3), 428-453.

[5] DellaVigna Stefano and Matthew Gentzkow. 2010. “Persuasion: Empirical Evidence.”

Annual Review of Economics, 2, 643-669.

[6] DellaVigna Stefano, Ethan Kaplan. 2007. “The Fox News eﬀect: Media Bias and Voting.”

Quarterly Journal of Economics, 122(3), 1187-234.

[7] Di Tella, Rafael, and Ignacio Franceschelli. 2011. “Government Advertising and Media

Cov erage of Corruption Scandals.” American Economic Journal: Applied Economics 3(4),

119-151.

[8] Dobrescu, Loretti, Michael Luca, and Alberto Motta. 2013. “What makes a critic tick?

connected authors and the determinan ts of book reviews.” Journal of Economic Behavior

and Organization, 96, 85-103.

[9] Dougal, Casey, Joseph Engelberg, Diego Garcia, and Christopher Parsons. 2012. “Jour-

nalists and the Stock Mark et” Review of Financial Studies, 25(3), 639-679.

[10] Dranove, David and Ginger Zhe Jin. 2010. “Quality Disclosure and Certiﬁcation: Theory

and Practice,” Journal of Economic Literature, 48(4), 935-63.

[11] Durante, Ruben and Brian Knight. 2012. “P artisan Con trol, Media Bias, and Viewer

Responses: Evidence from Berlusconi’s Italy.”Journal of the European Economic Associ-

ation, 10(3), 451-481.

[12] Ellman, Matthew and Fabrizio Germano. 2009. “What do the Papers Sell? A Model of

Advertising and Media Bias”, Economic Journal, 119(537), 680-704.

[13] Enikolopov, Ruben, Maria Petrova, and Ekaterina V. Zhuravskay a. 2011. “Media and

Political P ersuasion: Evidence from Russia.”American Economic Review, 101(7), 325 3-

3285.

[14] Gentzk ow, Matthew and Jesse Shapiro. 2006. “Media Bias and Reputation ”Journal of

Political Economy, 114(2), 280-316.

[15] Gentzkow, Matthew and Jesse Shapiro. 2010. “What Drives Media Slan t? Evidence from

U.S. Daily Newspapers ”Econometrica, 78(1), 35-71.

[16] George, Lisa and Joel Waldfogel. 2003. “Who Aﬀects Whom in Daily Newspaper Mar-

kets?” Journal of Political Economy, 111(4), 765-784.

[17] George, Lisa and Joel Waldfogel. 2006. “The New York Times and the Market for Local

Newspapers” American Economic Review, Vol. 96(1), 435-447.

[18] Gilens, Martin and Craig Hertzman. 2000. “Corporate Ownership and News Bias: News-

paper Coverage of the 1996 Telecommunications Act ” Journal of Politics 62(2), 369-86.

[19] Gooslbee, Austan. 2007. Vertical Integration and the Market for Broadcast and Cable

Television Programming. Working paper.

[20] Groseclose, Tim and Jeﬀrey Milyo. 2005. “A Measure of Media Bias”. Quarterly Journal

of Economics, 120(4), 1191-1237.

[21] Hamilton, James T. 2003. All the News That’s Fit to Sell: How the Market Transforms

Information into News. Princeton: Princeton University Press.

[22] Knight, Brian and Chun-Fang Chiang. 2011. “Media Bias and Inﬂuence: Evidence from

Newspaper Endorsements.” Review of Economic Studies, 78(3), 795-820.

[23] Larcinese, Valentino, Ricardo Puglisi, and James M. Snyder. 2011. “Partisan Bias in

Economic News: Evidence on the Agenda-Setting Beha vior of U.S. Newspapers.”Journal

of Public Economics, 95(9-10): 1178-1189.

[24] Milgrom, Paul and John Roberts. 1986. “Relying on the Information of Interested Parties”,

RAND Journal of Economics, 17(1), 18-32.

[25] Mullainathan, Sendhil, Joshua Schwartzstein, and Andrei Shleifer. 2008. “Coarse Thinking

and Persuasion.” Quarterly Journal of Economics, 123(2), 577-619

[26] Reinstein, David and Christopher Snyder. 2005. “The Inﬂuence of Expert Reviews on

Consumer Demand for Experience Goods: A Case Study of Movie Critics” Journal of

Industrial Economics, 53(1), 27-51.

[27] Ravid, S. Abraham, John Wald, and Suman Basuroy. 2006. “Distributors and ﬁlm critics:

does it take two to Tango?” Journal of Cultural Economics, 30(3), 201-218.

[28] Reuter, Jonathan and Eric Zitzewitz. 2006. “Do Ads Inﬂuence Editors? Advertising and

Bias in the Financial Media.”The Quarterly Journal of Economics, 121(1): 197-227.

[29] Rossman, Gabriel. 2011. The Inﬂuence of Ownership on the Valence of Media Content:

The Case of Movie Reviews. Working Paper #27, Fall 2011.

[30] Str¨omberg, David. 2004. “Radio’s Impact on Public Spending.”Quarterly Journal of Eco-

nomics, 119(1): 189-221.

Figures 1a-1f. Documenting the Quality of Movie Matches

Figure 1a-b. Similarity to Match: Movie Genre

Figure 1c-d. Similarity to Match: MPAA Rating

Figure 1e-f. Similarity to Match: Number of Theaters in Opening Weekend

Notes: Figures 1a-b display the distribution of movie genre for the movies by 20

Century Fox and Warner Bros., the movie matches,

and movies which are not matches. Figures 1c-d and 1e-f display parallel evidence for the distribution of MPAA ratings and the

number of theaters at opening. For the purpose of assigning movies to a particular quintile in Figures 1e-f, 5-year bins are formed and

the quintile of a particular movie is determined within each bin.

Figure 2a. Average bias in movie ratings: News Corp.-affiliated outlets

Figure 2b. Average bias in movie ratings: Time Warner-affiliated outlets

Notes: Figures 2a and 2b report the average review score on a 0 to 100 scale. Figure 2a is split by whether the movies are reviewed by

News Corp. or other outlets. Each subpanel shows two differently colored bars indicating either movies distributed by 20

Century

Fox (dark blue bar) or movie matches (light blue bar). Figure 2b displays parallel evidence for Time Warner outlets and Warner Bros.

movies.

Figure 3a-b. Bias by Quality: News Corp.-owned outlets (3a) and Time Warner-

owned outlets (3b)

Notes: Figures 3a reports local polynomial regressions with Epanechnikov kernel with bandwidth of 5 and 1

degree polynomial of

the average review score (on a 0 to 100 scale) by News Corp. outlets on the average movie review score by all other outlets. We do

separate regressions for 20

Century Fox movies (dark blue line) and the matching movies distributed by other studios (light blue

line). Figure 3b reports the same polynomial regressions for Time Warner outlets and Warner Bros. movies. The sample only contains

movies with an average review score in the range of 30 to 80.

Figure 4a-b. Bias by Studio: News Corp.-owned outlets (3a) and Time Warner-

owned outlets (3b)

Notes: Figure 4a displays the average review score (on a 0 to 100 scale) by News Corp. outlets against the average review score by

other outlets conditional on the distributing studio of the movies reviewed. Colors indicate whether a particular studio is owned by

News Corp. (red dots) or is one of the other nine biggest studios (excluding Time Warner studios) (gray dots). Dot sizes are

proportional to the square root of the number of reviews by News Corp. outlets. Studios with a number of reviews by News Corp.

outlets less than 20 are excluded. Figure 4b shows parallel evidence for Time Warner outlets and Warner Bros. movies. Both figures

use the Metacritic/Rotten Tomatoes dataset without matches.

Figure 5a-b. Bias by Media Outlet: News Corp.-owned outlets (5a) and Time

Warner-owned outlets (5b)

Notes: Figure 5a displays the average review score (on a 0 to 100 scale) of 20

Century Fox movies against the average review score

of the associated movie matches for News Corp. outlets and outlets not owned by News Corp. Colors indicate whether a particular

outlet is owned by News Corp. (red dots) or is one of the other 200 biggest control outlets (gray dots). Outlets with a number of

reviews of 20

Century Fox movies less than 15 are excluded. Figure 5b displays parallel evidence for Warner Bros. movies and Time

Warner outlets.

Figure 6a-b. Editorial Bias in Assignment of Reviews

Notes: Figure 6a displays the share of score reviews (on a 0 to 100 scale) of 20

Century Fox movies for a given journalist employed

at a News Corp. outlet versus a measure for the journalist generosity. For each review an idiosyncratic score is calculated as the

review score minus the average review score for the corresponding movie. The mean of this variable is computed for each journalist

and outlet to calculate a measure of absolute generosity. The journalist generosity (relative to the media outlet average) is then defined

as the difference between the absolute generosity of a journalist minus the absolute generosity of the affiliated outlet. Dot sizes are

proportional to the square root of the number of reviews by a particular journalist. Journalists with a number of reviews less than 50

are excluded. Figure 6b displays the same relationship for journalists employed at a Time Warner outlet. For each journalist the

associated outlet is indicated in parentheses: BN: Beliefnet; CST: Chicago Sun-Times; CM: Cinematical; CNN: CNN; NOTW: News

of the World; EW: Entertainment Weekly; NYP: New York Post; ST: Sunday Times; Time: Time; Times: Times; TVG: TV Guide;

WSJ: Wall Street Journal.

Figure 7a-b. Selective coverage -- Probability of review by movie quality (rating)

Notes: Figure 7a reports local polynomial regressions with Epanechnikov kernel with a bandwidth of 5 and 1

degree polynomial of

an indicator for whether the movie was reviewed (score on a 0 to 100 scale or ‘freshness’ indicator) on the average movie review

score. Colors distinguish separate polynomials for either 20

Century Fox movies or associated movie matches. Solid lines

characterize regressions for outlets owned by News Corp., dashed lines those for the ten best matching media. For each News Corp.

outlet those are determined by minimizing the distance to the particular News Corp. outlet in the review probability only for the

matching movies. For this purpose, bins with a width of 5 score points are formed and the distance in the review probability for each

bin is weighted by the number of matching movies in the particular bin. We only keep movies released in the time period of News

Corp. ownership of a particular outlet. Figure 7b displays parallel evidence for Warner Bros. movies and Time Warner outlets. The

sample only contains movies with an average review score in the range of 30 to 80. For additional information on how the data is

generated see text.

Figure 8a-b. Selective coverage -- Probability of review by movie quality,

Comparison to other media

Notes: Figure 8a depicts the sensitivity of the review probability of 20

Century Fox movies to the average review score for all outlets

owned by News Corp. as well as for 200 control outlets (selected as those with the highest number of reviews). Sensitivity is measured

as the slope coefficient of a linear regression of the review probability on the average review score of a movie. The y-axis shows the

sensitivity in a particular outlet while the x-axis measures the sensitivity in the 10 outlet matches determined as for Figure 7. A

positive (negative) sensitivity measure indicates that the review probability is increasing (decreasing) in the average movie review

score. Additionally, the graph contains a linear fit. Dots above the line indicate that the likelihood of a particular outlet reviewing a

Century Fox movie increases more strongly in the average review score relative to its outlet matches. Figure 8b shows parallel

evidence for Warner Bros. movies and Time Warner outlets.

Figures 9a-9b. Conflict of Interest in Rotten Tomatoes (Newscorp 2006-09)

Notes: Figure 9a reports a local polynomial regression with Epanechnikov kernel with a bandwidth of 5 and a 1

degree polynomial of

an indicator for ‘freshness’ rating of a movie in Rotten Tomatoes on the corresponding movie review score. The sample includes the

period in which Rotten Tomatoes is owned by News Corp. (2006-09, continuous lines) and the remaining period (dotted lines), and

plots separate regressions for 20

Century Fox movies (dark blue lines) and movie matches (light blue lines). Figure 9b reports the

estimated coefficients from an event study regression of the freshness score in Rotten Tomatoes on the quantitative score and year

fixed effects interacted with an indicator for a 20

Century Fox movie (dark blue lines) and year fixed effects interacted with an

indicator for movie matches (light blue lines).

Table 1. Data set Formation, Example

Notes: Table 1 shows the construction of the main sample. For every movie distributed by a News Corp. studio (for simplicity called

Century Fox) or Time Warner studio (for simplicity called Warner Bros.) which is covered by at least one of the three datasets

Netflix, Flixster, or MovieLens the ten best movie matches are determined as those with the minimum distance in individual user

ratings. This data provides movie groups consisting of a particular 20

Century Fox or Warner Bros. movie and its 10 best matches.

The information is combined with the movie reviews provided by MetaCritic and Rotten Tomatoes. The resulting movie-media groups

contain all reviews of movies in a certain movie group by a particular outlet. The upper part of Table 1 illustrates this formation for

the example of the comedy movie “Black Knight” by 20

Century Fox. Gray color indicates that an outlet, in this case New York Post,

belongs to the same conglomerate. Darker gray color indicates that a certain review is at conflict of interest. The lower part of Table 1

shows the equivalent for the comedy movie “Scooby-Doo” by Warner Bros. Note that a certain movie can be a match to several

movies, and thus its reviews can be appear more than once in the main sample, as is the case for the comedy movie “102 Dalmatians”

by Walt Disney.

(1) (2) (3) (4)

-0.1288 -0.1898

(Measure of Confl

of In

eres

for News Corp.) [1.1166] [1.0596]

-0.4033 -0.7560

[1.0131] [0.5297]

-4.3446*** 0.3348

[0.6234] [1.2639]

-0.4368 -0.0188

(Measure of Confl

of In

eres

for T

arner) [0.9468] [0.8762]

-0.8251 -0.6220

[0.9017] [0.4413]

2.5421*** -26.2505***

[0.5382] [5.6107]

Movie-Media Group Fixed Effects

0 0.45 0 0.47

Number of reviews at conflict of interest 421 421 685 685

291124 291124 450699 450699

0.0330 0.0156

(Measure of Confl

of In

eres

for News Corp.) [0.0277] [0.0286]

-0.0005 -0.0138

[0.0215] [0.0115]

-0.0790*** 0.0304

[0.0149] [0.0520]

-0.0068 -0.0113

(Measure of Confl

of In

eres

for T

arner) [0.0216] [0.0227]

-0.0227 -0.0188**

[0.0179] [0.0093]

0.0103 -0.5340***

[0.0133] [0.1026]

Movie-Media Group Fixed Effects

0 0.37 0 0.38

Number of reviews at conflict of interest 360 360 642 642

280797 280797 435242 435242

Indicator for Media Outlet Owned by Time Warner

Panel B. Dep. Var.: Freshness Score

Panel A. Dep. Var.: 0-100 Score for Movie

N (number of reviews)

Notes

An observa

ion is a

ovie review

edia ou

from 1985 to 2010. The de

enden

varia

le is a

ovie review conver

he 0-100 scale devise

y m

tac

iti

.com . The

standard errors are clustered by movie.

Indicator for Media Outlet Owned by News Corp.

Indicator for Warner Bros. Movie on TW-Owned Outlet

Indicator for Warner Brothers Movie

TABLE 2

THE EFFECT OF CONFLICT OF INTEREST ON MOVIE REVIEWS: AVERAGE BIAS

egress

ons

Indicator for Fox Movie on News Corp.-Owned Outlet

Indicator for 20th Century Fox Movie

fica

on:

* significant at 10%; ** significant at 5%; *** significant at 1%

Indicator for Warner Brothers Movie

Indicator for Media Outlet Owned by Time Warner

Indicator for Media Outlet Owned by News Corp.

Indicator for Fox Movie on News Corp.-Owned Outlet

Indicator for Warner Bros. Movie on TW-Owned Outlet

N (number of reviews)

Indicator for 20th Century Fox Movie

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Panel A. Newscorp. Media

-0.1898 -0.1898 -0.1898 0.713 0.3875 -0.3810 0.1913 0.2261 -0.5384

(Measure of Conflic

of In

eres

for News Corp.) [1.0596] [0.8868] [1.0230] [1.0134] [1.1291] [1.2089] [1.3622] [1.1532] [1.2192]

0.45 0.45 0.45 0.6 0.39 0.46 0.64 0.45 0.55

Number of reviews at conflict of interest 421 421 421 421 300 350 411 420 460

291124 291124 291124 291124 72610 239994 106941 278433 163993

Panel B. Time Warner Media

-0.0188 -0.0188 -0.0188 -0.0558 0.4248 -0.0184 0.6316 0.8640 0.6458

(Measure of Conflic

of In

eres

e Warner) [0.8762] [0.5170] [0.4886] [0.8233] [0.8739] [1.0742] [1.0787] [0.9181] [1.0126]

0.47 0.47 0.47 0.6 0.42 0.48 0.64 0.46 0.56

Number of reviews at conflict of interest 685 685 685 685 594 485 657 685 711

450699 450699 450699 450699 115940 359201 163937 413953 237182

Movie-Media Group Fixed Effects

X XXX XX XX X

Movie Fixed Effects

Extra

Controls

Benchmar

Cluster by

Studio

Cluster by

Media

Extra

Controls

MetaCritic

Sample

Only

Rotten

Tomatoes

Sample

Only

Match Uses

3 (not 10)

Best

Matches

Match Uses

Correlation

in Reviews

Match Does

Not Use Year

and Ratings

No.

SampleStandard Errors Matching Procedure

* significant at 10%; ** significant at 5%; *** significant at 1%

N (number of reviews)

Note

An observation is a

ovie review

edia out

et from 1985 to 2010. The de

endent varia

le is a

ovie review converte

on the 0-100 scale devi

y m

tac

itic.com.The

tandar

errors are c

ustere

ovie unle

stated otherwi

e. In Columns

(2)-(3) robustness to alternative ways of clustering is tested. In Column (4) movie fixed effects are added to the regression as additional controls. In column (5)-(6) only reviews either of the Metacritic or the Rotten Tomatoes dataset are included. Column (7)

uses three compared to the 10 best movie matches as control movies. In Column (8) movie matches are determined based on the correlation of the individual user ratings compared to the minimum distance measure in the benchmark specification. Restrictions

on potential matches explained in Section 2.2. are applied in determining the movie matches. In column (9) we do not restrict the set of potential movie matches based on the closeness in the release year and number of user ratings as explained in section 2.2.

Indicator for Warner Bros. Movie on TW Outlet

Control Variables:

Specification:

Robustness Check:

TABLE 3

THE EFFECT OF CONFLICT OF INTEREST ON MOVIE REVIE

S: ROBUSTNESS

S Regressions

Movie Review on a 0-100 Scale for Movie m in Media Outlet o

Indicator for Fox Movie on News Corp. Outlet

Dep. Var.:

Specifica

ion:

Specification:

Dependent Variable:

(1) (2) (3) (4) (5)

1.2906 2.0460*** 2.0720*** 1.8516** 0.618

(Measure of Conflic

of In

eres

for News Corp.) [0.9450] [0.7410] [0.7382] [0.8741] [0.9372]

-2.4572***

[0.7556]

-4.4576*** -1.7134*** -1.5201*** -1.5688*** -1.2759*

[0.2357] [0.4544] [0.4567] [0.4602] [0.7293]

0 0.45 0.45 0.45 0.46

Number of Reviews at Conflict of Interest 620 620 620 421 421

469252 469252 408695 401595 154572

-0.9727 -0.4512 -0.4515 -0.8342 0.0866

(Measure of Conflic

of In

eres

arner) [0.7597] [0.6606] [0.6563] [0.7209] [0.7864]

-2.9171***

[0.6546]

3.5985*** 1.6477 0.2341 0.217 -11.3558***

[0.2688] [1.4247] [1.7685] [1.8088] [1.1743]

0 0.45 0.45 0.45 0.46

Number of Reviews at Conflict of Interest 842 842 842 685 685

469252 469252 369490 361575 185262

Movie Fixed Effects

XXXX

Media Outlet Fixed Effects

XXXX

e on

y mov

es w

eas

one rev

y an ou

ent

conflict of interest

XXX

Exclude Fox/TW movies not in Filixter/Netflix/MovieLens data set

Exclude movies that are never matches to a Fox/TW movie

Sample:

Panel B. Time Warner

* significant at 10%; ** significant at 5%; *** significant at 1%

Note

n obs

rvat

on i

a movi

revi

w by a m

dia out

t from 1985 to July 2011. The de

endent varia

amovi

revi

w convert

d on the 0-100 sca

devi

d by m

tac

iti

.com . The standard errors are c

ust

red by movi

Indicator for 20th Century Fox Movie

Indicator for Warner Brothers Movie

Indicator for Media Outlet Owned by News Corp.

Indicator for Media Outlet Owned by Time Warner

Control Variables:

TABLE 4

THE EFFECT OF CONFLICT OF INTEREST ON MOVIE REVIE

S: CROSS-SECTIONAL ESTIMATES

OLS Regressions

Movie Review on a 0-100 Scale for Movie m in Media Outlet o

Indicator for Fox Movie on News Corp.-Owned Outlet

Indicator for Warner Bros. Movie on TW-Owned Outlet

Panel A. News Corp.

Specification:

Dependen

Varia

le:

(1) (2) (3) (4)

-0.00026 -0.00008 0.00011 0.00034

Average Movie Rating [0.00069] [0.00067] [0.00068] [0.00060]

-0.00033 -0.00065 -0.00073 -0.00080*

Average Movie Rating [0.00074] [0.00058] [0.00069] [0.00043]

-0.00009 -0.00003 0.00108*** 0.00041

Average Movie Rating [0.00033] [0.00039] [0.00037] [0.00043]

0.01137 0.00598 0.00583 -0.00294

[0.03947] [0.03960] [0.03774] [0.03330]

0.02843 0.05148 0.06177 0.07236***

[0.04327] [0.03256] [0.03833] [0.02410]

0.0064 -0.05288**

[0.01911] [0.02122]

0.00412*** 0.00310*** 0.00565*** 0.00413***

[0.00037] [0.00033] [0.00037] [0.00033]

Movie-Media Group Fixed Effects

0.01 0.55 0.03 0.48

1142 1142 2020 2020

120692 120692 234388 234388

* significant at 10%; ** significant at 5%; *** significant at 1%

Sample:

Average Movie Review Score

Potential review in featured media and in each of 10 matched media, with

match based on minimum distance in probability of review.

TABLE 5

CONFLICT OF INTEREST AND OMISSION BIAS: PROBABILITY OF REVIEW

Regressions

ews Co

p. Conflic

of Interes

ime

Conflic

of Interes

Notes

Each colum

is a separate regression inc

uding as observations potentia

ovie revie

y the feature

edia out

ets, or

y any of 10

atche

edia, with match

ase

on minimum distance in the probability of review as described in Figure 7. The sample only includes years in which the media featured in the relevant column are owned by

News Corp. or Time Warner. The average score is computed as the average 0-100 score for a movie from all media outlets. All specifications include fixed effects for the movie-

media group. The standard errors are clustered by movie.

or var

or rev

ew o

a mov

e m

y ou

Indicator for Conflict of Interest

Media Owned by Conglomera

Control Variables:

Indicator for Conflict of Interest *

Num

er of po

reviews a

conflic

interest

Media Owned by Conglomerate*

Distributed by Conglomerate

Distributed by Conglomerate*

(1) (2) (3) (4) (5)

0.00315 -0.00816 0.000352 -0.0572 -0.00338

(RottenTomatoes owned by Newscorp.: 2006-09) [0.00832] [0.00685] [0.00822] [0.0352] [0.0183]

-0.00204 -0.00638* -0.00579 -0.0474** -0.0271**

[0.00436] [0.00372] [0.00477] [0.0184] [0.0106]

0.0188*** 0.0183*** 0.0409***

[0.000140] [0.0000535] [0.000131]

0.0173***

[0.0000928]

Year Fixed Effects

XXXXX

Media Outlet Fixed Effects

XXXX

Movia-Media Group Fixed Effects

All Reviews,

Matching Group

Sample

All Reviews

Scored in RT

Only Reviews

with

50<=Score<=70

0.74 0.65 0.59 0.05 0.56

295400 397420 153229 97940 28225

0-100 Review Score

TABLE 6

BIAS IN REVIE

AGGREGATORS: EFFECT OF NE

SCORP. O

NERSHIP OF ROTTEN TOMATOES

egress

ons

RottenTomatoes 0-1 "Freshness" indicator

Indicator for 20th Century Fox Movie

Depedent Variable:

ecification:

Indicator for 20th Century Fox Movie *

* significant at 10%; ** significant at 5%; *** significant at 1%

0-100 MetaCritic Review Score

Only Reviews Unscored in RTSample:

Notes

An observa

on is a

ovie review from 1985 to

2011. Colum

(1) uses

he FOX sa

le o

ching da

ase

whi

e colum

(2)-(5) use

he ful

acri

c/ Rot

Toma

oes da

aset.The

dependent variable is an indicator variable for 'freshness' of a movie according to review in Rotten Tomatoes. The key indepedendent variables are indicators for movies distributed by 20th Century Fox and an

interaction of this indicator with the years in which Rotten Tomatoes is owned by News Corp. (2006-09). The standard errors are clustered by movie.

Control Variables:

No. Reviews

While Owned

(At Conflict of

Interest)

No. Reviews

While Not

Owned

No. Reviews

While Owned

(At Conflict of

Interest)

No. Reviews

While Not

Owned

News Corp. All Reviews 335 media 1985-2010 Varies Varies 75326/49208 -

270147/280797 -

54349/49208

4058 (421) 287066 3410 (360) 277387

News Corp. Beliefnet Website 1994-2010 News Corp.

from '08 to '10

Ato F(+/-

allowed)

0/0 - 302/289 - 0/0 241 (21) 61 230 (21) 59 Ne

Minow (302/289)

News Corp. Chicago Sun-Times Newsp. 1985-2010 News Corp.

until 1986

0 to 4 stars (1/2

allowed)

2176/1622 - 2436/1930 -

2103/1622

67 (13) 2442 54 (13) 1876 Roger Ebert (2281/1706)

News Corp. New York Post Newsp. 1989-2010 News Corp.

from 1993

0 to 4 stars (1/2

allowed)

2168/1990 - 2159/2088 -

2085/1990

2241 (218) 1 2088 (211) -

ou Lumenick (1087/1029),

yle

Smith (485/477)

News Corp. News Of The World Newsp. (U

) 1985-2010 News Corp. 1 to 5 stars 0/0 - 87/84 - 0/0 87 (8)

84 (8)

Robbie Col

in (87/84)

News Corp. Sunday Times Newsp. (U

) 1985-2009 News Corp. 0 to 5 stars (1/2

allowed)

0/0 - 206/162 - 0/0 206 (27)

162 (27)

Shannon J. Harvey (156/117)

News Corp. TV Guide Weekly 1985-2009 New Corp. 1988-

0 to 4 stars (1/2

allowed)

2632/1855 - 2015/1897 -

1969/1855

588 (78) 2090 129 (21) 1768 Mait

and McDonagh (1388/1104),

Ken Fox (473/405)

News Corp. Times Newsp. (U

) 1985-2010 News Corp. 0 to 5 stars 0/0 - 437/495 - 0/0 437 (40)

495 (45)

endy Ide (132/159), Ja

Christopher (185/222)

News Corp. Wall Street Journal Newsp. 1985-2010 News Corp.

from 2008

Qualitative 1197/328 - 345/336 -

345/328

191 (16) 1006 168 (14) 168 Joe Morgenstern (1045/282)

News Corp. Other Reviews 327 media 1985-2010

Varies 67153/43413 -

262160/273516 -

47847/43413

- 281466 - 273516

Time Warner All Reviews 335 media 1985-2010 Varies Varies 118538/76549 -

417346/435242 -

85185/76549

6168 (685) 444531 5461 (642) 429781

Time Warner Cinematical Website 2004-2010 Ti

arner

until 2009

0 to 5 stars (1/2

allowed)

0/0 - 584/689 - 0/0 575 (61) 9 676 (76) 13

es Rocchi (127/159), Scott

Weinberg (123/122)

Time Warner CNN.com Website/Radio 1996-2007 Ti

arne

Qua

itative 0/0 - 42/929 - 0/0 42 (5)

929 (120)

aul Clinton (0/596)

Time Warner Entertainment Weekly Weekly 1990-2010 Ti

arner

from 1990

Ato F(+/-

allowed)

4171/3140 - 3445/3240 -

3314/3140

4302 (463) - 3240 (349) - Owen G

an (2038/1483),

Lisa Schwarzbaum (1626/1318)

Time Warner Time Weekly 1990-2010 Ti

arner

from 1990

Qua

itative 1249/474 - 507/616 -

507/474

1249 (156) - 616 (97) - Richard Corliss (654/328), Richard

Schickel (568/269)

Time Warner Other Reviews 331 media 1985-2010

Varies 113118/72935 -

412768/429768 -

81364/72935

- 444522 - 429768

Notes: The sources of the movie revie

a are www.m

tac

iti

.com (a

brevia

ed MC) and ww

oma

.com (a

brevia

ed RT). The da

acoversrevie

sava

le from 1985 un

2010. See

ext for addi

ona

informa

on.

Yea rs O wn er

Usual Rating

System

Data Source (MC Score (0-

100)/ MC Fresh Ind. - RT

Score (0-100)/ RT Fresh

Ind. - Both Score (0-100)/

Both Fresh Ind.)

ore Var

res

ator

APPENDIX TABLE 1, PANEL A

SUMMARY STATISTICS: MEDIA SOURCES OF MOVIE REVIEWS

Sample Media Outlet Media Type

Most Common Reviewers

(Reviews Score (0-100)/ Reviews

Fresh Ind.)

Sample

Distrib

tor of Movie

(Studio) Studio Type Years Owner

No. of Reviews (S

ore 0-

100/ Fresh Ind.)

No. of Movies (S

ore 0-

100/ Fresh Ind.)

News Corp. All Studios Varies 1985-2010 Varies 291124/280797 1593/1278

News Corp. 20th Century Fox Major 1985-2010 News Corp. 21080/20500 236/189

News Corp. Fox Searchlight Independent 1995-2010 News Corp. 8020/8221 72/61

News Corp. Other Studios Varies 1985-2010

262024/252076 1285/1028

Time Warner

ll Stu

ios Varies Varies 450699/435242 1847/1532

Time Warner Warner Bros. Major 1990-2010 Ti

eWarner

from 1989

30195/30808 288/243

Time Warner Fine Line Independent 1991-2005 Ti

eWarner

from 1989

2474/1968 33/20

Time Warner HBO Other 1997-2003 Ti

eWarner

from 1989

76/30 3/1

Time Warner New Line Independent 1992-2008 Ti

eWarner

from 1996

12135/12431 114/98

Time Warner Picturehouse Independent 2005-2008 Ti

eWarner

from 1989

892/998 8/8

Time Warner Warner Independent Independent 2004-2006 Ti

eWarne

1444/1591 11/11

Time Warner Warner Home Video Other 1994-1999 Ti

eWarner

from 1989

63/149 2/4

Time Warner Other Studios Varies 1985-2010

403420/387267 1388/1147

Notes: The sources of the movie revie

data are www.m

tac

itic.com (a

breviated MC) and www.

ott

ntomatoe

.com (a

breviated RT). The data covers revie

savai

le from 1985 unti

2010. See text for additiona

information.

APPENDIX TABLE 1, PANEL B

SUMMARY STATISTICS: STUDIOS

Appendix Figure 1a-b. Bias by Journalist: News Corp.-owned outlets (1a) and Time

Warner-owned outlets (1b)

Notes: Appendix Figure 1a displays the average review score (on a 0 to 100 scale) of 20

Century Fox movies against the average

review score of the associated movie matches for News Corp. journalists and journalists not employed at a News Corp. outlet. Colors

indicate whether a particular journalist is employed at a News Corp. outlet or is one of the other 500 journalists with the most reviews.

News Corp. journalists with a number of reviews of 20

Century Fox movies less than 15 are excluded. Appendix Figure 1b displays

the parallel evidence for journalists employed at a Time Warner outlet and Warner Bros. movies.

Online Appendix Figures 1a-1f. Documenting the Quality of Movie Matches: Micro-level

Comparison

Online Appendix Figure 1a-b. Similarity to Match: Movie Genre

Online Appendix Figure 1c-d. Similarity to Match: MPAA Rating

Online Appendix Figure 1e-f. Similarity to Match: Theaters at Openings

Notes: Online Appendix Figures 1a-b display the fraction of movie matches in a particular genre conditional on the genre of the associated movie

distributed by 20

Century Fox or Warner Bros. Online Appendix Figures 1c-d display the fraction of movie matches with a particular MPAA

rating conditional on the MPAA rating of the associated movie distributed by 20

Century Fox or Time Warner. Online Appendix Figures 1e-f

report local polynomial regressions with Epanechnikov kernel with bandwidth of 250 and 1

degree polynomial of the number of theaters at

opening of movie matches on the number of theaters at opening of associated movies distributed by 20

Century Fox or Warner Bros.

Online Appendix Figures 2a-d. Documenting Quality of Movie Matches on Outcomes: 0-

100 Review Score

Notes: Online Appendix Figure 2a displays the fraction of movies in the quintiles of the average review score from critical review for 20

Century Fox movies, the associated movie matches, and movies which are not matches. For the purpose of assigning movies to a particular

quintile, 5-year bins are formed and the quintile of a particular movie is determined within a bin. Online Appendix Figure 2b presents the parallel

evidence for the Warner Bros. movies. Online Appendix Figure 2c reports a local polynomial regression with Epanechnikov kernel with a

bandwidth of 5 and 1

degree polynomial of the average review score of movie matches on the average review score of the associated 20

Century Fox movies. Online Appendix Figure 2d displays the same relationship for Warner Bros. movies.

Online Appendix Figures 3a-d. Documenting Quality of Movie Matches on Outcomes:

Probability of Review

Notes: Online Appendix Figure 3a displays the fraction of movies in the quintiles of the review probability for the movies by 20

Century Fox,

the associated movie matches, and movies which are not matches. For the purpose of assigning movies to a particular quintile, 5-year bins are

formed and the quintile of a particular movie is determined within a bin. Online Appendix Figure 3b presents the parallel evidence for the Warner

Bros. movies. Online Appendix Figure 3c reports a local polynomial regression with Epanechnikov kernel with bandwidth of 0.05 and 1

degree

polynomial of the review probability of movie matches on the review probability of the associated 20

Century Fox movies. Online Appendix

Figure 3d displays the same relationship for Warner Bros. movies.

Online Appendix Figure 4a-d. Selective coverage -- Probability of review by movie quality:

News Corp. outlets

Notes: Online Appendix Figures 4a-d segregates the evidence of Figure 7a for News Corp. outlets. Outlets with less than 50 potential movies to

review are not included.

Online Appendix Figure 5a-d. Selective coverage -- Probability of review by movie quality:

Time Warner outlets

Notes: Online Appendix Figures 5a-d segregates the evidence of Figure 7b for Time Warner outlets.

Online Appendix Figure 6a-c. Selective coverage for Time magazine, Placeboes

Online Appendix Figure 6a-b. Period of Time Warner Ownership (6a) and pre-period (6b)

Online Appendix Figure 6c. Probability of Review of 20

Century Fox movies

Notes: Online Appendix Figure 6 presents a placebo test for potential omission bias in the Time Warner outlet Time. Online Appendix Figure 6a

reproduces the evidence in Online Appendix Figure 5d. Online Appendix Figure 6b shows a placebo test for the period before 1990 in which the

Time Warner conglomerate has not been established. For this purpose, movie matches are determined for movies distributed by studios which

merged or were acquired by the Time Warner conglomerate at the time of establishment. For simplicity these movies are labeled as TW. Figure

6c shows a placebo test of omission bias by Time of 20

Century Fox movies during ownership.

Online Appendix Table 1. Documenting the Netflix, Flixter data

Notes: Online Appendix Table 1 reports summary statistics for the three user rating datasets Flixster, MovieLens, and Netflix. Only movies are kept which can be identified with a movie title.

Furthermore, for the MovieLens and Netflix datasets for which information on the release year of the movie is available, we only keep movies for which this information is provided.

(1) (2) (3) (4) (5)

Panel A. Newscorp. Media

-0.1898 0.4393 0.4966 0.8704 -0.0394

(Measure of Conflict of Interest for News Corp.) [1.0596] [1.2296] [1.1706] [1.4028] [1.0478]

0.45 0.49 0.42 0.55 0.39

Number of reviews at conflict of interest 421 233 358 189 514

291124 169575 253726 129111 322808

Panel B. Time Warner Media

-0.0188 -0.8084 0.8813 -1.3899 0.6058

(Measure of Conflict of Interest Time Warner) [0.8762] [1.1111] [1.0275] [1.1943] [0.8355]

0.47 0.51 0.41 0.54 0.4

Number of reviews at conflict of interest 685 433 567 349 772

450699 289958 376283 225849 490668

Movie-Media Group Fixed Effects

XXXXX

Benchmark

Match Uses Only

Netflix Data

Match Uses Only

Flixter Data

Match Uses Only

MovieLens Data

Match Uses

Likelihood Ratio

Method

Notes

An observa

on is a

ovie review

edia ou

from 1985 to 2010. The de

endent varia

le is a

ovie review conver

he 0-100 scale devise

y m

tac

iti

.com.Thes

andard errors are c

ere

ovie.

Columns (2)-(4) only use one of the three user rating datasets to determine the ten best movie matches of 20th Century Fox or Warner Bros. movies. In column (5) movie matches are determined for each of the three user rating

datasets based on a probability measure: For each potential movie match M and 20th Century Fox or Warner Bros. movie F, we calculate the probability of rating M condititional on rating F as P(M|F), as well as the unconditional

rating probability P(M). For each F we select the ten movies with the highest value of P(M|F)/P(M) as movie matches. We apply the constraints on the matching procedure explained in Section 2.2, but do not restrict the sample o

potential movie matches on those with at least 40 common user ratings. Compared to the baseline scenario this mathing procedure does not take into account the closeness of ratings but the probability of rating.

Control Variables:

Robustness Check:

* significant at 10%; ** significant at 5%; *** significant at 1%

Indicator for Fox Movie on News Corp. Outlet

N (number of reviews)

Indicator for Warner Bros. Movie on TW Outlet

N (number of reviews)

Specification:

Dependent Variable:

ONLINE APPENDIX TABLE 2

THE EFFECT OF CONFLICT OF INTEREST ON MOVIE REVIEWS: ADDITIONAL ROBUSTNESS

OLS Regressions

Movie Review on a 0-100 Scale for Movie m in Media Outlet o

(1) (2) (3) (4) (5)

-2.1225*** -2.0614*** -1.6554*** -2.0787*** -1.8792***

[0.4972] [0.5158] [0.4968] [0.6664] [0.5135]

1.5952***

Movie Ma

o a 20th Century Fox Movie [0.4680]

0.4868** -0.1505 -0.1873 0.2459

Predic

ed Prob. Of Be

ng a 20

h Cen

ury Fox Movie Based on

Characteristic (Standardized)

[0.2251] [0.2178] [0.3137] [0.2169]

0.45 0.45 0.45 0.46 0.45

147.31 60.53 17.51 151.05

429994 351293 378975 224816 358940

2.2712 1.0276 1.7354 0.2018 0.4371

[1.4649] [1.5769] [1.5121] [2.0935] [1.5784]

-1.4856***

Movie Ma

o a Warner Bros. Movie [0.4882]

-0.5626** -0.5647** -0.0197 0.1692

Predic

ed Prob. Of Be

ng a Warner Bros. Movie Based on

Characteristic (Standardized)

[0.2264] [0.2271] [0.3297] [0.2409]

0.45 0.45 0.45 0.46 0.45

104.74 29.31 100.64 338.48

409096 332649 358257 210258 340145

Sample:

Proxy Used and Controls:

Charac

eris

c of Movie Used

o Predic

Proba

y of Be

ng a

20th Century Fox (Warner Bros.) Movie

Genre MPAA Rating

Dec

eof Budge

Production

Dec

e of No. of

Theaters at Opening

Movie and Media Fixed Effects

XXXX

Indica

or for T

e Warne

-Owned Ou

Indica

or for T

e Warne

-Owned Ou

Indica

or for T

e Warne

-Owned Ou

Panel B. Time Warner

N (num

er of reviews)

Dependent Variable:

CONFLICT OF INTEREST IN MOVIE REVIEWS: EVIDENCE ON CORRELATED TASTES

Panel A. News Corp.

Indica

or for News Corp.-Owned Ou

Indica

or for News Corp.-Owned Ou

Movie Review on a 0-100 Sca

efor Moviem in Media Ou

Indica

or for News Corp.-Owned Ou

ONLINE APPENDIX TABLE 3

egress

ons

Exc

udes Movies a

Confl

of In

eres

(20

h Cen

ury Fox movies in Pane

A and Warner Bros. movies in

Panel B)

Note

An observa

ion is a

ovie review

edia out

from 1985 to 2010. In column (1) a 0-1 indica

for whe

ovie is a match to a 20th Cen

ury Fox o

Warne

Bros.

ovie is in

erac

with

he dum

indica

ing

ews Corp. o

eWarne

ownership. In columns (2)-(5) the ownership variable is interacted with a probability measure of being a 20th Century Fox or Warner Bros. movie. This measure is obtained after predicting from a probit regression and standardized by subtracting the mean and

dividing by the standard deviation. In columns (4)-(5) the probability is estimated from a probit regression of an indicator for a 20th Century Fox or Warner Bros. movie on the decile of the budget of production and the number of theaters at opening as a

categorical variable. Deciles are determined within 5-year intervals. In columns (2)-(3) the probability is estimated using the categories for genre and MPAA Rating respectively. Standard errors are clustered at the movie level.

can

10%

;

can

;

***

cant a

age Chi

N (num

er of reviews)

Spec

fica

on:

(1) (2) (3) (4)

-1.1754 -0.4953 -2.4027 -3.1797

(Measure of Conflict of Interest for News Corp.) [2.0402] [2.2611] [2.0722] [2.2124]

1.4372 0.2996 0.3916 0.5539

roxy for Higher Return to Biased Review [2.3503] [0.3845] [0.3529] [0.3730]

0.45 0.45 0.45 0.45

Number of reviews at conflict of interest 421 308 410 416

291124 231731 282908 285171

-0.6096 -2.6253 -0.4526 -0.7714

(Measure of Conflict of Interest Ti

e Warner) [1.2862] [1.9628] [1.6518] [1.8568]

[10.3605] 0.3941 0.1042 0.1568

roxy for Higher Return to Biased Review 0.9219 [0.3088] [0.2626] [0.2824]

Proxy Used and Controls:

Proxy for Higher Return to Biased Review

Indicator for Major

Studio

Decile of Budget of

Production

Decile of No. of

Theaters at Opening

Decile of Domestic

Box Office

Movie-Media Group Fixed Effects

XXXX

Interaction of Proxy with Fox Movie Indicator and with

Newscorp Outlet Dummy

XXXX

0.47 0.47 0.47 0.47

Number of reviews at conflict of interest 685 508 651 664

450699 362124 437617 438888

egress

ons

Indicator for Warner Bros. Movie on TW Out

et*

N (num

er of reviews)

Notes

An observa

on is a

ovie review

edia out

from 1985 to 2010. In colum

(1) i

each

ovie group a

ndica

whe

he 20th Cen

ury Fox o

Warne

Bros.

ovie is dis

ute

ajo

udio i

interacted with the independent variables. In column (2)-(4) for each movie group the decile of the 20th Century Fox or Warner Bros. movie (on a 1-10 scale) is interacted with the independent variables. Deciles ar

determined by grouping the 20th Century Fox or Time Warner movies in 5-year bins and calculating the deciles within each bin. Note that the indicator or decile is unique in each movie group, and thus drops out given the

movie-media group fixed effects. The standard errors are clustered by movie.

cant a

10%

;

cant a

;

***

cant a

Indicator for Fox Movie on News Corp.-Owned Out

et *

N (num

er of reviews)

Panel B. Time Warner

Indicator for Warner Bros. Movie on TW Out

ONLINE APPENDIX TABLE 4

CONFLICT OF INTEREST IN MOVIE REVIE

S: COMPARATIVE STATICS ON RETURN TO BIAS

Movie Review on a 0-100 Sca

e for Movie m in Media Out

et o

Panel A. News Corp.

Indicator for Fox Movie on News Corp.-Owned Out

Dependent Variable:

ecification:

Specification:

Dependent Variable:

Panel A.

Chicago

SunTimes

New York

Post

News of the

World TV Guide Times (UK)

Wall Street

Journal Beliefnet

Sunday

Times

(1)(2)(3)(4)(5)(6)(7)(8)

1.1914 0.9127 -6.1731 1.9354 0.8042 -4.9835 6.2866* -12.3594

[11.8712] [1.3914] [5.2753] [1.9623] [3.5214] [6.1281] [3.3805] [8.6897]

Movie-Media Group Fixed Effects

XXXXXXXX

0.73 0.45 0.56 0.51 0.5 0.55 0.43 0.69

Number of reviews at conflict of interest 13 218 7 73 39 16 21 16

796 243400 2146 24535 22372 9741 11184 3129

Panel B.

CNN.com

Entertainme

nt Weekly Time Cinematical

(1) (2) (3) (4)

0.2035 0.4245 -1.1355 -0.0143

[7.5661] [1.0085] [1.8649] [3.1377]

Movie-Media Group Fixed Effects

XXXX

0.48 0.47 0.52 0.53

Number of reviews at conflict of interest 2 463 146 60

553 446840 79944 30074

* significant at 10%; ** significant at 5%; *** significant at 1%

Notes: An observation is a movie review by a media outlet from 1985 to 2010. Each column is a separate regression including as observations only movies with at least one review by the featured outlet, and as independent

variables indicator variables for the outlet and for production by the conflicted distributing company (20th Century Fox and Warner Bros.). All specifications include fixed effects for the movie-media group.The standard

errors in parentheses are clustered by movie.

ONLINE APPENDIX TABLE 5

THE EFFECT OF CONFLICT OF INTEREST ON MOVIE REVIEWS: BY MEDIA

Indicator for Conflict of Interest

Movie Review on a 0-100 Scale for Movie m in Media Outlet o

Control Variables:

OLS Regressions

Time Warner Conflict of Interest

Indicator for Conflict of Interest

News Corp. Conflict of Interest

Specification:

Dependent Variable:

Panel A.

TV Guide Beliefnet Times (UK)

Journal

Lou Loumenick Kyle Smith

Jonathan

Foreman

Megan

Lehmann

Mait

and

McDonagh

Nell Minow

James

Christopher

Joe

Morgenstern

(1) (2) (3) (4) (5) (6) (7) (8)

1.8109 0.339 -3.8099 -0.368 6.3402** 6.2866* 0.7005 -4.5947

[1.9946] [4.2489] [4.3026] [4.1527] [2.9709] [3.3805] [6.6586] [6.4825]

Movie-Media Group Fixed Effects

XXXX X X

0.5 0.51 0.68 0.68 0.54 0.43 0.64 0.61

Number of reviews at conflict of interest 103 46 28 20 29 21 18 15

73641 24666 10115 8249 7818 11184 6814 7794

Panel B.

Owen

Gleiberman

Lisa

Schwarzbaum

Richard Corliss

Richard

Schickel

(1) (2) (3) (4)

0.7818 0.7173 1.423 -5.9921*

[1.5304] [1.5697] [2.6838] [3.4257]

Movie-Media Group Fixed Effects

XXXX

0.55 0.56 0.56 0.67

Number of reviews at conflict of interest 225 174 74 50

117097 100662 33894 15064

OLS Regressions

* significant at 10%; ** significant at 5%; *** significant at 1%

ONLINE APPENDIX TABLE 6

THE EFFECT OF CONFLICT OF INTEREST ON MOVIE REVIEWS: BY REVIEWER

Indicator for Conflict of Interest

New York Post

Control Variables:

Movie Review on a 0-100 Scale for Movie m in Media Outlet o

News Corp. Conflict of Interest

Indicator for Conflict of Interest

Note

An observa

is a

ovie review

y a journali

from 1985 to 2010. Each colum

is a

ara

eregre

ion including as observa

ions onl

ovie

with a

leas

one review

he fea

ure

journali

and as inde

endent varia

indica

varia

for

he ou

and for

roduc

conflicted distributing company (20th Century Fox and Warner Bros.). Reviews by other journalist from the same conglomerate are taken out. All specifications include fixed effects for the movie-media group.The standard errors in parentheses are clustered by movie.

Control Variables:

Time Warner Conflict of Interest

Ente

ainmen

eek

yTime

ONLINE APPENDIX TABLE 7

CONFLICT OF INTEREST AND OMISSION BIAS: BY MEDIA OUTLET

Panel A.

Chi

ago

SunTimes New York Post

News of

World TV Guide Times (UK)

all S

Journal Beliefnet Sunday Times

(1)(2)(3)(4)(5)(6)(7)(8)

-0.02578*** 0.00186 -0.00164 -0.00041 -0.00232 -0.0013 -0.00315 0.00048

Average Rating [0.00857] [0.00124] [0.00375] [0.00150] [0.00207] [0.00448] [0.00542] [0.00107]

1.67750*** -0.10598 0.12088 0.00573 0.13175 -0.01133 0.17137 -0.01561

[0.49981] [0.07619] [0.21264] [0.08316] [0.11634] [0.23296] [0.29734] [0.05892]

Movie-Media Group F.E.

XXXXXXXX

0.37 0.37 0.24 0.34 0.26 0.27 0.2 0.28

Number of Fox movies in sample 19 285 44 161 193 35 35 370

1045 32285 3036 16940 20537 2695 2684 41470

Panel B.

CNN.com

nte

ainmen

Weekly Time Cinematical

(1)(2)(3)(4)

0.00088 -0.00097 0.00239** -0.00426

Average Rating [0.00129] [0.00072] [0.00109] [0.00288]

-0.03184 0.06471 -0.09498* 0.1907

[0.07376] [0.04185] [0.05696] [0.16413]

Movie-Media Group F.E.

XXXX

0.28 0.44 0.34 0.26

Number of Warner Bros. movies in s

517 660 660 183

58916 78529 78529 18414

* significant at 10%; ** significant at 5%; *** significant at 1%

Confl

of Interes

Confl

of Interes

Control Varia

les:

Confl

of Interes

Confl

of Interes

ent

review in fea

ured media and in 10 ma

ched media

(match based on minimum distance in review probability)

Notes

Each colum

is a separate regress

uding as observat

ons potent

ovie reviews

y the feature

edia out

et, o

y any of 10

atche

edia, with match

ase

um distance in proba

of reviews as descri

in F

gure 7.

The sample only includes years in which the media featured in the relevant column is owned by News Corp. or Time Warner. The average score is computed as the average 0-100 score for a movie from all media outlets. The standard errors in

parentheses are clustered by movie.

le:

otent

review in fea

ured media and in 10 ma

ched media (ma

ch based on m

um dis

ance in review proba

Control Varia

les:

Sample:

OLS Regressions

Indicator variable for review of movie m by media outlet o

News Corp. Conflict of Interest

Time Warner Conflict of Interest

Specification:

Dependent Variable: