From agenda-setting to furtive manipulation: The role of (social) media in political polarization
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
| Title |
| |
| Title of Series | ||
| Number of Parts | 8 | |
| Author | ||
| License | CC Attribution - NonCommercial - ShareAlike 3.0 Germany: You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal and non-commercial purpose as long as the work is attributed to the author in the manner specified by the author or licensor and the work or content is shared also in adapted form only under the conditions of this | |
| Identifiers | 10.5446/54795 (DOI) | |
| Publisher | ||
| Release Date | ||
| Language |
Content Metadata
| Subject Area | ||
| Genre | ||
| Abstract |
|
3
00:00
Computer animation
09:57
Computer animation
19:50
Computer animation
26:58
Computer animation
29:19
Computer animation
36:55
Computer animation
44:31
Computer animation
Transcript: English(auto-generated)
00:04
What I want to talk about today is social media and the effects on polarization. And I'm going to depart from two assumptions. The first one is that democracy is currently
00:21
in retreat worldwide. This is a map by the V-Dem Institute for Democracy in Sweden that looked at the trends between 2009 and 2019 and discovered that when you look around the globe, there is a number of countries where democracy is in decline.
00:42
Those are the orange countries, including some in Europe and concerningly, including the United States. And there are fewer countries in which democracy has been increasing. These are the same data yet again shown at the level of individual countries and comparing 2009 to 2019.
01:07
And any point that is below the diagonal is representing a decline in democratic health during that 10-year period. And if you look around that space below the diagonal,
01:22
then you find a lot of countries that are close to us here in Germany or that are very important to us, like the United States. Now, there are some other countries where things have improved over time, but by and large, in key countries
01:40
for us in the Western world, I would argue that there has been a decline in the health of democracy. Equally, there has been increasing polarization, again, in certain key countries. These are the data for the United States in a recently published report by Boxell et al.
02:04
And you can see how affective polarization has been increasing over the last 30 or 40 years. Affective polarization is the difference in feelings of warmth towards people from your own party versus the other party.
02:22
And on a scale from 0 to 100, the difference between the parties is now around 45, which is a lot. I mean, that means if you're a Democrat, you really don't like Republicans and vice versa. In fact, there are other data suggesting that political polarization, affective polarization
02:43
in the United States is greater for politics than it is for race. People are more divided now in America along party lines than they are along race. Now, an interesting aspect of this polarization is that at least at the level of political leadership,
03:02
it is not symmetrical. What I'm showing you here are data by Hare and Poole that was published some time ago, where they did a very nice statistical analysis of the relative positions of the two parties in the United States over time. In fact, going back to 1879.
03:23
So we're talking about 120 years of history here. And what you can see is that since about the 1970s, the Republicans have become more extreme or have migrated further away from the center than the Democrats have, who are relatively stable.
03:40
So I think it's important to keep that in mind that just because there's polarization doesn't necessarily mean it is symmetrical and that people walk away from the center equally. It could also be that one side is moving more than the other. Now, I have to add an important qualification
04:01
to this, which is that polarization is not increasing everywhere in the same way that democracy is not failing everywhere. For example, these are the same data from Boxall et al. In Germany, affective polarization, according to these data, has been decreasing.
04:21
The same is true for Sweden, Norway. It's flat in Australia. Unlike Australia, Canada, and Switzerland, where it has been, I think I said Australia, but it's obviously the US, Canada, and Switzerland where it's been increasing. So the pattern is heterogeneous.
04:43
But I think that if we focus on countries such as the United States that are very important to the world, then it is very clear that democracy is in retreat and polarization is increasing. And in Europe, within the European Union,
05:01
there's at least one country that by most accounts is no longer considered to be a democracy. That's Hungary. And yet, it remains a member of the European Union. So those are my departure points for what I want to talk about. So are social media to blame for this?
05:23
Well, a lot of people say yes, some other people say no. There's a lot of debate about it. And to me, the important thing to start out with is to acknowledge that there really is no binary answer for this.
05:40
It's not a yes or a no. Instead, we have to break the problem down. We have to say, well, what is it that social media might be responsible for, and how? And how would we even know that it is social media? So we can achieve partial answers to this, I think,
06:02
but not an overall, I don't want to come to an overall conclusion that says yes or no today. What I want to do instead is I want to focus on these three issues, agenda-setting power, micro-targeting, and then the whole notion of establishing causality in all the research we do
06:22
and how we might do that. So here we go, first issue, political agenda-setting. What do I mean by that? Well, the conventional wisdom, and by that I mean 20 or 30 years of research in political science,
06:41
the conventional wisdom is that the media are the principal agent of agenda-setting and politics. It's not actually the politicians, according to the conventional view. It is mainly the media which collectively set the political agenda. And yes, the politicians can then influence that,
07:02
but typically they themselves are insufficient to set the agenda, and just to illustrate why we know this, here's a couple of examples. One causal effect was shown in an experiment published in Science a few years ago
07:21
where they actually, the experimenters randomly launched topics in local media in the United States and then observed on Twitter what happened, and guess what? Those topics were picked up by people in general suggesting that the political agenda was set by the media.
07:41
Likewise, the New York Times coverage of terrorism leads to more terrorist attacks. You can show that, and it is a very elegant study using an instrumental variable paradigm. But enter Trump and Twitter. And as I would argue, everything is now different,
08:04
and the conventional wisdom is no longer supported. Now, I don't know if you remember this. This is five years ago, so that's in the Jurassic period before Trump assumed office, but after he had been elected,
08:20
and no one remembers this now in amidst all the other stuff that's been going on, but Donald Trump got very exercised over the cast of a Broadway play, Hamilton, which after a performance, pleaded with vice president-elect Pence, who happened to be in the audience
08:40
for a diverse and democratic America. I mean, I guess they knew what was coming, so they said, you know, let's preserve American democracy. Donald Trump got very exercised over this, very excited, and went on at considerable length on Twitter to attack, really, that performance and the actors.
09:05
Now, if you look at Google Trends for that time period, for the keywords Trump-Hamilton, you find that there was a massive spike in interest, in public interest, around the time that he tweeted.
09:21
Now, Google Trends data tell us what people are searching for, so this is telling us that people were searching for Trump and Hamilton. The absolute numbers are not available, but the maximum is always expressed as 100%, and everything else is scaled relative to that. Now, why does this matter?
09:40
Well, it matters because at the same time, on the same day that Trump engaged in this Twitter activity, he settled a lawsuit against him over the so-called Trump University for $25 million,
10:01
including a $1 million penalty to the state of New York, so basically admitting culpability in this fraud lawsuit. And the blue line shows public interest in the Trump University settlement. It's about, I don't know, 5% of the interest
10:21
in Trump-Hamilton, so is this coincidence? Did he just tweet something to distract people from news he didn't like? Well, we don't know, but my collaborators and I wanted to examine this possibility more formally. Now, to do this,
10:41
you have to postulate some sort of conceptual model, and this is what that model is. Our presumption was, well, if Donald Trump is diverting the media, then whenever the media covers something that he doesn't like, he's gonna start tweeting about something different. And if that diversion is successful, then maybe the media drops this harmful issue
11:02
or at least reduces it. So basically, it's a trivial conceptual model which you can implement in a regression equation and where you expect one positive and one negative coefficient, and we operationalize this using an event in ancient history known as the Mueller investigation
11:24
involving potential collusion between Donald Trump's campaign and the Russian government. So we postulated furthermore that whenever, and this clearly, by the way, was damaging to Trump, and he knew that. He hated it.
11:41
It was damaging to Trump, so I think no one can say that we picked the wrong negative event for Donald Trump. We then went to his record and his campaign literature and everything else, and we examined what we considered or tried to discover what his political strengths were, and at the time, pre-pandemic,
12:02
it was certainly jobs, the economy. He was going on about this all the time. It was China. He was always an antagonist of China. North Korea, you may remember that Donald Trump has a bigger button than the other guy when it comes to nuclear power, and of course, immigration was his core platform.
12:22
So we expected a lot of activity on those topics that were congenial to Donald Trump's political future, and if that were successful, then maybe the media would drop or reduce coverage of Russia and Mueller. So let me now tell you about a study
12:40
that we conducted some time ago where we scraped all of the New York Times, all of the ABC News statements, the headlines, and all of Donald Trump's tweets for the first two years of his presidency, which is when the Mueller investigation was taken place,
13:02
and we then did some very simple, relatively simple statistics that related Mueller-Russia coverage in the media to Donald Trump tweeting about his presumed scientific strengths, and so here's the first set of regression models.
13:23
Do Trump's tweets divert? Well, you can't see the numbers because they're too small. Don't worry, I have a magnification here. All I want you to take away from this is that the numbers are positive, and they're as significant as indicated by the asterisks. What does that mean?
13:41
Well, that means whenever the media increase Russia-Mueller coverage, Donald Trump increases tweets about jobs, immigration, China, et cetera, et cetera. That's what this means, statistically. Now, does that work? Well, this is the second regression model from our conceptual model,
14:02
where we're relating yesterday's tweets on those diversionary topics to today's coverage. So what happens after he tweets? What does the New York Times do? What does ABC News do? What do they do together? Well, what they do is less.
14:23
Here are three significant, small, but significant coefficients that are all negative, which indicates that the diversion works. The media are dropping Russia-Mueller in response to Donald Trump's diversions. Now, that was the first analysis
14:42
based on pre-selected keywords where we presume to pick up Donald Trump's political strengths, but we pushed this further by examining his entire vocabulary that Donald Trump was tweeting about, excluding things related to the Mueller investigation,
15:04
because that would just contaminate the whole thing. When he's talking about his own investigation, we were not interested in that. We were interested in understanding what he'd be talking about and how much in response to coverage about Russia-Mueller.
15:21
And we also wanted to know what happens if the coverage in the media is neutral, just for comparison purposes. Now, I'm gonna show you a bunch of graphs that summarizes these data quite strikingly, I think. But I can only do that if I explain how we did the plots and that'll take me a minute just to walk you through it.
15:42
So, here we have that first regression model I was talking about. How many times does Donald Trump tweet a random pair of words, call that X, in response to the New York Times or ABC coverage of Russia-Mueller, okay?
16:02
Well, we look at that for each pair of words in his tweets, we estimate that regression coefficient and we plot it on the graph. Well, actually, we plot the T value so we can indicate significance. But effectively, it's just the transformation of the coefficient.
16:21
And then we look at what the media do the next day to Russia-Mueller in response to that very same pair, pair X. And we estimate another coefficient which we plot on the ordinate and bingo, we have a space in which we can present
16:41
each pair of words in his tweet. So, this is pair X, that's just some pair of words but of course, we also have Y and we have said, and in fact, we have a 1400 of these because he tweets about a lot of different things and they come in lots of different pairs. Now, we can formulate some expectations
17:03
about what should happen. Now, if nothing happens, then we should have a blob of points in the middle. And we should observe that either if there's just no effect or we should observe it for neutral items that Donald Trump doesn't care about. There's stuff in life Donald Trump doesn't care about
17:21
and I'll show you in a moment what they are. Now, if on the other hand, something is exciting him, then he should tweet more in response to coverage in the media. More means the blob should move to the right.
17:40
Now, if that is successful, so the media then stop talking about Russia and Mueller, then the blob should move to the bottom as well because that means the media are reporting less the next day in response to the tweets. So, what we're looking for is a blob in the middle
18:00
to indicate nothing, something to the right to indicate Donald Trump getting very excited and something to the bottom right if in fact Donald Trump's diversion is successful and the media reduced their coverage, okay? Here are things Donald Trump doesn't care about much.
18:21
Well, he cares about the economy but it doesn't actually affect his tweeting too much. Well, he tweets a bit more but the media don't pick it up. Football leaves him unfazed. Gardening, well, if anything, it puts him to sleep because he tweets less. There are some points out here on the left.
18:42
Gardening is not his big thing. Skiing is also not his big thing. Nothing happens. So, with neutral terms, you get what you'd expect, the blob in the middle. And just to illustrate, the word clouds down here are the words from those articles in the New York Times on those topics and I put it there
19:00
so you can confirm that we picked the right thing. The skiing stuff really is about skiing. Olympics, mountains, snow. I mean, it's got all the right things in there. What about Russia and Mueller? Well, here we go. This is the New York Times reporting on Russia and Mueller.
19:21
And guess what? There is this point cloud in the southeast which are word pairs that Donald Trump is tweeting about significantly more in response to Russia Mueller coverage and the New York Times the next day reduces its coverage significantly
19:41
in response to those tweets. And here it is for ABC. So, the same thing for both ABC and the New York Times. We get this reduction in coverage the next day. So, to summarize as a caricature, what we find is that whenever the New York Times talks about Russia Mueller using these words over here,
20:04
Donald Trump starts tweeting about that. These are the words from the tweets that are in that southeast corner on the preceding plot. So, he talks about Korea, China, jobs, tax,
20:20
you know, et cetera, Republicans and so on. And in response to that the next day, New York Times talks less about Russia and Mueller. Now, the effect, second effect is actually quite small. So, I'm exaggerating what's going on here but that's simply a caricature visualization
20:41
of the effects that we've observed. So, what does that tell us? Well, it tells us that Donald Trump, while he had access to Twitter, was able to set the political agenda and to influence New York Times and ABC media coverage.
21:01
It appeared that way anyway. Of course, we're not claiming causality in any of this. Gotta be super careful here. I'm not saying that, you know, Donald Trump is telling the New York Times what to do and they obey. No, it's just there's reliable statistical association that is compatible with the idea
21:20
that Donald Trump is indeed setting the political agenda. So, that's one effect of social media that I think is reasonably strong and something worth keeping in mind for future discussion. Now, the second thing I wanna take up is micro-targeting.
21:44
Now, micro-targeting is the idea that you can address persuasive messages to people online based on certain characteristics. Now, one of those characteristics is personality
22:00
and here are data from a paper by Yu Yu et al. in 2015 that showed that if you have access to 300 Facebook likes by a person, you can predict their personality better than their own spouse.
22:20
By the time you get to 300, you outperform the spouse. If you have 200, actually no, less, this one here, 100, you do better than other members of the family and with 10 likes, you're already doing better than work colleagues.
22:41
So, Facebook likes, if you have access to that information, allows you to know a person's personality and once you know a person's personality, you could then perhaps manipulate them better because you understand what their vulnerabilities are. Does this work? What do we know about the effectiveness of micro-targeting?
23:03
Well, I would argue that there's pretty good evidence to suggest that it does work based on this study by Matz et al, which included more than three million participants, so I don't think power, statistical power is much of an issue in this study and what they did was to expose their participants
23:24
on Facebook to cosmetic ads and these were real ads and they actually sold real stuff on Facebook with those ads and the ads were designed to appeal to extroverts or introverts
23:40
and they had a complicated way of validating that. I'll show you some examples in a moment so you can get an idea of how this was done and the audience was also selected to be extroverted or introverted. How? Well, by using their likes. We know what likes an introvert has on average and an extrovert,
24:02
so all we gotta do is target an advertising, an ad to people with that profile of likes, whatever it is. What did the data show? Well, first the stimuli. Which one is what? Well, now that I've primed you, you probably also think that the one on the left
24:22
appeals to extroverts and that one to introverts and indeed, if you do the validation study, then that is what you find. And you sell more stuff if you match the audience to the ad. What I'm showing you here are the conversion rates.
24:41
This is the click through rate from each ad. You can express the same data also in pounds, in pound sterling, which is the amount they actually sold. You get the same result and if you match the ad to the audience, you sell more. So introverted ads are over here. Introverted audience is green.
25:02
Well, you sell more than if you have an extroverted audience. And the reverse is true for the extroverted ads. If you send it to an extroverted audience, you sell more than if you send it to an introverted audience. So targeting works. There's other evidence from other studies
25:21
done in the laboratory that pretty much supports that. The persuasive messages can be tailored to a person's personality. And of course online, if you do that, you have a lot of impact to the point where we have to examine the relationship between Facebook and democracy.
25:45
Now, I just talked about cosmetic ads and I'm pretty confident about what happens with cosmetic ads. What I don't know is what happens with political messages because to my knowledge, no one has done the research
26:00
and we know way too little about what is actually going on on Facebook. But even if we don't know exactly what's going on, I think we can say that micro-targeted political messages are a serious problem for democracy because if only the target
26:20
and the originator know of the existence of the message, then a political opponent has no opportunity for rebuttal. Hillary Clinton had no idea what was going on on Facebook and what was being said about her by Russian trolls and God knows who else. Well, that is not democracy because democracy relies on a public exchange of ideas,
26:45
a free marketplace of ideas so that people can make a choice. You can't do that with micro-targeted messages. So irrespective of any empirical data on that, I think we have a problem for democracy.
27:02
Those of you who are interested in this sort of fundamental relationship between technology and democracy, this report that I was a lead author on for the European Commission about a year ago is available at that web link and that goes into all these issues in great depth and it also contains policy recommendations incidentally
27:23
about how we might get out of this mess. But that all takes time. The European Union is currently working on a lot of legislative proposals. I know that because I spent two months in Brussels earlier this year working on those initiatives
27:42
but they will take years to come out. So what do we do in the meantime? Well, one thing we can do that colleagues at the MPI in Berlin including Ralph Hertwig and I worked on, one of the things we can do is to reverse engineer quote unquote micro-targeting
28:02
by telling people something about themselves that sensitizes them to being targeted. That's the basic idea and I'm pretty sure Ralph is gonna talk about that more at length in his keynote tomorrow. So I'm just gonna give you a very brief
28:20
thumbnail sketch of this one study that just came out a few weeks ago where what we did was to get people into the lab so to speak, online of course and we gave them a personality test. We ascertained the extraversion and introversion. Now in one condition we did that first
28:43
and we gave them feedback on their extraversion, introversion. In another condition we administered an irrelevant personality test and we then asked people to classify ads as being targeted at them or not.
29:02
So in other words the only task was for people to say hmm, yeah I think that ad is being, that's me, they're targeting me. Or no, they're not. After we give them feedback on their personality in one condition but not the control.
29:21
And how did we give them feedback? Well we gave them a brief description of what an extrovert is and what an introvert is. Now it was a lot longer than just this of course but that gives you a sense of what we were telling them and they had to read this. We made sure they actually spent time reading this explanation.
29:42
We then provided correct feedback on their introversion, extroversion score. So here's a sort of an extroverted person. They would get a feedback that places them on the 74th percentile and you know blah, blah, blah
30:01
was all explained, this is where you are. A highly introverted person, this would probably be a mathematician, they would be given this feedback here saying 98% of people your age are less introverted than you.
30:22
So that's the feedback. What happened after that? Well this is the task. Is this ad targeted at you, yes or no? That's the only question we asked. Picture the sound bite, yes, no, targeted.
30:41
Now if you're an extrovert what are you gonna say? Well yes, if you're an introvert probably not. Well that's exactly what happened. After people were given personality feedback that's the experimental condition. The modal accuracy was 100% for people in that condition
31:03
and the average was around 90%. Now that is 30 percentage points higher than in the control condition. In the control condition people are above chance. Chance would be 50%, here the average is 60 something and it's above chance statistically
31:22
but it's 30 percentage points below the boosting that you get in the experimental condition. So it's a massive effect and we've replicated this and extended it a few times but the bottom line is that messages are targeted.
31:45
There's no question about that and the targeting has been shown to be effective for commercial ads. I don't know of anything for political messages off the top of my head. We also know that people can learn to detect advertisements that are based on their personality.
32:05
At least for cosmetics, well that's a start and maybe that is a first step towards resilience. What we're doing now is the obvious next thing which is to say well will people be less persuaded to buy cosmetics once they find out
32:24
that they're being targeted. Now with cosmetic ads it's not a ethically complicated issue really, I mean to my mind it doesn't matter whether or not people are persuaded to buy one lipstick or another based on micro-targeting.
32:40
Where it is crucial of course is for political messages because that is where things become ethically far more dubious. Now as an interesting aside, when you ask people, when you ask the public about micro-targeting it turns out that in Germany, the UK and the US
33:03
people uniformly do not like being targeted on the basis of their personality or other inferred characteristics when it comes to politics. We just published that paper a few weeks ago in Humanities and Social Sciences Communication
33:21
and that was clear across all different countries. So the public actually doesn't like being targeted for political messages. Now the final question I wanna address briefly in all of this is well how can we be sure
33:41
or how can we make a causal attribution to any of this? Now in an experiment such as the one I just told you about where we had a control condition and an experimental condition, we can make a causal inference because we are randomly assigning participants
34:01
to one condition or the other. So we have a control condition, experimental condition, an outcome. If there is a difference, we know it is due to our intervention. Now that's sort of causation 101 and it's an oversimplification and it's not the whole story but as a first approximation, that is correct.
34:20
And the key thing here is that the randomization disrupts all other variables that might otherwise be responsible for the outcome. Now in observational studies, we don't do that or we can't do it because what we do
34:40
in observational studies is that we look at naturally occurring variation. So we look at people who have different shoe sizes or different genders or different political opinions and we just sort of look at them and measure that and we relate it to an outcome such as maybe polarization. Well that's okay but that does not allow us
35:02
to make inferences of a causal nature because well lots of reasons. One of them is that the causation could be the other way around, that what we think is an outcome is actually causing some of the variation we're observing or perhaps more likely, there are these nasty hidden variables out there
35:22
that might instead cause the outcome and there's absolutely nothing we can do about that. Well, or is there? Well I would argue that actually there is because there are some ways in which you can come closer to a causal interpretation and the one I just want
35:40
to illustrate is what's called an instrumental variables approach. Now what does that mean? Well it's an observational setting so we have naturally occurring variation and we have an outcome that we're observing and we'd like to know if this causes that without running an experiment, how do we do that?
36:03
Well we can't run the experiment but maybe what we can do is we can find a so-called instrumental variable that induces exogenous variation that makes this thing vary and in so doing, perhaps that's almost like running an experiment. Instead of assigning people to different interventions,
36:22
maybe there's another variable that causes these interventions or makes these interventions happen on our behalf and if this exogenous variation cannot reasonably be related to the outcome directly, well then the only causal path is the one
36:40
that we're trying to establish. Now that's sort of true. I'm giving you a thumbnail sketch here so you can always make it more complicated but I think as a first approximation, that's pretty reasonable. Now here's one example of a study that was published last year that did this
37:02
by Simonov et al. and what they looked at was the effect of Fox News consumption in the United States on compliance with social distancing. So their presumed causal variable
37:23
was how much people watched Fox and their measure was how much people stayed at home compared to baseline pre-pandemic. And the hypothesis was that the more people watch Fox News, the less likely they are to stay home because Fox News tells them that COVID is a hoax,
37:42
effectively, that's the sort of simplified model. Now if you look at the data, then without doing anything else, that's precisely what you find. You know, the more people watch Fox, the less likely they are to comply. Well is that causal or is it because the people who don't like complying with anything are choosing to watch Fox?
38:01
Well we don't know that, except what we can do and this is really clever, I love this. What you can do instead is to say aha, in each of thousands of cable markets in the United States, the position of the channels on the menu
38:21
for your cable TV is quasi-random. So in some places in West Virginia, Fox might be number one. You go to Maryland in some county somewhere, some small market, it might be number 23. And whether it's one or 23 or 15 or 30 or anything in between is sort of random. I mean it's not entirely,
38:41
I don't think anybody's drawing random numbers to do the channel assignment. But you know, there's no systematic relationship of a station to its position on the menu or button on the remote. And this is crucial because the channel position then
39:03
cannot conceivably directly influence social distancing. I mean how would the menu for your cable TV make you stay at home or not on its own? Well no, it doesn't. But what it does do, and there's independent evidence for that, is that the higher up the channel is,
39:22
the more people watch that station. So on top of people's preference for Fox, whatever it may be, there is that added boost that Fox gets if they're high up in the menu as opposed to being lower. And that is that little bit of exogenous variation
39:41
that is added to the naturally occurring variation in people's preferences for Fox. And the moment you do that, you can look at those marginal effects that are due to the exogenous variation, and you can then look at social distancing for that added extra bit of exogenously caused variation.
40:08
And when you find that, you find that Fox is causing people not to stay home. The more people watch Fox, the less likely they are to comply with social distancing.
40:21
And this isn't the only study, there's another one that came out at the same time that showed the same thing, focusing on two different commentators within Fox who either did or did not consider the pandemic to be a hoax and you find the same effect. So that's my way of background
40:42
to give you a flavor of how one can do this. Now, we have an effort forthcoming that will be submitted in hopefully not too long, once I stop going to conferences. And where we did a systematic review of the literature
41:02
on the effects of social media on outcomes relevant to democracy. Now, there's nearly 400 papers on this, but only 25 of them used an approach that allowed us to identify causality and we focused mainly on those.
41:20
Now, the conclusion that we draw from the systematic review is that, and it's a bit nuanced because the difference between where you are in the world, but in established Western democracy, so Europe, the United States, social media use causes increased political polarization
41:41
and also some other adverse effects. And that's true even when users are exposed to opposing views online. So it's not an echo chamber phenomenon. It is simply the use of social media translates causally into increased political polarization.
42:03
Now, this conclusion comes with a caveat, which is that the number of existing studies is small and that causal inferences, even with instrumental variables and so on are not 100%. So there's an escape hatch built into this,
42:22
but I personally find it at least suggestive. And just to illustrate what other things might happen, there's a recent paper that was done here in Germany where they looked at a random outages of Facebook in selected towns in Germany.
42:42
Now, it turns out that in Germany, I guess is elsewhere, occasionally your internet fails, right? You may have noticed that the ISP just falls over for a couple of hours or for a day, Facebook maybe down for maintenance or something. So now this happens presumably at random.
43:01
I mean, I don't think there is a mastermind, a Facebook outage master who's selectively switching off tubing in every Tuesday afternoon. You know, that's not what's happening. It's random, pretty much. Well, it turns out that whenever Facebook is down, there are fewer hate crimes against refugees in Germany.
43:21
It's as simple as that. You turn it off and fewer people get beaten up. That's one of the effects of social media. I'm simplifying a little, but not much. That actually is what they showed. So I think my 45 minutes are up.
43:42
What are my conclusions? What is it that I'm, what are the claims I'm making? Well, I think the first claim is that democracy is in retreat in certain key countries. Polarization is likewise increasing in certain key countries. And in part, that is due to social media
44:06
through processes such as permitting agenda setting, targeted advertising, and indeed, just usage of social media which, for which some causal effects have been established.
44:22
And we can now discuss the nuances or your disagreements with my conclusions. Thank you for your attention.