Sander van der Linden is Professor of Social Psychology in Society in the Department of Psychology at the University of Cambridge and Director of the Cambridge Social Decision-Making Lab. His research interests center around the psychology of human judgment, communication, and decision-making. In particular, he is interested in the influence and persuasion process and how people gain resistance to persuasion (by misinformation) through psychological inoculation. He is also interested in the psychology of fake news, media effects, and belief systems (e.g., conspiracy theories), as well as the emergence of social norms and networks, attitudes and polarization, reasoning about evidence, and the public understanding of risk and uncertainty. In all of this work, he looks at how these factors shape
Episode Description
Social media are complicated. With some research suggesting they're important spaces for digital community building and other scholars pointing out how social media can serve to actually disconnect people from one another. A growing concern among both academics in the public is the ways in which misinformation and conspiracies move through social media networks. That is the focus of this episode of Stats+Stories with guest Sander van der Linden.
+Full Transcript
Rosemary Pennington Social media are complicated. With some research suggesting they're important spaces for digital community building and other scholars pointing out how social media can serve to actually disconnect people from one another. A growing concern among both academics in the public is the ways in which misinformation and conspiracies move through social media networks. That is the focus of this episode of Stats+Stories with guest Sander van der Linden. Sander van der Linden is Professor of Social Psychology in Society in the Department of Psychology at the University of Cambridge and Director of the Cambridge Social Decision-Making Lab. His research interests center around the psychology of human judgment, communication, and decision-making. In particular, he is interested in the influence and persuasion process and how people gain resistance to persuasion (by misinformation) through psychological inoculation. He is also interested in the psychology of fake news, media effects, and belief systems (e.g., conspiracy theories), as well as the emergence of social norms and networks, attitudes and polarization, reasoning about evidence, and the public understanding of risk and uncertainty. In all of this work, he looks at how these factors shape. Sander Thank you for being here today.
Sander van der Linden
My pleasure, great to be on the show.
Pennington
You recently co authored a paper about how conspiracy moves through Twitter. And I wondered if you could talk a bit about sort of what sparked your interest in that particular topic? And maybe even that topic in that space?
Linden
Yeah, well, you know, my interest in conspiracy theories has been long, long standing. Because when I was young, one of my brothers was really enthralled in the whole truth or movement. We had very, very long debates, and we would get our information from very different sources. And, you know, this, this was around the time that, you know, I was in high school, maybe or first year of college and, and so I became really fascinated by people's belief systems and the psychology of it. And, you know, I wouldn't say that that motivated me to study psychology necessarily, but it was interesting to have those discussions. And so that maybe that was an early sort of motivator. But for that study that you mentioned, particularly, it's, if you look at a lot of the research, you know, that people do on this topic, then often, what we do is we ask a bunch of students to rate you know, how likely they are to believe in some conspiracy theory, and then we look at what correlates with with that belief. And one of the things that we got frustrated with is that real conspiracy theorists don't want to come to our lab.
Bailer
Microchip implanted if they did,
Linden
Experiencing scientists and everything. And so I mean, are some techniques people have used avid colleague infiltrated the flat earth conference and things like that, but, but we hear but then the problem is that, you know, go through ethics committees, and you have to disclose what, why you're asking people questions, and then they still don't want to participate. So one way around that, for us was to use the sort of new computational methods that allow you to scrape you know, I guess, quote, unquote, big data that everyone uses that word, you know, really depends on what's big, what's not big, but for us, for us, it's vague, because we usually deal with, you know, 100 students, 200 students. And so we were able to actually look at the top accounts of the major conspiracy theorists on Twitter. So the actual, you know, main sort of spreaders, and we scraped all of their timeline data. And then also we looked at a random subset of their followers. And we tried to sort of map out the structure of their social network and what they were talking about. And, one of the things we and we did the same for popular scientists, by the way, so we looked at who the top popular scientists are, and see what they're talking about, and what their followers are talking about it. There were some really interesting, interesting differences. So for example, we found that the conspiracy clusters really had higher scores on themes like fear and anxiety, they were promoting information that contained those types of themes. They were also looking at, you know, things like, there was a lot of distress, I think she wouldn't you would expect it also some things that we didn't necessarily anticipate, for example, they were not as high on promoting certainty, as we thought, in fact, scientists also had a lot of certainty in their language is interesting, because scientific processes is very uncertain in some ways. And so there were some interesting differences. But by and large, we found that, you know, they were very high on the use of profanities. In the language, death, religion were big themes. And, you know, kind of makes sense, you know, it's always it's always about some mysterious cover up of somebody who died, or, you know, whether it's Princess Diana, Osama bin Laden, and so, you know, and what we found is that the followers adopt a lot of this language, too. And that's kind of why it's widespread. I mean, it wasn't a causal study. So you know, we can't say that, you know, one cause the other but it was, it was interesting that there was such an association between what the involved people were talking about and how their followers were sort of echoing those concerns. And so very, very negative conversations overall.
Pennington
How did you identify the top conspiracy spreaders? I imagine that identifying popular scientists would be fairly, fairly easy-ish, right? Because they identify themselves as scientists, you can look and see how big their following is and how much they're tweeting. I wonder how you identify people who are pushing conspiracy theories? Because I would imagine they don't say like, I'm a conspiracy theorist. Right. So I wonder how you identified them going into the study?
Linden
Yeah, so basically, what we did I mean, it's just as what you said, just now we looked at one key metric and that is the number of followers that they had at the time of the study. And a lot of them myself, I mean, they're, they're, I guess, not self declared, but, but they're widely known as conspiracy theories. So, I would name the names, but our ethics application suggested that we keep it anonymous, right? So I, you know, I can't share the names of the people we included, but we did have blacked out Twitter handles. And so for the top scientists, for example, there would be a guy who likes science, and so and so you know, and so it's not completely anonymous in terms of who it was. But it would be the same for the conspiracy sort of theorists. And in fact, some of them are no longer on Twitter, because they've been banned for floating conspiratorial material, but too much. And so there were pretty much that the people everyone are aware of in terms of, you know, the, the sort of the big producers of conspiratorial content and then a random sample of their followers. And it was just by popularity metric, which is interesting, because the actual list was heavily biased towards pretty much white males. And so it wasn't, it wasn't a representative sample. Particularly because he had all sorts of people. So conspiracy theorists, I suppose. But the most popular ones on Twitter at that time all had that same sort of characteristic.
Bailer
You know, this is, I have so many questions. I mean, this is such a cool study, you know, the idea of, of diving in and sort of defining the this particular collection of people that are starting these conspiracy tweets, and then kind of then tracking their followers to see how that behavior occurs, I really liked the idea that, that you're using this paradigm of this case control, kind of model. And and, and how you describe this, it's neat to see this kind of carryover of methods that might be used in other contexts now being used to define a, you know, the individual is essentially the source of conspiracy as a case and the controls being scientists. Did you do any additional matching of those of those of the scientist controls to these conspiracy cases?
Linden
Yeah, it's a great question. And in fact, it's really interesting you say this, because we thought it was cool to adopt a case control study, as they're called in this context. But one of the reviewers didn't think it was that cool. And then that's that all story about our subpar causal inference strategy with the case control design and whatnot. And then we said, look, you know, I mean, if you do a highly controlled experiment, we can have an unbiased causal estimator, right? But if we're going to do something messy in the real world, I mean, this is it. And so this is this is as a, you know, as a, as, as clean as we could get it. But we didn't do a whole lot of matching No. And so this is, I think this would be a confound to keep in mind. Because, yeah, we were balancing the objectivity of the metrics. So we had some concerns from reviewers that, you know, other other criteria, you know, when we sort of pre-registered the design, there were concerns that people Oh, but how you select people based on what criteria and it could all seem subjective. And so the only criteria that everyone agreed on was objective was the number of followers that also we couldn't match, you know, because the top 10, let's say, scientists, you know, didn't, for example, included some, some, some, some females, but the conspiracy less didn't include any females. And so we could have matched, we could have found, let's say, a female conspiracy theorist, but they would have a very different number of followers. And this, you know, we went through the whole sort of range of these concerns, and in the end, we decided to just stick with the number of followers as the metric. So that did create some, yeah, matching that wasn't perfect. So people should keep that in mind for sure.
Pennington
But in a paper, you talk about measuring cognitive processing. And I wonder, how do you do that when you're looking at people's tweets, I imagine in a lab, right, there are ways that you could, you know, hook people up to psychometric stuff and sort of get their processing, but But what does it mean when you are looking at tweets, and trying to understand cognition or or things related to cognition?
Linden
Yeah, that's a great question. And I would say, you know, it's not it's not perfect. I mean, these methods are constantly developing, and we have better methods coming out, you know, every every few years, but the main tool that we use is called the loop dictionary. It's linguistic as I called the linguistic inquiry word count. And so it's basically a psychological dictionary. So what, what this team did, and they've been doing this for a long time is they classified words into psychological themes, group them, and based on millions and millions of data points have kind of validated these data. That's across hundreds of different studies. So it's a very well known sort of dictionary, because that's the problem with some of these dictionaries, some, some aren't used a lot, some unvalidated. So this is kind of one of the biggest ones. And then you get these fancy sounding themes, like cognitive processes, but I think, you know, in the end, the sub themes are a bit clearer. So under cognitive processes, we would be looking for things like causal language. And so the reason why that's interesting for conspiracy theorists is that a lot of the theories suggest that the reason why people, you know, adopt conspiratorial beliefs is because they want to make causal connections where there are none. So 5g towers and Coronavirus cases could be coincidence, right? But no, there's a causal connection. And so we thought that this search for causality would show up in the language that people use, for example. And, you know, so that's an example. And then an example of the words in the tweets that would be classified as having scores that are higher on causality, you know, would obviously be the word cause itself, but also related words, like effect, predict, and these these sorts of words. Surprisingly, we actually found that causal language wasn't a major factor. And I think coming back to the other point, this could be due to the case control design, right? Because our control was scientists who also tend to use a bit of causal language. And so maybe if you had used a different group, let's say lay lay audience, maybe the conspiracy folks would have stood out more so that's another discussion. But compared to scientists, we didn't actually find even though we hypothesize that they used much more causal sort of language, there was some certainty seeking language, which is also part of the conspiratorial worldview, that they want certain answers, you know, it's it's deterministic, these group of people are responsible for it. So that's, that's sort of how the words come together, and so forth. For other team themes like anger, you know, there would, there would be words like hate, that are classified under anger. And so that's how these dictionaries work. They have more sophisticated ones now that use the whole semantic context of a sentence to infer the meaning. And so there's a very fancy kind of topic model, as they're called. We tried that. But we ran our study, actually, before Twitter, increased the word count. And so as you know, you might know on Twitter, before the workout, it was a very short, short sort of structure. And so those topic models really didn't reveal anything interesting for us because there wasn't enough text to really work with. So we start with a relatively simple method for our study. But there are certainly other studies now. Using, you know, longer Facebook posts and things like that, where you can, you can get more intricate themes.
Pennington
You're listening to Stats and Stories. And today we're talking to University of Cambridge's Sander van der Linden, about conspiracy and misinformation in social media. I wonder, given the work you've done, what advice you would have for journalists who are covering the floating of conspiracy theories on social media, because it's something that journalists love to cover and talk about. And I wonder if you have sort of, you know, best practices about how to cover it well and thoughtfully.
Linden
Yeah, and it is, it is such a big topic. I think from some of our findings from our research indicate that it's, it's not good to repeat falsehoods a lot or myths, because what happens is, you have not always but often the way that our memories work is that when you repeat something that's already known to people, you increase its familiarity, or we call the fluency with which people process things. So the reason why we know that two plus two is four is because we've heard of it a lot, and it feels familiar, and we know the answer. And that's, that's a good thing. But unfortunately, it works the same way with bad information. And so you know, the more we repeat something, the more familiar it becomes in the brain, kind of mistakes familiarity for, for having truth value. And so that's, that's a problem. And this operates, regardless of factual knowledge. So even when people know the correct answer, it's hard to resist it when you hear something a lot. So you're still likely to think it's, you know, true, even though you know that it's not. And one example of this that I like to use is, you know, if I, if I tell you that, you know, one of the restaurants I went to one of the restaurants on your street, had a good time, but I got this terrible food poisoning was really terrible. Then two weeks later, I tell you, you know what, I made a mistake. It wasn't your street, it was somebody else's street, but every time you go past that restaurant, now you're gonna think of food poisoning, and so and so that's kind of how that works. And for that reason, I think it's important to sort of, you know, pre bunk misconceptions when you write an article and try to not repeat the myth too often. So when you cover the conspiracy, obviously you don't want to spreaded by by repeating it too often, but also, instead of debunking it somewhere at the end of the story where people might not go on and actually read it, you want to be upfront, and kind of what we call inoculate or immunize people, before they then go on to read about the conspiracy, because you don't want to run the risk of, you know, it sort of catching on. So I think that's one of the best practices, sort of, you know, ingredients that we've developed is don't repeat it too often. Pre-bunk it in advance. And if you do have to debunk it, you have to give people a credible alternative. That's kind of the last thing I'll say about that. Because one of the things about why corrections and fact checking isn't always effective is because they don't always provide people with an alternative explanation. So when I say that something is wrong, your memory will encode what you know as being incorrect. But if there's no alternative, it's going to continue to make inferences from that knowledge. And so we need an alternative to replace it with and that's, that's often not present in the articles when they're covered. You know, you might say, oh, there's this satanic pedophile ring operating, you know, within governments somewhere. But you know what people say, if that's not happening, then what is happening, and so people need an alternative.
Pennington
That's really interesting. I have a colleague here in actually MJF at Miami, Andrew Peck, who's a folklorist who studies communication. He has published an article about this idea of amplification, which I think connects to this idea of repetition, where you amplify a message, and people hear it over and over and over again. And that sort of helps speed, you know, from this more qualitative perspective, stuff sort of moving, because people hear it, there's no alternative, it gets repeated, and it starts feeling true. And so something that was a very, very small rumor gets amplified into this giant thing, because it keeps getting repeated without being sort of questioned. That's really interesting how that connects.
Bailer
Absolutely. So you know what, I want to pivot in a second to the the study the study, the experimental work that you did in terms of shift of perception, you know, some of the work where you looked at kind of, you know, contrasting the simple facts, fake news, and then kind of partial and full vaccination, I think that's, that's gonna be a fun piece, but I can't go there yet. I can't, because I'm gonna get there in a second. And but when you're talking about this linguistic analysis, and thinking about this, I found myself thinking, Boy, there's lots of really challenging aspects of it. I mean, one, you've talked about having a dictionary for mapping topics and content, there's, there's also a sense of, of intensity or valence of, of kind of the expression, it's one thing to say you hate something, it's another to say you dislike it, you know, so even on that scale, when you start saying there's a negative, you know, negative aspect, negative emotion associated, there's a there's an intensity. So my, my questions, you know, this is gonna drive Rosemary crazy, cuz I always do this, there's sort of two components here to this question. What is this issue of kind of intensity of feeling? But then you also begged another question, which was, what do you do now, in terms of replicating this? You know, you've told us that you did this study at a certain point in time, prior to some of the social media outlets, you know, restricting access for people that are deemed, you know, kind of promoting conspiracy? What are you doing now that there's been a change to 200? And, you know, 280 versus 140 characters? So, are there plans to replicate? So the two part question is, kind of Valence or intensity of emotion. And then the second part is replication to see whether or not this holds up into the future?
Linden
Yeah, both are great points. I think we didn't look at this. But it's certainly true that there's a second parameter here, which is the valence riding on the intensity of the emotion. And there's other research which has looked at that. So for example, they have found that the the valence of the negative emotions is higher among conspiracy theorists and online networks, when compared to control groups. So they have found that means, we didn't do that. But other research has found that. And the question about replication is a great one. And it's, you know, in a study we write it was kind of a unique period, because some of the people that we followed on Twitter are no longer on Twitter, they've been kicked off. So you know, how do you replicate that the character count has changed. And so I think it is very difficult to replicate these kinds of studies. And it's interesting, because we do engage in, in regular sort of replication studies of, you know, sort of do our little bit of contribution to the field. And it's relatively straightforward to replicate, you know, controlled laboratory experiments where we can recreate these conditions, but the field studies, and that's why you see that it's actually very difficult more generally to replicate field studies, not just online but also offline when you go to a store or something like that. This is very, very good. Got to recreate these conditions. And so it's it's it's challenging, I think in terms of how we would do it is that I, you know, our approach was relatively, even though the the computational stuff is often exploratory bottom up, we did have some strong theoretical dimensions that we started with, even though they weren't all confirmed, of course, but I would expect that other research on even on other platforms, using different kinds of posts would replicate these these general patterns. So to me, it would seem, it would seem like this basically, that, you know, if there is, and this is what we're trying to get at the paper, if there's some common, you know, semantic structure to the language of conspiracy, then this should emerge in different contexts on different platforms in language use. And even though it might not be exactly the same, it would have to be a kind of conceptual type of replication, which uses a similar but not the exact same method. And so far, that's pretty much what you see. I mean, every study that I've looked at on Reddit, Facebook, Twitter, finds if conspiracy theorists are much more emotional in the expression, particularly negative emotions, like anger and anxiety, so that seems to replicate. Even in all these sorts of different studies using different platforms, the findings around the sort of our grouping are pretty consistent, too. And things that are less consistent around cognitive, you know, because of the uncertainty that actually seems less consistent across different studies. And so then it's difficult to tell whether it's because of the studies or the designs, and then you get these sorts of comparability issues. But overall, I think how we would replicate it is we would do the same thing with a new, you know, with a slightly adjusted set of influencers using the new word count, but the same other otherwise, things exactly the same? And we would expect, you know, similar results, I mean, if we didn't find similar results, I would, I would, you know, put a note in saying that, you know, there seems to be high variability here that, that's either attributable to, to the, you know, to the previous method, or the particular case control, or the, you know, designs or there's something more generalizable but from what I've seen on all these different studies is that these results appear to be fairly generalizable, even though they've used different kinds of, of methods. But that's the last thing I'll say about this. So interesting, because in psychology, if you compare data, no identity physics, you know, social conversations change over time. And so, you know, in terms of replicating sentiment and things like that, it's tricky. And I'm not even sure if we all know what the standards for that should, should, should be. But it's, you know, a common characteristic of social media. And another study we did, where we looked at millions of users on both Twitter and Facebook, led by my graduate student, we looked at whether the predictors of virality are on these platforms. Yeah, and so this was, this was the main all US Congressional social media accounts, and all major media outlets in the US and we scraped, you know, all of the posts, we could get in a specific time period over a year, on both Facebook and Twitter. And then we looked at what the predictors of shit resharing and reposting were, and we were replicating previous research which finds that the more emotional the tone of the post, the more it goes by, you know, quote unquote, goes viral. And that is using standard emotional dictionaries like, you know, things like Empath, which is a Python sort of software package for the geeks we're interested in. But moral words, and especially to combo moral emotional words. So So you know, when you combine hate with murder, and things like that, and and that was well established. But what we found, in addition, was that the biggest predictor, by far, was our group language. So if you're a liberal, and you were posting about a conservative and vice versa, that really caused a huge spike in retweets and shares on top of those other factors, and especially when it was negative, so when you were dunking the other side, and that was consistent on Facebook, on Twitter, for media accounts for congressional accounts, that was just the biggest again, it was associative, but it was the biggest pattern and the largest odds ratios, the likelihood across all of the predictors, how much you were you were basically trash talking about the other side?
Bailer
Well, you have my head spinning here, Sander, this is really...Yeah, I gotta tell you Rosemary and I are duking it out to see who can ask the next question, which tells you as that's a good sign in terms of this. I, you know, I found myself thinking, Gosh, I wonder what you started to touch on this with the characteristics of kind of who is doing this type of retweeting and who's the characteristics you alluded to lose some of the characteristics of the conspiracy, originators? I'm curious about whether or not this holds. If you're looking outside, if this is sort of an internationally true characteristic, you've mentioned a lot of this, had you primarily been studying the kind of US conspiracy, US originators. And with this, with this also, if so would this apply if you were looking at the UK, kind of originators of conspiracy as well as scientists? I'm just curious how that plays out.
Linden
Yeah, it's, it's really interesting. You mean for conspiratorial content? Yes. Yes. Yes. Yeah. Well, I should say that, that there's that generally there is the group that does tend to be this sort of, you know, white, male YouTube type of audience. That's, you know, that that's, that's well defined. But then there are also minorities who are spreading conspiracy theories. But that's for totally different reasons. Right? That's because of a complicated history with minorities actually being persecuted, persecuted, but actually experimented on, for example, because those conspiracies tend to be specifically about how, because of you know, that, you know, that historical reasons. And so that's, I think, a very separate issue to address in terms of why that content is spreading. And that, you know, that's, that's not only within the US, but also in other countries, but if you if you zoom out, and look at those sort of, you know, mainstream conspiracy content and why that spreads, you find that those characteristics are pretty similar in the UK, the US and Australia. So the Commonwealth and us are actually pretty, pretty similar. But, um, the US is unique in, in, in the level and intensity and I think in the quantity of conspiratorial content. So do you look at international analysis, for example, about people endorsing conspiracies about climate change and things like that? The US stands out as being one of the only countries where that is, you know, so high and so, so prevalent. And same with things like Q Anon, even though I'm originally from the Netherlands, we had one Q anon case that made it to the Netherlands where people were digging up graves, because they thought there was some satanic coloring going on. And yeah, I mean, it's bizarre how, how it makes its way across the Atlantic. But, that was an isolated case. And so I think, you know, by and large, what you see is that the US is pretty special. In terms of Yeah, of conspiracy theories, and the people who, who spread them. So it's not it's not necessarily the same. And in other countries. It's much more diverse, I think, in other countries, the makeup of the audience. Yeah.
Pennington
Well, that's all the time we have for this episode of Stats and Short Stories. Thank you so much for joining us today.
Bailer
Thank you, Sander.
Linden
My pleasure.
Pennington Stats and Stories is a partnership between Miami University’s Departments of Statistics, and Media, Journalism and Film, and the American Statistical Association. You can follow us on Twitter, Apple podcasts, or other places you can find podcasts. If you’d like to share your thoughts on the program send your email to statsandstories@miamioh.edu or check us out at statsandstories.net, and be sure to listen for future editions of Stats and Stories, where we discuss the statistics behind the stories and the stories behind the statistics.