Dr. Lisa Oswald: How Digital Media Shape Our Political Behaviors

Show Notes for Episode #1 of Unraveling Behavior

In this episode, I am joined by Dr. Lisa Oswald, a computational social scientist and expert on public discourse in online environments. We dive into an impactful study she coauthored in the journal Nature Human Behaviour, which examines the complex relationship between digital media and democracy (Lorenz-Spreen, Oswald, Lewandowsky, & Hertwig, 2022). Our discussion covers how digital platforms can both empower political participation and contribute to political polarization. We clarify the differences between correlational and causal evidence, explaining why causal evidence is scarce and challenging to obtain when studying the effects of social media use on democracy. An important takeaway is the skewed nature of content production on social media—where a small number of users generate the majority of content, often amplifying extreme or biased viewpoints. We also explore the importance of political context, the role of corporate responsibility, and the necessity for transparency and content moderation policies to safeguard democracy.

Watch the interview on YouTube. Listen to the episode on Spotify, Apple Podcasts, or on your favorite podcast platform.

Links 

• Connect with Dr. Lisa Oswald: Homepage, X, Google Scholar
• A short clip introducing Lisa Oswald
• The scientific study discussed in this episode
• Press release about the study
• The Digital Services Act of the European Union

Timestamps

• 00:00 Introduction
• 01:46 The potential of digital media to erode democracy
• 06:33 Research question motivating the study
• 07:35 Importance of a systematic literature review
• 10:39 Types of digital media and political behaviors analyzed
• 14:13 Correlational vs. causal evidence
• 18:31 Selection of articles in the review
• 19:41 Scarcity of causal evidence and its alignment with correlational findings
• 21:03 Beneficial effects of digital media use on political participation
• 22:46 Increased digital media use linked to decreased trust in governments
• 24:11 Increased digital media use linked to greater political polarization
• 25:14 The skewed nature of content production on social media
• 28:43 The role of political context
• 31:57 Internet companies’ reactions to scientific findings
• 33:26 The need for data access for researchers and greater transparency for users
• 34:19 Balancing user awareness with corporate and governmental responsibility
• 37:15 Conclusion

Transcript (edited)

Sofia Morais: Lisa, thank you so much for joining us today. It’s truly an honor to have you on the podcast.

Lisa Oswald: Thanks so much for having me.

Sofia Morais: A good place for us to start our discussion today could be to talk about the reasons why digital media have this potential to erode democracy in ways that the traditional media like TV, the radio, newspapers, generally do not. So, what is different about them?

Lisa Oswald: Good question to start with. I would say, they share a lot of features. So, both traditional and digital media provide information for public discourse. And also newspapers have to be sold and read by people in order to keep their business running. But with digital media and, I would now say especially social media, we see some structures that are really fundamentally different. So the first thing is the new prevalence of user generated content. So pretty much everybody can produce content, put it online, publish things. And so we are moving from this traditional one-to-many communication situation with traditional media, like journalists putting things together, publishers publishing things, now to a many-to-many communication dynamic, which is complicating things, and also takes away this previous gatekeeper function. And so, basically, nobody is systematically fact checking things before they go online. And then we also have, usually, the shortening of formats. Like, a tweet is much, much shorter than a newspaper article. And also, short videos on Tik Tok are very different from documentaries in television.

Sofia Morais: And when you mentioned that there’s nobody fact checking the content that people publish on social media, it’s also that the information just spreads too quickly for fact checkers or responsible journalists to have time to do that.

Lisa Oswald: Exactly. Especially since the overall volume of information is just at an unprecedented scale now.

Sofia Morais: Is the issue of personalized content also a problem?

Lisa Oswald: Yes, absolutely. Through algorithms that sort of try to cater to this attention economy, they sort of tailor feeds to people to make it even more likely that people stay engaged, that people stay on platforms, consume more content. So we have a personalization of information. So not everybody sees the same information and has the same basis for discourse in the end.

Sofia Morais: And it’s even a personalization, since we are comparing it to traditional media, that is really beyond the scope of your awareness. Because you could say, well, “I watch this news channel because I like it, or because it confirms with my views of the world. But if I want to get this other type of news, there is this other channel.” So there, there is at least a conscious choice of what to consume. But with social media, you’re just less aware of what’s filtered and how.

Lisa Oswald: Yes, absolutely. And I think you’re pointing to a super important phenomenon. Because, on the one hand, you have a huge diversification of sources and content producers. So you can find someone who is sort of producing content that is exactly what you want to read. But on the other hand, you also have a very interesting pattern of centralization because the platforms hosting the content are huge, and also very untransparent to governance or democratic oversight.

Sofia Morais: I don’t know if it’s correct to put it in these terms, please tell me. But it seems that there’s a constellation of risk factors that make a negative outcome for democracy likely.

Lisa Oswald: I would say, the negativity of the outcome is precisely why we did the study, because we wanted to see, what is actually overwhelming, or how is this complex landscape looking like. But it is extremely difficult to draw one bottom line and one outcome in the end. And then, both the media landscape, which now consists of online and offline sources, as well as democracy overall are maybe two of the most complex phenomena you can try to study.

Sofia Morais: Could you tell us about the main question that your study aimed to address and why?

Lisa Oswald: As a research question, we were asking, whether or to what extent and in which context, digital media has or may have detrimental effects to democracy. And this was sort of a two stage question. First, it was generally the question, how do the directions of association look like? Is it predominantly detrimental or beneficial, when you look at different political outcomes? And also then to explore this across political contexts. Because we’re not talking just about effects on democracy, the world does not consist of just democracies. And also, I would say that the hypotheses that we were just talking about are, fundamentally, in discussion in established democracies.

Sofia Morais: Now talking about the methods that you used in your study. You used a systematic literature review to uncover the details of the relationship between digital media and democracy. Could you explain what a systematic literature review is and why you opted for this approach? Instead of, you know, doing a brand new, original study, for example. Why did you opt for this methodology?

Lisa Oswald: Very good question. Given the state of the current debate, or the debate at the time when we sort of designed the study, it was very clear that we wanted to be very systematic about this because this field of evidence is just so big and so complex. And, no matter which position you take, so… For example, let’s say you are in a regulatory position and your position is that you want to regulate digital media. You could find a number of studies that support your position. And if you were a tech company, you could also just find a number of papers that support your position. So I would say that in this field of evidence, it’s not this one study that can answer the question of whether digital media is a threat to democracy. That’s completely impossible. This field really learns through evidence accumulation. And we need evidence accumulation and then also synthesis in some way. But it was very clear that a classic meta-analysis, where you would take statistical coefficients… So what is the correlation between those two variables and other variables in different studies? And synthesizing those values, those numerical values, that this will not be possible in this field, because measurements and variables and definitions of variables were just so heterogeneous and diverse. So you can just imagine that there were, of course, lots of surveys, often also representative samples, or these very large samples in different countries. But then you have survey self-reports of, let’s say, online news consumption and support for populists, or something like this. And then you have, on the other hand, behavioral measures of how people interact with online interfaces, or you have a sample of 2 million tweets. And how would you compare those 2000 survey responses to those two million tweets?… And the coefficients that are derived from such studies? So it was unfortunately completely impossible to do a meta-analysis. So we said, okay, we take the thing in the middle, the systematic review, and try to sort of structure the field of evidence as good as we can.

Sofia Morais: So now that we are going into the details of your study, I’d like to ask you to define, in the context of the study, what you mean by digital media. We’ve also used the term social media so far in our conversation, and also democracy.

Lisa Oswald: I mean, defining democracy, maybe to start with the latter is I also a million-dollar question in itself. And it’s of course, a very multifaceted concept that cannot be measured as well with something. There’s no thermometer to put on the country to say, okay, today democracy looks better, or worse. But, usually how it’s done is, is that it’s broken into different facets and different dimensions. We just said, okay, we are interested in different variables. Outcome variables, political outcome variables, that can be, broadly speaking, be categorized in either being, perceived as detrimental to democracy or beneficial to democracy. So broadly speaking, an informed public is something that is necessary for democracy to function. So, political knowledge is one of those outcome variables that is sort of in the positive box. And then, political polarization, especially if you look at affective polarization, having very negative perceptions of the political outgroup that go beyond the mere opinions… This can be a threat to democracy or was previously regarded as definitely detrimental to democracy. Same with misinformation, hate speech, populism. So, of course, that’s more like a split between variables. But this is how we went about systematizing all the outcome variables that we see in this landscape. And then talking about digital media, I would say we took a very broad approach. So any type of medium that is hosted via the internet and accessed via the internet, counted. So we did, for example, not include traditional TV news or print, printing press. But, of course, we have social media, which is one category, of digital media. And, I would say, a very important one in this discourse and also this field. But also the online outlets of newspapers counted as digital medium. A podcast would be a digital medium…

Sofia Morais: In the systematic review, as you’re saying, you could not analyze, how large the observed effects are, but you could still look at the direction of the effects, right? You could still look at the literature and say: In this paper, is the association between consumption of digital media and a certain political variable beneficial or detrimental to democracy? So this you could still do.

Lisa Oswald: Exactly. Because we only included, of course, original, empirical evidence. So those were all quantitative studies. Studies that gathered data, analyzed data, and came up with some form of estimate of association or effect.

Sofia Morais: An important analysis in your study is that you examined both correlational and causal evidence. Could you please help us understand the difference between these two types of evidence? Perhaps using an example. And why you looked at both, why that was important.

Lisa Oswald: Yes, absolutely. So correlation generally, and I would say that this is the vast majority of studies that we had in our sample, those studies, they observe variables. They measure variables, for example, social media use, maybe even as people self-reported: “How many times per day, per week, do you use social media?” And then they measure an outcome variable. In our context, an outcome variable could be political opinions or something like political extremism on a scale, right? And then you have those two variables and a positive association or a positive correlation in that sense would mean that, if you have high values on one variable, so if you use social media a lot, then this would imply that you should also have high values on the other. That maybe your political extremism is also high. And the same for low values. But this is a statistical observational finding, this correlation, and could also be explained by third variables. So this does not necessarily have to mean that social media use causes political extremism. By no means. There could also be a variable and I’m just making this up, that is maybe, unemployment or having too much time, which causes you to use social media a lot. You have a lot of time. And also, maybe you are unsatisfied with your life, and this is why you’re becoming politically extreme. And then also, it could all go in the other direction. Just because you’re so extreme, you use social media, because you don’t have any trust in established media anymore. So, correlations are a statistical pattern, but cannot really tell what is cause and what is effect or whether there are third variables. And this is pretty much the how this field looks like. It’s extremely difficult to get causal evidence. Because if you want to prove that one variable causes the other, you usually need an experiment and you have to randomly assign the cause variable. So you would in this context, you would have to randomly assign people to use social media or not, or use it more or use it less. And then then measure the outcome. If you just think about research designs. How does this work? Pretty much everyone uses social media. So you can’t say, “now I just take all those people who’ve never engaged with social media”, and then, I take one group and you know, people are randomly assigned to both groups. And then I take one group and say: “Now use social media” and we see what happens. This sample would be a very weird sample, probably 85-year-olds and five-year-olds. So, this doesn’t make sense. The most prominent approach to do this, to try to get down to causality with social media use, if that’s your treatment variable or your cause, is to do deactivation experiments. So, you take a sample, a group of people, and you randomly assign one group to a conditioning which you should deactivate, let’s say, your Facebook account or your Instagram account for six weeks. And the other group, can just continue using it. And, this way you can sort of understand what the causal effect of the deactivation is, which is already a big step.

Sofia Morais: In your literature review, you were able to uncover nearly 500 articles. How did you search for all these articles? How did you select them? How was that process?

Lisa Oswald: Basically how you start this entire process is with a research question: What’s the impact or what’s the link between digital media and democracy? And then you design a search query that contains different keywords. So for us, those were keywords about different types of digital media. Then you have a set of keywords about different elements of democracy and the outcome variables that we were interested in. And because we were really focused on original, empirical evidence, not other conceptual, theoretical papers, we also had a set of keywords about methods. So something like “survey”, “experiment”, … And then you have this big query and you run it just like you would in any search engine. You run it on big databases. And we did this on Web of Science and Scopus. And then you have a bit list of articles.

Sofia Morais: We were saying that your analysis had two stages. First, you analyzed the full set of correlational evidence, and after that, you focused on the subset of articles that also reported causal evidence. Before we go into the details of your findings, could you tell us whether these two analyses converged more or less? How did things look like?

Lisa Oswald: Broadly speaking, we could just say they mostly converged. At least there was no systematic pattern of one looking very different than the other.

Sofia Morais: But there’s still a difference in the amount of evidence.

Lisa Oswald: Yes, absolutely. I mean it was a very, very small subset of causal evidence. And also, there we didn’t replicate those studies, right? We also just went with the authors’ conclusions, and probably also there were instances of not perfect causal inference in the end. The bottom line of this entire study is that causal evidence is extremely rare.

Sofia Morais: Let’s talk about the main effects, what you actually found. What were the main effects of digital media on democracy? Do you want to start with the beneficial effects or the detrimental ones?

Lisa Oswald: I’d say, start with the positive, right? And actually it’s even the most researched outcome variable, namely, political participation. And here I have to say, participation, very broadly defined. This could be voting in elections, but could also be participation in protests. And then also many forms of online participation, so signing an online petition, for example. These are all different forms of political participation. We found, or the papers that we reviewed found, largely positive associations. So the more use of digital media, in various forms, the higher political participation. So it actually illustrates the early hypothesis that the internet has a big enabling function for citizenship and for democratic participation. I also found it interesting because, again, looking at political regime heterogeneity… Most studies in democracies looked at voting and, in authoritarian and emerging democracies, authoritarian contexts, researchers mostly looked at protest participation. So these are very different phenomena. But still, in both contexts, there were predominantly positive associations.

Sofia Morais: But there were also detrimental ones, right?

Lisa Oswald: So one that was a bit surprising to us even, because we didn’t have the variable of “trust” explicitly in our keywords… Trust in governments, in politics, in democratic institutions, and could also be trust in mainstream media. For example, in Germany that could be trust in public broadcasting. But mostly trust in institutions. This variable is super important, but we didn’t have the word “trust” in the original search query, yet it came out quite a bit. So people were researching this phenomenon quite a bit, and so we also had it as a category in the outcomes. And there were really largely detrimental associations visible. Also some causal evidence in this field. But here I would really again like to emphasize the political context. So, if I lose trust in functioning democratic institutions in a democracy, then that’s of course, a somewhat worrying sign for democracy. But if you live in an autocracy, then losing trust in a dictatorship or in the institutions or, for example, state-controlled media, then this is the first step towards democratization. So, even though we find mostly detrimental associations, keeping context in mind here is really important.

Sofia Morais: And then there is also political polarization.

Lisa Oswald: I think this is probably the one that I’m asked about the most. We find mostly detrimental associations between digital media use and political polarization. But here it’s mostly correlational evidence. I already brought this example when talking about correlation versus causation… From the literature that we see in this review, we cannot really say is it really digital media making people more polarized, or whether it is rather polarized people being more active on social media. So this direction of causality is still quite unclear. Or at least at the time of writing was not so clear. And now, as already more evidence accumulates in the very recent past, it becomes more and more visible that it is more difficult to actually causally identify the effects of, and now I am saying, social media on political polarization, especially affective polarization, not so much issue polarization. But the opposite direction is quite visible in different types of data. So basically the selection effect, that the more extreme you are in your attitude towards the political outgroup, the more active you are in voicing this opinion, also the more toxic you are in voicing this opinion. And this sort of comes together in a very skewed content production mechanism. So, that social media and discussions on social media that you can see as an ordinary citizen or that everybody can see are really just no representative image of public discourse or public opinion. The vast majority of content that you can see there is produced by a small minority of very active and, oftentimes also, quite radical individuals. At least, this is my reading of the literature. However, I would also say that the interpretation of this diverges quite a bit, in public discourse as well. Because, let’s say I’m a tech company, and I would just say: “See, it’s not us, we are not polarizing people. It’s just the people who are crazy, who come here and voice their opinion and that’s the world.” And maybe that’s correct. But I would still say this phenomenon of this selection effect and this skewed representation of public discourse, is really ground for concern. Because we are very much social beings, right? We don’t form our political opinions in a vacuum. This does not come out of us, somehow, but we also just observe our social environment. And we see, okay, what is being said in public, how do people speak about certain issues, what is the tone when discussing certain political issues. And there are just so many feedback loops in many different directions. So it’s very difficult to disentangle cause from effect and also to identify the boundaries of causal effects here.

Sofia Morais: You were saying that most of the evidence you found is correlational, that there’s not enough causal evidence about the link between digital media and democracy, and that you found both, beneficial, positive effects and detrimental ones. The most positive effect you found was in terms of digital media increasing political participation. And, on the negative side, there’s decreasing trust in governments and institutions and there is also increasing political polarization. So these are the two negative effects that you’ve observed. Talking more specifically about different parts of the world, where did you observe more beneficial effects and more detrimental effects?

Lisa Oswald: The most beneficial effects were found in the autocratic regimes and emerging democracies. But I would also say, of course, what this heterogeneity shows… And even if we would say this is all causal evidence, those are all flawless studies… Of course, this also shows the selection of research questions by researchers. So, let’s say if you are concerned about the political landscape in the U.S. right now… And also, I have to say that the U. S. evidences really overrepresented in the sample, there is still a big range of different countries that were included in the review. But there a lot of papers published in the U. S. And if you just look at this phenomenon, and also maybe see how research funding is distributed, you can see that there’s a focus on the detrimental effects of digital media on democracy in democracies. Because also funders are concerned about the state of democracy. And this is why we decided to also examine this in more detail. And maybe in autocratic countries, and now here I’m hypothesizing, but maybe, especially some of the early studies were looking at the sort of enabling functions of the internet for organizing protests, for finding regime critical information. And also, enabling anonymous participation, for example, to reduce the risk.

Sofia Morais: I think this ties in nicely to some of the research done here at our institute that shows that if you really want to understand the mechanisms behind human behavior, that you really need to look at our surroundings.

Lisa Oswald: Yes, very much. And also not just the political context, but also, of course, the media system, they are just so different in different countries. And of course, they also have past trajectories. The established traditional media also sort of have their legacy in the online world. For example, in the U.S., you have a two-party system and you also have quite a polarized media system, whereas in multi-party systems such as Germany, we have public broadcasting, which is completely different. And public broadcasting also has online outlets now and also has social media channels. So, there’s just huge complexity. And also, I think, like the history, in the political sense and in the media sense… This is why I said, you know, it’s very difficult to simulate in a lab!

Sofia Morais: Given the state of the evidence (i.e., that there is not so much causal evidence, most of it is correlational), do you think it is still right for digital media platforms to say “the science is not settled”?

Lisa Oswald: So I would say our review probably pretty clearly showed that we know more than nothing. It’s not that there is a blank slate and we think that there is just no link between, let’s say social media use and democratic outcomes. I think this is off the table, and I would say, and I think this was already before that off the table. It is very unclear, precisely, what the mechanism is and what the scale of the effects is. So I think, there, we need more. But then also, on the other hand, I often wonder and I mean… That’s definitely a political question and not my question to answer, but I’m just wondering: So what is the null hypothesis? Like, do we have to be certain that there’s absolutely no harm to not act, or the opposite?

Sofia Morais: Like, how much harm do you need to document?

Lisa Oswald: Yes, exactly. And so I think this debate is quite skewed by very powerful actors. And there’s already action being taken, right? The European Digital Services Act is, I think, a very good step in the first direction, especially to increase transparency. It should also ensure, if this all works out now, researcher access to digital platform data. So we should now have a better or an easier time to actually demonstrate such causal effects. This does not, and by no means, mean that we can now run experiments on platforms to demonstrate causal evidence, and then again, we have this problem, but still, we should probably be able to just continue research on demonstrating also associations.

Sofia Morais: And when you mentioned the issue of transparency, is that also transparency towards the users themselves?

Lisa Oswald: Yes, both. Like, to individuals to sort of enable more choices. And maybe also, so for example, to say, “Hey, I want to have just the chronological feed. I don’t want to have any personalized advertisements. I don’t want to have personalized content, but just the people that I follow, in reverse chronological order, full stop.” Some already offer this. But yes, having more autonomy as an individual is of course one thing. But then of course, I think that also many decisions are political in nature, more systemic, and cannot be given to the individual. The individual cannot take responsibility for many meta-effects. And so this is what I mean with transparency: That we, as researchers, can do research in a good way to understand what are macroscopic patterns, to then inform policymakers on where they should take action. And one thing that I always like to emphasize, regarding individual social media use is to understand the dynamics, on a meta-level, for example, the skewed nature of content production. So that I, as a regular individual user, that I understand that what I’m seeing on a social media platform is not representative of public opinion, but that the people who post there have very different, of course, interests and motivations to post things there. But they are probably not the average individual. And that I can take this into account when reflecting on my own personal political opinion. And, I would say, this comes down to digital literacy. Understanding that there are more and less trustworthy sources. And understanding the dynamics of content production. If you think about issues such as fact checking, it’s also super hard to put this responsibility on the individual user to fact check every single piece of information that you see somewhere. So I think also content moderation policies have to be negotiated in some way. And at best also informed by research, which is a very difficult thing to do, both in terms of misinformation and also in terms of online hate speech.

Sofia Morais: Lisa, thank you so much for joining us today, taking the time to share your knowledge. It’s been a pleasure having you.

Lisa Oswald: Thanks so much for having me.

[End of transcript]