The 12th Law & Economics Conference on Media Funding took place on Jan 23-24 2023.
Follow this link for more information on the conference: https://www.zew.de/VA4257-1?_gl=1*n2f2mf*_ga*MTk4MTEwNTM2LjE2ODM4MDQ1ODg.*_ga_KFD4G5CY27*MTcwNDg4NjM2Mi43Mi4xLjE3MDQ4ODY0ODYuMC4wLjA.
Please check our website for upcoming events:
https://www.macci-mannheim.eu/macci-mannheim-centre-for-competition-and-innovation/events
Podcast theme music by Transistor.fm. Learn how to start a podcast here:
https://transistor.fm/how-to-start-a-podcast/?via=music
[Applause] [Applause] Okay thank you thanks a lot for uh the invitation I’m glad to be there I’m going to present you this paper on hosting media bias uh which is joint work with Maurice Andel uh who was a PhD student at Sho and who actually just uh join the European Union to work on DSA
Data uh Nikolai who is a computer scientist at The Institute National Visual and K that you all know here in the here in the in the room okay it works uh I think if you if you look at the the state of the media industry a little bit everywhere in the
World or at least in Western democracies you see that the state of the industry is pretty bad uh we see that with the increasing concentration of the of the industry you can look at at the case of the us when you see a small number of conglomerate buying more and more local
Newspapers which is also true for the local TV Market uh this is also the case uh in France uh when there there were a recent inquiry uh into VI this takeover of a number of different media Outlets including like newspaper radio TV in fact the European commission just gave
The green light a week ago uh and there was this issue of the lack of pluralism which is not specific to the US or to France that we also find for example in the in the UK for example there were a number of issues against muruk and
Decisions taken by both ofcom and the regulation Authority uh in the in the UK this issue of like media concentration like pluralism of the media who owns the media in a sense is not new it’s pretty like old issue in the econ literature I can see like the the first paper I guess
Would be like jov won the media in 2003 with a number of Coors of course there was a famous Fox newspaper by deia and capap on the on the consequences of change in ownership but the thing is that uh we have like this growing concerns about the extent of Media
Consolidation in particular because if I compared in the number of countries the state of the media industry 20 years ago and today it is worse today than what it was 20 years ago it might seem strange to you because a lot of people are going to argue this is a strange overview of
The media industry look at the number of media outlets around okay go on the internet uh open a website you will find like uh 100 different news websites uh and you will find also a lot of like TV channels and you can find some like TV channels on YouTube so there are like
Much more media today than they were in the past so we should see less concentration I want to highlight two things here first this is not true in the majority of the case in terms of market share second even if it was true in ter of market share market share is
Not enough at all to consider here and what we really need to look at is what Andrea Prat first used to call attention shares in his Media power paper and this attention shares by the way what was was used in the UK recently in the case
Against Murdoch so the issue is not so much to know how many media do you have or what are the market shares of the different media Outlets you want also to you to look at the way people consume news and it turns out that in today’s world despite the huge supply of
Different media Outlets a lot of people they still just consume one or two media and for example in Andrea’s work we see very well that a lot of people for example just watch Fox News so you don’t really care that Fox News is competing with CNN in terms of position to
Pluralism you want to understand the kind of uh content that is produced of Fox News to know whether or not people get different point of view on uh what is happening in the uh in the world okay and this is why when we talk about pluralism in general there are two kind
Of regulations that can be implemented one kind of Regulation is what we can call external pluralism so this is really where you want to limit Market concentration so basically you will have some rules we have rules for Market concentration for all uh the different sectors in general in the number of
Countries you have extra rule for the media where you will say that for example like a given owner cannot own more than 30% of the total market share or that you can own a TV and a newspaper but then you cannot own a radio this kind of like cross ownership rules but
These rules might not be sufficient and this this is where a number of countries thought it was of importance to also consider internal pluralism which is to ensure diversity of point of view within a given media outlets in particular for TV and also for radio because we know that the
Number of people will just watch one TV channel and not many TV channel this has first been the case in fact in the US like the US tends to innovate on number of topic but very often they introduce early regulation and then they changed their mind also early in the US this was
The fairness Doctrine introduced in the 1950s by the FCC the Federal Communication Commission it was repealed at the end of the 1980s uh following re ran uh policies and the fact that he changed the head of the FCC when he made it to power but you still have the
Equivalence of the fairness Doctrine in the UK for example with the ofcom or in France uh with with the arom and in a sense this will be kind of the focus of this paper when we are going to look at pluralism of viewpoints broadcast in each media Outlets I’m also quickly
Going to discuss because this is linked to our findings other tools that you can use to ensure pluralism and in particular the fact of reinforcing protections for editorial Independence in The Newsroom and I again that this is at the middle of the discussion of the European media uh Freedom uh Freedom Act
Okay one of the issue why might what might have have when we talk about media concentration people might say okay you are afraid about media concentration because you think that if there is a lot of concentration one single owner is going to determine the editorial line of
Media outlet and so this will have important consequences but then people might tell you yeah okay you have one owner but then you have 100 people or 200 or 500 or 800 different journalist working for a media outlet and these different journalist might have a variety of viewpoints and they might use these
Viewpoints in a sense to drive the editorial guidelines depending on their preferences okay and in a sense the variety of journalists can compensate the fact that we only have a few owners and this is what we’re going to measure in this in this paper okay what we are
Going to measure in this paper in a sense is the extent to which journalist have some agency and the extent to which journalist can buyas the news depending on their own preferences in independently of the editorial line of the media outlets for that we are going to focus on TV and
Radio and we are going to focus on TV and radio because the way we are going to measure bias in this paper is by looking at the guests who are invited to to speak in the different uh shows okay so basically we are going to tackle the extent to which journalist have some
Agency when decided who speaks in their show for that what we did with SC Moritz and uh Nicola is that we built a completely new data set that basically include the universe of shows broadcast on the main French TV and radio channels between 2002 in 2020 so we have 20
Outlets so basically the idea is to have all the generalist TV channels and radio stations I will come back to the data a little bit later on uh over during this time period uh we have data on all the apparences of guests on this show so we
Have 2.3 million guests uh we have 39 distinct host and we have uh two60 distinct guest at the end of the of the day when I say I’m going to look at uh the decision of who I want uh to invite on my show I’m going to focus here on
Like political bias so what we need to do once we have the identity of the guest is to classify them and in particular to classify them politically we will do it in two way and this is pretty important for us the first thing that we will do is to classify
Politicians and then we will also classify those that we call in the paper are the Pinups for politically engaged non-politician so this can be like activist this would be pandits so people who are not politicians who are not candidate who never run for campaign but who are politically engage and you will
See that this matter a lot in particular in countries like France but also the UK when you have some regulation on the speaking time of politicians and where channels may decide to use a Pinups as a way to escape regulation because in none of these countries and in none of the
Countries I know of in fact you have speaking time regulation for the pinup okay and so by using this classification of gu between like non-politician politicians Pinups we can compute the time share of each political family for each of the of the show okay and then we
Are going to look at variation across channels in the political leaning of guest and in particular we are going to see whether or not channels have different OST whether they impose different guidelines and whether also us they tends to sort on different channels depending on their political uh
Political preferences okay how can we identify that okay the role played by OST versus the role played by channels and other factors we can do it because we have a lot of movers so we have a lot of journalist in our data set who either like within a given year or from one
Year to the other are going to move from one channel to the other so we can use this move to see whether they are going to change the kind of host the invite upon move depending on the channel to join okay and this is going to allow us like technically this would be
Identified with the two-way fixed effect fixed effect model this is going to allow us to explain the share of the variance between channels that come to the channel from the channel fixed effect and the channel fixed effect can reflect either the owners preferences or the test of of the
Audience that I will come back to it or whether this is due to the journalist preferences or whether this is due to sorting so sorting is okay right-wing journalist decide to join rightwing Channel leftwing journalist they decide to join leftwing Channel and so there is a coincidence between the preferences of
Of uh of the two okay so this is what we do in the first part of the paper and what I will present to you in the first part of the talk okay and then the second thing that we do that we will use an event study design so we will use an
Important change in media ownership in recent years in France and in particular the fact that vanon b so vanon B gave rise to a lot of media coverage by International newspapers and basically all the newspapers in the UK in the US they call him the French Murdoch so
Let’s call him the French Murdoch because this is a good way to define the guy uh he he he he took cover three TV channel in 2015 uh he has a pretty strong radical uh right wing uh stance and we will see the extent to which first thing we will
Quantify the extent to which the take over by using a difference in difference design shift the set of guest invited to the right so that we will quantify it but then we will go one step further and I think this is really where our paper contribute with a
Two-part is that we will understand the mechanisms through which media bias work and in particular will kind of open the black box of media bias by looking at the host who State looking at the host who left following uh the the the Takeover looking at their destination Channel
And see in fact the extent to which the REM REM remaining H um comply or not with the new editorial line and I think this is an important contribution if I’m just telling you okay you have a new owner okay you have like B it’s like you
Have Murdock and Fox News you have a shift to the right that’s nice we can quantify it that might be of interest for for French regulator but that’s not really new what is really new here is ready to go one step further and to understand the way bias work okay and a
Lot of people don’t know how bias work you can Wonder like whether this is about the owner showing up into The Newsroom every morning saying okay let’s look at what we going to broadcast tonight or whether this is about in picking the right guys at the right place which is something that we’re
Going to document and you will see that also and this is kind of key the agency of the journalist so the characteristics of the journalist who stayed who left who comply more or less it depends on the number of characteristics it will depend on their gender to begin with it
Will also depends of their experiment it will depends on their ratings so you will see that in a sense the agency of the journalist or the way they can fight against the change in editorial line like all the journalists are not equal and you you can like have policy
Implications from this kind of uh of findings just to give you a brief overview of the results the first result that we have is that basically uh there is a lot of compliance so the channel fixed effect overall they explain around 90% of the observed variance between
Channels which is huge and not only is huge but it has increased over time and increase we are going to relate to the fact that the market is becoming more and more concentrated with less and less journalists job so at the end of the day journalist when they have a job they
Follow the editorial line uh composition so the the the the role played by journalist preferences is very low like around 3% and we do have a little bit of uh of sorting in terms of how us react to ownership change you will see that uh there is a lot of compliance something
That you see that following the Takeover you have an increase by nearly 50% in the speaking time of the radical right and if you look at the journalist who State the magnitude of the increase is nearly as high as overall so it really mean that the journalist who stay
They just comply with the new editorial line and we will also document uh some uh sorting with the leing of a number of uh host so compared to the existing literature I guess we contribute mostly to two kind of different literature the first one is a literature on media bias
So there is a lot of literature that try to measure uh media bias you can do that by looking looking at endorsement the literature working on historical newspaper have done that a lot in particular in the US more and more thanks to the progress of natural language uh uh processing you can look
At the language used by the newspapers and the more we will have transcript the more we will also look at the language used by the TV and radio stations you can look look at Agenda setting uh you have also new work that is pretty nice
Uh on visual bias in this paper what we look at is a choice of guest so in the sense what we do the closest work in the literature is the one by reben D and Brian night on beron TV with two main contribution that are important uh the
First contribution is really the idea of not focusing only on politicians and this really matter you need to consider all the invited guest and the second one that is even more important in a sense that in general when people tackle this issue in political economy they just consider
Newscast but we would show you that we should not focus only on newscast we should focus on other the kind of shows in particular because on a lot of TV channels in including 24hour News Channel you have less and less newscast more and more talk shows because it’s
Less expensive very often to produce a talk show than to produce some news on the field and this is also easy way for you to do some politics because you take two guys they fight together they’re not politicians they Pinups they are not regulated you can push an agenda this
Way okay the second literature we contribute to is the literatur on the determinance of media bias so at least since 2010 against kapiro papers there was a huge fight whether this is about demand or about Supply you can have theoretical model that are going to back up both kind of assumptions will say
That against Shapiro claim mainly in favor of the demon site and Demon May matter a little bit but we have more and more evidence on Fox historically now on slair that at the end of the day what matters is also owners so Choice driven by the supply site this paper we
Consider like a third driver of bias which are journalists and we measure whether or not journalists have agency within their organization in fact there are two old the iCal papers that we making this assumption never brought it to the data so we are like the first one to do that
And we think that this is important because it also allow us to better understand the mechanisms through which owners May bias the news okay I don’t know whether you have questions about that or okay so this is the the run map for for what I’m going to
Do first going to present you the data then we will do the first part let’s say when we show the relative importance of Channel fixed effect journalist fixed effect and sorting and then we will turn uh to the B takeover as a case study of what
Happen when we have a change uh in uh in ownership so in terms of data we have this new data set that covers 20 uh TV channels and radio stations from 2002 to 2020 so basically the idea was to take all the main generalist TV channels and
Radio stations for example what we won’t have is a music only radio stations that won be included uh for all of this channel uh we have their content so in terms of coverage we have nearly everything uh with at least uh one guest and one host so we will have news talk
Shows infotainment documentaries with guests Etc what is not included here is fiction games and Sport okay if we if we have no guest or no host basically this won’t be introduced in the data if you have at least one guest or one Host this
Will be part of the of the of the data so this is the overall sample we use in particular for descriptive statistics for identification we will mainly focus on 2005 2019 we do that for two reason first of all a lot of channels enter into the French market that were
Completely changed in 2005 so basically in 2005 you have a huge change in the competitiveness of the market with the entry of many different TV channels which make it like strange to compare uh the pre2 2005 to the post 2005 period are as if they were like equivalent
Market the reason to stop in 2019 is not such a good reason it’s rather a bad reason the fact that the data we use here is data that is collected mainly from The Institute National Audi visual so the national Audi visual Institute which is the repository of the French TV
And radio and where Nicola arve is a computer science and basically the way the data is built uh that it’s entirely documented manually by archist who watch all the shows and we document the number of gu the number of O the topic of the show this kind of stuff and in
2019 uh they decid they decided to save some money and they fired some activist some archivist some activist too perhaps I don’t know uh and so they reduced the stuff and the number of channels became no longer to be documented so at least we have this full complete exhaustive
Sample for 14 year covering 2005 2019 okay for this data we have uh the following information we have information about the host which is pretty detailed so this is not only about like the the main host this is also about like all the uh segments of
The shows so like think about like the morning newscast that can last for one hour and a half you know you have the the main encor and then you have journalist interviewing people we will have all this information minute by minute so we have the amount of time in
Fact spent by each of the OST at the end of the day we have 39 distinct OST uh we will use 13,000 of them in the estimation sample given that the estimation sample we will only keep host who interviewed at least twice a politically classified guest okay 46% of
Them appear on distinct Outlets uh and 68,000 per of them appear on the same network in two different year Time season this is important for us because we need to have enough movers so 46,000 are going to move from one channel to to the other over our 20 years time period but we
Also need them to appear more than once because if they’re just here for like one or two years then their effect will be completely like absorbed by the by the time fixed effect uh in terms of guest so we have information on all the guests by the way whether or not they’re
In the studio or like during some doing some press conferences or this kind of stuff we have two 160,000 distinct guest for more than 2.3 million apparences of guest obviously like all the guests do not have the same probability to appear and we will have some like top guests one of
The most visible if I take the five most visible guest in the in all the shows in data they either either uh uh president or like a prime minister uh so 25% of the guest in terms of professions are politicians we have like uh 25% also for the media publishing industry 13% from
The entertainment sector we have academic and expert sport etc etc okay so then what do we do from that we want to classify the guest and to determine their political leaning if any and I might repeat it but our measures of political leaning of each of the guests
Obviously do vary over time this is true for politicians that can move from one party to the other I don’t know whether like this often happen in German politics except when parties split which happens apparently but this happen often like in French politics okay they can move from
One party to the other and the other thing that it’s not because you are like an economics professor and you decide to be involved in in politics in 2015 that you will still be involved in politics in 2022 if you stop doing politics in between okay so we are
Really going like to check at a very fine grain level your probability to be involved in politics and for which kind of political parties I’m going to enter into the details before that just look at kind of something looking like the French political landscape so it’s a
Little bit of a mess and it’s already outdated compared to the time when we first did this plot uh because parties they keep changing names we really decided to tie our end in the way we classify the parties from the left to the right so we rely on the Chapel Hill
Expert survey and we classify the different political group from the radical left to the radical right okay how do we proceed uh the simple part of what we did was to classify the politicians so for that we first use a candidate list at election so we took
Like since 200 uh 2000 yes all the candidate at national elections so presidential election legislative elections then we look at European election senate election Regional elections Municipal elections Etc then we look at all the members of parliamentary groups and we also look at government members uh this is not so
Easy in a sense so this is pretty easy basically if you are called Kam because we do not have like 16 million Kam in French politics okay if you are called Christian marttin it’s a little bit more complicated or Jean marttin because we tends to have a lot of Jean Martin in
France okay and it’s not because you have one jean Martin who run for a municipal election from the Socialist Party that you want all the Jean Martin showing up on TV uh to be classified as left wing so the second thing that we did uh is that we did after the F name
Matching a lot of manual check and in particular we checked manually all the guest in our sample would change from one party to the other to check whether this was like real change or whether this was due to the fact that these are like two different Johan Martin okay at
The end of the day we classify 8,900 politicians accounting for more than six uh 100,000 apparences regarding the Pinups we proceeded in three ways so it’s not always easy for people who are not politicians to determine their political bias and whether or not they’re politically engaged so we decided to use
Uh three different things the first thing that we use are the party summer meeting participants so you have this big uh summer meeting every year for all the main political parties in France so we look since 2000 at all the guests invited during these meetings and basically if you
Participate in one of these meetings at a given point of time you will be Associated uh to the political parties organizing the meeting the second thing that we look at is that we first established a a list of all the French s tanks then in particular by using their
Funding we um uh decided the political leaning of the Sy tanks then we look at the contributors of the different Sy tongs and then if you contribute to a leftwing Sy Tong you will be like classify on the left to a rightwing Sy Tong classify on the right
And the last thing that we look at is uh the the list of uh signatories of oped uning candidates in the first one of the presidential elections okay so if you appear in one of these three categories you will be linked uh to the related political parties but
Again these are punctual events so we are going to combine together all these punctual events in the probabilistic model so we decided to have kind of like a a sharp Decay before so like just imagine you appear in a like summer University of a party okay your
Probability to be from this party won’t be zero the day before but like the the shape of the distribution will decline very very quick so for a couple of months you will like a lower lower lower probability to be attached to this party and then you will have a long tail after
That so for a couple of months with a decreasing probability you will be attached to this party obviously if I see you like to the same summer universities every year you intervene in all the party uh uh summer meetings plus you contribute to the S tunks every
Month then your probability will be one all the time okay but if not it’s really going to be uh time varying and this is how we are going to classify uh the the different uh uh Pinups in our in our data set okay so I don’t know whether this is clear or not
Okay one example just that I I like to give because it really see a relevance not only of the approach from a technical point of view but also of changing the regulation uh in the last presidential elections the candidate who ranked fourth at the end of the day Eric
Zur is from like a new right party the guy went up to 18% of the pool he did pretty well uh he was a pendit before the last presidential election in France they took place in May uh 2022 until September 2021 the guy was on TV on one of the Bol
Channel I’m going to come back to that but he was on TV uh one hour a day and by arom the French regulator it was not not classified politically if we look at all data even the guy participated twice in the summer universities of the radical right party
It was classified already at radical right so we did the change before arome in a sense that only decided in October 2021 to consider the guy as a politicians because he say at the time that he will run for the presidential uh elections so you see the relevance you
Know of not only like focusing on official candidates when you want to understand political bias on media outlets at the end of the day or perhaps let me show that to you in a graphical way you know in the Ina data we know the profession of the guest so we know
Whether they’re politicians or not politicians we classify politically uh 92% poit of the politicians as politician okay those we do not is really link to the fact that the information on politician is time invariant the information on the profession is time invariant in the inad DAT so if you take today’s Minister of
Justice in France okay now he classified as politican as a Prof as a profession but before that he was just a lawyer okay so when we see him on TV as a lawyer in order we do not classify himl politically but now we do okay and the
Second thing that among those who are not politicians we classify around 5% of the guest from a political point of view and 3% of them as uh as Pinups okay one of the thing that I also want tolight over time is that if you look at
The speaking time share of peot over the total speaking time share uh in the in the data it more than doubled during our time period so basically it was below 10% now it’s around 20% so if you do not look at Pinups you really miss part of the story this is
Really linked to the fact that according to us Pinups are used by the channel as a way uh to escape as a way to escape regulation okay last point that I want to highlight it’s a little bit technical but I think it’s of importance so the so
We measure the political leaning of the Channel with the speaking time given to each of the guests and we we can do that uh because we have the length of each of the show we also have the number of guests but I am going to assume that we
Are treated equally so just imagine Kam you have with Martin in the show and perhaps K you speak 90% of the time and Martin only 10% if I look at the data I would say okay you had 13 minute and had 13 minut okay so is it a good proxy or
Not to know whether this is the case we use like recent data so for one year of data uh we have information not only on the speaking time share we also use information uh based on face recognition so how long I see your face on TV okay
And so when you look at the correlation between the Our Guest time share and the share of your face on TV you know we have like a very strong positive relationship with a slope of s so this is not a perfect uh test in the sense
That I can see your face what Kami is speaking and like oh she’s speaking too much uh but still you know it mean that what we have is kind of a good proxy of what is actually happening during the show if if I am on the kavit side one of
The other Cavit we have is that I don’t know how the host is treating you and whether he’s nice with you or not nice with you with a lot of interruption this kind of stuff but on average we think that the importance you give to some
Political parties on the show is a good proxy of the the the importance of of the bias you have in favor of this political party okay so if I first begin with some descriptive statistics so this is the overall overall time share of political groups uh in France what you see here
Basically that you have a lot of variation links to the political cycle so the right was in power between 2002 in 2012 and so you have the blue line when there had around 45% of the political time share then the left won in 2012 and 2017 they had like most political time
Share and then you know the liberal is maon won in 2017 and then then their time share increased this is partly due to regulation so regulation in France I can answer more questions if you want but in a nut Shield it’s works as follow for the political time share you need to devote
Oneir of the total political time share to members of the government so this is the official government time share and the remaining two3 to all political parties including political party in power depending on their importance but the thing is that importance is not well defined they say Okay importance depend
On pools past political uh results contribution to public debate and so there is a lot of agency led to the channel to shoose the the one third of the time given to government it matters you see that if you move from that to this one when we exclude the government
Members then you see that you know we lose a lot of the political uh uh variation and in particular that there is not a lot of difference between the political time given to the left and to the uh to the radical uh to the radical right we have this strong regulation but
It is strong and not so strong at the same time that you can see if you look like descriptively at the different channels so here we just rank the channels uh according to the overall time speaking time share devoted to leftwing parties okay if you take a TV
Channel like LC for example on average from 2000 to 2002 to 2020 they devoted 40% of the speaking time share to the left if you take France culture we are around 65% so the difference is 25 percentage Point okay the same for each party so you know that
Despite the existence of a regulation you still have lot of variations from one channel to the other and this is what we try to explain in the rest of the paper what explained this differences is it something due to composition effect so different channels have different host with different taste
Uh is it just due to compliance so channels have strong guidelines and journalists they follow these guidelines okay or do we see some sorting that should amplify the previous two uh the previous two effect okay any questions or SC okay so how do you do that uh I’m going to turn to the
Empiric in a minute I just want to give you the intuition of what we do before the intuition is the following we have a host the host is going to move from a channel that has 40% left speaking time on average to a channel that have 50 so the question is that following
This move is he going to devote more speaking time to leftwing guest okay so to do so we are going to look at the uh uh gap between the speaking time share of the left between the destination Channel and the origin Channel okay if the Mover invite similar guests after the
Move then we’ll have a zero slop and it really mean that what matter are the preferences of the journalist if the Mover fully adapts to the new channel guidelines then the slope should be one it mean that the journalist has basically no agency okay if you look at
The road data then we will do it properly with a lot of like fixed effect controlling for like year fixed Effect season fixed effect but just the road dat to get a sense of what is happening whether or not you look at the leftwing Pary time share or the right wi Pary
Time share you see that the slope is around 60% so just descriptively the channel level decision explained around 60% of the variation in Channel level representation of political groups so moving us adapt to the guidelines set by the channels yeah so you mentioned that politicians change parties journalists may change
Their views and in response leave the channel and uh are you able to identify that because I think you put this now all in kind of Channel compliance but in a way that particular host changes the composition might be feels constrained his views have evolved and as a response
Changes channel in that sense the move from uh changing the composition reflects then his or her changes in political views so that I would show you how we how we deal with that uh in the in the in the in the in the identification strategy the assumption
That we need on the exogeneity of the move uh in uh in now in a sense if you want then I think it would be like easier to show that to you with the specification that we have so basically for those of you who did some labor economics going to
Say Ah that’s nice you’re just doing an Akim and that’s true we are doing something that can be equivalent to a Time varying akm model that is used by labor Economist since a number of years in particular to understand uh the uh uh uh um role play play by firms and
Workers uh in productivity and wage so we are going to adapt this kind of framework to understand the determinance of media pet so in a nut Shield what do we have on the left hand side we have the speaking time share of a given political group in the shows h by host I
On Outlet C at time T okay and what do we have on the right hand side so we’ll have the host component so o fixed effect that’s true that one of the limit that we have but I will come back to that that we do not have time varying o
Fixed effect but what we do have and this is very important is that the the channel component is time varying okay we like two seasons Channel effects and this is important because the editorial line of the channel M may change over time like to begin with because you
Might have some ownership uh change okay and we have time component in the time component we put a lot of stuff uh we have day fixed effect so this matters a lot okay if a day fixed effect date fixed effect to be more uh specific uh
Because from one day to the other okay depending on what is happening on the news you might invite moreas from the right from the rical right from the left and you don’t want to capture that in all data and we interact it with our uh we do interact it with our because we
Want to control for the audience and this is not the same thing to invite a guest at uh noon uh 8:00 pm or midnight okay so we don’t have the audience a show by show we have that a little bit we will play a little bit at
The end with that for TV we don’t have that for radio but and and then on top of that the host and guest might influence audience but we have the average audience on TV and on radio uh depending on the hour of the day so we
Will uh control for we will control for that okay so given that what we have our uh identification assumption will be that us moves are as good as random conditional on the o fixed effect which are time invariant on the two season Channel fixed effect and on this
Date hours times platform so radio or TV uh fixed effect okay uh this will be based on the fact that we use host that are observed on multiple different channels okay but that’s true that uh if with from one season to the other like a OST move uh
Because of a change of of his own preferences that we won’t be uh able to capture it because we assume that the political preferences of the host uh do not vary over time to be totally honest in this kind of model given that we allow Channel fixed effect to vary over
Time I don’t think that we can allow both Channel fixed effect and host fixed effect to uh to vary over time if you want to identify everything anything but that’s strong assumption on which we rely yeah yes so this is also on that so we are so the nice thing that we have
And this is a huge progress in the recent literature so before if you were looking at akm literature basically you you the both the host fixed effect and the channel fixed effect where time invariant here what we do that we allow Channel fix effect to uh vary over time
Which allow us to identify not only on movers but only on stayers because then even that we have this time varing chel fixed effect as long as uh H stay for more than two seasons we can uh use it uh to identify uh the uh role played by the different jour
In terms of variance okay uh so uh just brief summary of this table so before I was telling you if you take all the O who have 39,000 o in our sample if we limit to those that have at least two political guest we are done to 13,000
This is L like the drob between that is that these are journalists who often appear when we look at Host we look at all the host so if you have I don’t know like tayor Swift in the US one day deciding to host the Super Bowl okay I
Guess this is only popular program I know in the US she will appear in the data as the host but she’s not a journalist okay so when we will move from the set of host to the set of like host uh we can use for identification we’re going to lose all the
Tor Swift so okay because I guess she’s the most famous singer in the world should be able to pronounce her name properly uh and so this is where we see the decrease in the size of the of the sample then there are like two categories of O that we can use
Zo who appear in two distinct twoyear season so these are the stayers but we need them to stay enough to be uh used in the identification and we use also the the Mover so these are 6,000 so at the end of the day we can estimate with 6,000 mover and 8,000 different stayers
Okay here you have characteristics there is one thing I want to light here because this is very important is in terms of their political characteristics as measure by the time the devote to the radical left to the greens to the left to the Liberals to the right if you
Compare all the host or the estimation sample or those that are movers or stayers you don’t see differences that are statistically significant okay so this is not as if stayers are different preferences than movers because if it were to be the case then we would have a huge issue in terms of identification
We don’t see such a difference in the in the data here okay okay uh so then we perform our variance de composition exercise so to see the part of the difference in bias between channels that is due to channel OST or sorting of OST across uh Channel
Okay and so we will estimate then the the following uh the following model let me go okay now that I want to show that to you quickly through the variance de composition I just want to first show you the result of an event study on the
Move uh this might be linked also to your previous concern here this is not a perfect answer but we see some something here one of the thing we wanted to see is whether host tended to change their invitation patterns before the move so you can do that either to signal
To the new owner that hey I’m to the right you should invite me or perhaps is because you are moving to the right so you invite more and more right-wing guest and then you decide that it might be easier for you to like move to a rightwing channel okay so what we look
At here that we look at before the move and after the move uh whether um uh uh OST fit the destination editorial line and basically we see nothing no difference before the move and then a jump to uh toward the invitation patterns of the new channels just following the just
Following the move okay this is not a perfect test that your concern is not uh uh is not a concern but at least it’s reinsuring as to see that you do not have like three Trends in in the behavior of OST uh before the change uh
Channel okay when we turn to the proper variant de composition exercise how should you read this table you should focus on this three line sh of variant here year and year what do we have and that if you look you can do that for the left the right the radical
Right you won’t have a lot of difference the share of the variance that is explained by the channel fixed effect is around 90% a little bit more because here we have the entire time period the OST effect are very low okay between like 2.7 and four 4% and you have a
Little bit of sorting okay so the majority of the of the of the effect is really driven by channel here I’m not yet showing you that this is driven by owner’s preferences because the editorial line can just like uh uh uh reflect the taste of the audience but
What you will see with the example of B takeover is that even when you have like a change in editorial life not linked to change in demand you see this importance of uh the uh uh Channel fixed effect which in this case like reflect the owner preferences okay uh just one more
Thing I want to uh alight uh the first one is you know you can ask whether all the journalist have the same agency or whether some journalist are higher probability to deviate uh with respect to the channel editorial uh line what you see here uh is that the demographics the fact that uh
Looking at female they tends to invite overall more leftwing guest to what will be predicted by the channel they they work on okay this something we also find in our voting behavior on uh on average uh one more thing I want tolight or perhaps we see
That even better here this is a relative deviation with respect uh to the time s here this is absolute deviation with respect from the chel line one of the things that is of Interest here like two things okay you have a lot of stuff but
I want to light two things uh this one and this one the first one you did not ask me but from time to time when I present that people say okay that’s nice but at the end of the day do the journalist pick their host isn’t it something that
Is more about the producer it turns out that if you look at the main journalist in front the most famous One are both journalist and producers of their own shows okay they have the two jobs at the same time and one of the thing that you
See here and we will have this finding again when looking at B is that the producers they are the one that have the highest probability to deviate with respect to the chinel line compared to the other journalist okay because in the S they’re like more important have more
Weight and they should have more bargaining power we also find the same thing if we look at as a Leo we key entry what does it mean uh for each of the journalists we look at whether they have a page on Wikipedia and Libo is the French equivalent of the W is who
Basically they if they famous they have higher probability to have a Lio or Wikipedia entry and we also see that if you are like more famous as measured by that you have a higher probability uh to deviate from your channel line and I’m going to come back to that looking at B
Because it really mean that when you have a change for example takeover of a given Channel you don’t have like all the journalists are are not subject the same way to the change in editorial line so it means that some factors can protect better let’s say the independence of the journalist and you
Might want to generalize these factors like by LW rather than just like letting the like the journalist characteristics uh deter determining whether they can work uh in overall in terms of Independence the second thing that I want to highight and this is kind of striking and this is also linked to the
Fact that you have a higher chair of the uh variant that is explained today by the channels that we see really a polarization of the channel effects I guess this is really not specific to France like people like okay we still have to do that for the us but my guess
Would be that we will have the same thing in the in the US just look here you know in the with Square you have the channel fixed effect in 20057 with the diamond the channel fixed effect in 20179 if you look at the square the
Minimum we add uh in 2007 was a u RMC at minus one and the maximum it was ER RT at 0.07 now look at the size of the channel fixed effect we have now from minus5 to+ 15 so we have like much more polarization of channel uh of Channel effects and
This might also have uh consequences uh uh over time okay so for the uh last 15 minutes I I want to show you evidence on the B takeover so just in a in a nut chill this like B I can talk about B for
Five hours in a row so if you have any question about v b you should ask me so B is really seen as French Murdoch now INF France is both a leader in terms of media that we will focus on that he’s also the the leader of the publishing
Industry as of today to give you an idea of the importance of the guy in the in the in the public debate uh in 2015 it took over like the canal plus group with four three different TV channels Canal plus d it they changed their name so basically decided to
Rename everything with a C so d8 became C8 it became c as like in Canal it’s why we have this like two names uh more seriously like people realize that there were like a change in editorial line so part of what we do is to quantify
Something that we know we knew from a from a qualitative point of view uh uh they were the one of the longest at the time they were the longer strike uh in uh private media history in 2016 following the ech cover I say at the time because the longer one took place
This summer when B took control of a newspaper called J so the strike was even longer this time than in 2016 but it was as inefficient in terms of protecting the independence of journalist uh and then something that is well documented again from a qualitative
Point of view uh the fact that he played a pretty important role uh in the success of uh Eric Zur so you know the guy who was not classified as a politician during the last presidential election in in France okay so what do we do uh uh with our data to properly
Quantify the effect of the Takeover we do like a simple difference in difference so we are going to compare channel before after the Takeover in our control group we’ll have all the channels whose ownership structure did not change the treaty group will have all the channels acquired by Van
For the sake of uh Clarity and transparency to understand what happens we do that both in the short term 2015 17 and Midterm 20179 okay uh and we control so we have two different estimation strategy in the first one we control for channel fixed effect time fixed effect and the number
Of like different characteristics in the second specification we go a little bit further rather than having the channel fixed effect we have Channel o fixed effect so that we identify on H who stay so then we just only look in fact as a change in editorial line we look at the
Change in invitation pattern of the host who stay to see whether or not they do comply with the new editorial line just to be completely reing this channel they were not on the right before and there were no Trends towards more radical right guest before the Takeover the
Takeover took place in 2015 so in this Gray Line you see like no difference with other channels before and then a jump in the speaking time of the radical right following the the Takeover to be totally honest if anything if you look at our data ital was more on the left
Hand side of the political Spectrum before the Takeover and then he shift uh toward the radical right uh in terms of uh the the the the the the magnitude of the of the effect here if you look at the speaking time of the radical right
You had in the short term an increase by 1 68% in the medium run an increase by 4.46% of the speaking time of the radical right compared to a base line that was 8.6 percentage point so basically you have an increase by 50% of the speaking time of the radical right overall and
This come mainly at the expense of the right also a little bit of the radical left green and left but the effect is not statistically significant so this is overall with a channel fixed effect now if you look at compliance focus on the host who stayed
On the channel you see the same increase in the speaking time of the radical right and then you really see that it comes mostly at the expense of the radical left who got like less invitation uh and the the for for the right the change is no longer statistically uh significant okay this
Was expecting in a sense what is really new and to the extent of knowledge as never been done in the literature is to look at the kind of guest host who stayed and the kind of host who lived and also to try to understand where did they go following the uh the the
Takeover so the first thing that we did is something pretty simple in fact uh we did the same thing as before but on the left hand side uh we look at an indicator variable whether a journalist who was there during the previous quarter is still working on the channel
Uh during the following uh quarter so the probability of like like a turnover following the Takeover versus the probability of turnover on the other channels what do you find uh here you find the following basically again this is pretty flat before the Takeover so the probability from moving from one
Channel to the other was not higher on the channels acquired by B before the the Takeover okay were just behaving as the other channel but then you see that the probability of no longer working on the channel like dropped by 20 percentage Point compared to the other channel following the Takeover here if
If we just like take like all the all the host if we wait the host by their screening time uh by their screen time share we see that the effect is even more important okay and so you have this drop by nearly like 40% in the probability of uh leaving uh
Uh leaving the the the the the channel uh just to give you uh again an order of magnitude so you have this drop in the medium run by 133% of the probability of leaving the channel the the base line is 38.3 so you have an increase by one3 on
Average of the probability ofing leaving the channel due to Bol takeover then what is nice in what we can do that we can look at this probability of leaving the channel but we can also look at the probability of uh of at the characteristics of the guest will live
Like whether those are guest with more political host with more political guest whether those are journalist whether they were in charge of a newscast whether you have an higher probability for the male or female journalist whether they’re famous as measured by their uh uh um um who is entry uh look
At their ratings also this kind of stuff how do you okay if I want to summarize this table in one minute basically we see that the the the guest that are more likely to leave are those that are politically classified guests those that are journalist and those who show the newscast I really
Want tolight that it mean that this is not only a about a b taking control of a firm and firing people randomly this might happen in lot in a lot of firms okay and you might tell me okay the good Baseline is not to compare B channel to
Other channel because on other channel you have no change in ownership and when you have a new owners okay in any firms except University but we are not owned by anyone so that’s better uh you can fire people and have a lot of like turnover but what is striking here that
This is not random at all because those who live with the highest probability are those that were like more engaged with political programs have a news cast of our journalist or do invite political guests so it does not fire basically if I want to caricature a little bit the
One that was in charge of like showing Meto okay but you fire the guy that was in charge of like a newscast and this appear very clearly uh in the data uh in the data year so then the question is where do they go and what you see here
Which is a little bit depressing is that many OST simply simply quit journalism so to be more specific is a quit or sample but our sample include all the main TV and radio channels so they could have go and work for a newspapers for example but if I were to
Show you data on the evolution of the number of job in the journalist uh of journalist in the newspaper industry you will understand that this is not what happened to the majority of them and then we also have a lot of qualitative evidence uh so we have nearly an increase of like
30% of those of those journalist who just quit journalism which is bad per se I have to say because it mean that overall we have like less people working as journalist the other thing that we can do uh and I am going to uh uh finish
With that we we have those uh uh that stay in journalism we can look at the characteristics of the channel the join and in particular we can look at the political bias of the channel That join if you do that for example looking at right wing channel uh here you have uh
Those that are in the q1 is the bottom quartile of the channels in terms of rightwing invitation see that the lot of them move to this Channel with a low ratio of rightwing guest if you look at Q2 or the third quartile then you have no effect so basically what you show
That those who left but who are still journalist they tends to go mainly on Channels with lower speaking time for the right so basically those who don’t want to comply they don’t want to comply because they were not good in terms of sorting and so they join journalist that
Uh they join media Outlets that corespond better to their own uh to their own preferences okay just one word in terms of uh policy uh implication uh okay so we have a lot of work empirical evidence I mentioned some papers on Fox News Sinclair that show
That when you have a change in ownership you can have a change in media content uh and that this might also impact voters Behavior okay what we do and what is New Year is really to understand the mechanism through which uh it uh it happen what we cannot really
Do as is done for example in the deavia paper deavia Capon paper we cannot really look directly at the effect on voters so for example we have suggestive evidence I can document we can talk about that we try to convince the readers in the paper uh that the shift
Toward the radical right of B Channel it helps a lot but it’s hard to put that into the data because we do not have heterogeneity in the penetration of Senus it’s kind of an homogeneous shock over the French territory so we don’t have enough variation still thanks to
Richard we have the writer data that allow us to have like some qualitative evidence so what writers Institute does in the in the survey data is basically you have information on the news outlets consumed by the citizens and they also ask people to rank thems on um zero to
10 uh left wing scale so basically when you are at zero you are more on the left when you are at 10 you are more on the right and here I take the average depending on the kind of media Outlets you claim you consume okay this is just
Suggestive but I think it’s of interest if you look at sen new before the Takeover so I have data of 2013 we see that the rank on the left compared to the average population uh that is surve in the data okay these are below the average which is this
Dashed uh black line if you look at 2018 you see now they’re a little bit above if you look at 2019 they are even more above and if you look at 2020 the effect is even stronger so of course I’m not claiming that this move to the right is due to the same
Viewers watching Fox new watching Fox News watching sen news would change their mind because this is not panel data okay this is a repeated cross-section so this might also be due to the fact that you have new viewers watching the news but still this is interesting evidence of the fact that
You have also a shift in the characteristics of the viewers in terms of their uh political preferences the last thing I want to highlight because this is important and this is also linked to to that you might say just that okay B is a clever guy FR is turning more toward the radical
Right the guy wanted to make more money so he shift uh the editorial line of the channel to the right because he thought it was a way to acquire more audience what we do document here and I put that overall you see more of an effect in
Fact if you look like Channel by channel in particular for Canal plus what we see that following the Takeover compared to the audience of other channels you have a drop in audience so the shift to the right came at an audience cost so this is not something that is driven by o by
Demand or something that was like lead to a huge political uh positive shock in terms of audience people were so happy to see so many radical right G that they all shift to B Channel we see kind of the reverse in the data again there are
Limits in what we do I cannot claim that it mean that it was not profitable in particular because you have the issue of revenues and the issue of cost for revenues we know that revenues collapse because like Drop in audience was associated with a drop in advertising
Revenues we have that in the paper I don’t have time to present that to you here he also decided to do less news more talk shows so basically less journalist that that I show you so reduce on cost so at the end of the day in terms of overall profitability
Perhaps it was a little bit more uh complicated so uh in fact I am uh I am done uh uh so we are in this context of like growing concentration and polarization of the media uh with a lot of change in ownership here we want to understand the determinance of media
Bias we saw the importance of the role played by Channel fixed effect with the case of B we also see that a lot of these Channel fixed effect are driven by the preferences of the uh of the owners uh that uh host they need to comply with
The editorial line they do not have a lot of agency even if some of them have depending on their characteristics uh uh and we do that thanks to this uh Noel data that we collect and put together as well as to the use of this time varing model and
Now I see my coor that wants me to end so I’m ending here thanks a [Applause] [Applause] Lot