 I would like to introduce to you Anita Godes who's a postdoc at the University of Mannheim and also working for the Human Rights Data Analysis Group and she's going to give a talk about information control and state repression. So please give her a warm welcome applause. Thank you very much. Yeah, my name is Anita Godes and I'd like to say this is my first CCC and I'm pretty sure it's the best conference I've ever been to. Pretty cool. I work at the University of Mannheim as a conflict researcher and as a conflict researcher I'm interested in understanding how governments use violence, where they use violence and what types of violence they use. And for the last five years I've also been working for the Human Rights Data Analysis Group and what we do is we try to analyze statistically large-scale human rights violations and ultimately build evidence-based arguments that can help bring the perpetrators to justice. So the talk I want to give today is on information control and strategic repression. So the ability to connect via social media has been celebrated across the world by human rights groups, by researchers and by policymakers because it seems like we finally have a tool that can help ordinary citizens connect, overcome collective action problems and mobilize against repressive rulers. Now for all of you sitting in this room you're quite aware that this view on the digital revolution alone is deceptive and it's deceptive for one important reason and that is that governments principally still remain in control of network accessibility. So governments haven't just been sitting on the sidelines watching mass uprisings in their countries. They've been building an arsenal of tools to manipulate, censor and survey information exchange happening within their countries. Now we know this and we've learned a lot about this in the last few days and I'm really scared also about what I've learned here. But what we don't really know is how this control of the internet informs government repression. So how does it actually inform the violence happening to people in countries by governments? Right, so in the next 20 minutes I want to give you a very brief theoretical background, then discuss what I mean by different types of repression used by governments and then I want to discuss a very important topic which is how we deal with a dark figure of violence. So how do we figure out who is being killed and not documented? And then lastly I'll give you some results on how internet control and violence seem to coincide. Okay, so let's think about this from the perspective of governments. That's what we try to do as conflict researchers. Governments use repressions, for example, during riots. So there are two different types of threats that governments can encounter. On the one hand they're hidden threats. So there are people who are organizing to overthrow a government, to build an opposition group and the government doesn't really know exactly who the threat is and where the threat is located, right? So in these situations, state institutions need information. They need intelligence on where these people are, what they're organizing so they can eliminate it before it gets even bigger. There are other situations where the threat is already fully visible. The government knows exactly who its enemy is. It can already see thousands of people gathering on the streets fighting for change and in these situations they're likely to just say, okay, we're going to use any means necessary to crush down this protest. So if we think about these different types of responses to threats within the state we might think about how governments might use different digital responses as well. So on the one hand they could survey the internet in face of a threat. They can glean intelligence from people's Facebook profiles, from what people are Twittering, what they're planning on doing, from events and so on and so forth, from email exchanges and use that against opponents. The problem is as soon as you let people generate content online you give them a platform to mobilize and organize. So the other option would be to shut down the internet, shut down all communication and give them no way to organize virtually. Now these are obviously two very abstract concepts. Lots of governments survey and censor at the same time but these are kind of the abstract ideas that they could use. And what I'd like to argue is that the choice of digital response either using surveillance or using censorship is likely to inform the type of actual violent repression used by governments. So the case of my research has been the ongoing conflict in Syria. The Syrian civil war has been called the most socially mediated conflict in the history. What we see here is the Syrian presidency's Instagram account. And I don't know about you but I can't see any traces of conflict in this account. It looks like Assad is ruling a country that's united under him and his wife is smiling and they're having a good time. We can't see anything of death and destruction that's actually going on within the country. At the same time we know that the Syrian government is using different types of surveillance against its population. It's using shutdowns of the internet and it's manipulating information as well. So this picture is obviously highly deceptive. To investigate just how deceptive it is, my colleagues and I co-authored a report that was commissioned by the UN Human Rights Office that appeared in August 2014 where we looked at the number of documented killings that have occurred in the first three years of the conflict. And what we find is that we can uniquely identify almost 200,000 individuals who've been killed by the regime. 200,000 people who we know have been killed. So we know that the dark figure is likely to be much bigger because not everyone is documented. So when we hear about refugee flows coming to Europe, we know that this is happening for a reason. People are dying on the ground and they're dying every day. But these people haven't all been killed in the same way. Some people have been killed through explosions. Some people have been directly executed. And the question is how do we understand the different types of repression used by the Syrian government? Okay. So what we see here is a snapshot of our database. What we received were lists from four different human rights groups on individuals who were killed, including the location where they were killed, the date on which they were killed, sometimes information on their gender and their full name, which I've anonymized here. And we were able to match every individual in every group through this information so we could arrive at a number of uniquely identified killings. What you don't see in this snapshot is auxiliary information that tells us something about the circumstance of every individual's death. Okay. So if we think about different types of violence used by governments, in conflict research we often broadly distinguish between two different types of violence used by the government. The one we can call targeted violence. So this is violence that's used to eliminate individuals or groups based on certain characteristics. So examples would be a government targets someone for the information he or she has because that person is defected from the army, because that person is leading an opposition group. So these are people who eliminated because they share specific traits. Another example would be a government that is an ethnic minority and that is trying to target another ethnicity. These would be what we call targeted forms of violence. The other form is untargeted generalized terror. The entire civilian population is seen as a threat to the government and anyone is attacked. Now these are two different types of strategies that a government can use to suppress its population and oftentimes they're both used at different points in time within a conflict. And Assad's army has been using both types in the Syrian conflict as well. Okay. So the auxiliary information we get from these human rights groups can help us to understand the circumstances under which each individual was killed. So this is an example of one person who was recorded in three different sources A, B and C. And what this tells us is the first source tells us the person died under torture in the prison system, right? So that gives us information that that person was singled out, he was detained or she was detained and then tortured. The second source tells us this person was shot. So this person didn't just die from the effects of torture, this person was actually shot. The third source tells us where that person was arrested and where they were held and on what date they were arrested. So if we put this information together, it tells us something about the nature of repression used against that individual. So examples of words that we find when we look at individuals who were killed in a targeted way would be people who were previously arrested or detained, who were found with their hands and legs tied, who were tortured, who were directly executed or who were named to be defectors. And examples of untargeted violence would be people who were killed in explosions, in bombings or children, for example. Now I look at a period from mid-2013 to mid-2014 which means I have over 60,000 records of people who were killed by the government. And in order to classify these according to these two different categories, I draw a sample of 2,000 and I can code them to establish what I would consider to be targeted and untargeted violence. And then I use supervised machine learning to help me with the classification of the rest and I'm happy to talk about the details of the method afterwards. Okay. So the next big problem we have is that we have very good and very detailed information on people who were documented. But in every conflict there are people who fall under the radar of human rights groups, no matter how good a job they are doing. So depending on the time and location when someone is killed, the probability of he or she being documented by a group is likely to vary. And this is a problem when we're trying to understand patterns because the patterns might be biased if we only look at documented violence. What we can do is estimate this dark figure using a method called multiple systems estimation. If you attended CCC last year, you might have seen a talk by our founder of the human rights data analysis group, Patrick Ball, where he explained this method in much more detail and explained it on the case using Guatemala. So I encourage you to look at that talk if you're interested in the method. I'm not going to go into more detail on it now. Importantly, this method can help us identify this dark figure. And on the one hand, it's important to identify this dark figure for the means of analysis, but it's also important because we really need to think about accounting for each individual who was killed by the Syrian regime. So there's obviously a much bigger importance in doing this than just for analysis. Okay, so to give you an idea of how important it is to look at the dark figure of violence, this is a map of Syria by different governorates across the period of one year. And the darker areas show us where we know very little about what's actually happening on the ground. So the darker the areas are, the higher the percentage of undocumented killings. And as you can see, there's high variation both within governorates across time and across different governorates. So if you would have only used the documented killings, you would have gotten a very different picture about what's happening on the ground. Now, you can also imagine that someone who was intentionally detained and tortured by the government and then killed might have a different probability of being reported than someone who was killed in an explosion. So we need to estimate these different figures separately. And only once we have this information can we actually carry on with our analyses. So now we have information on different strategies of violence used by the Syrian government across the period of one year. The next thing we need is information on internet access in Syria. And for this, the Syrian Digital Security Monitor has been collecting information for the last one and a half years on the level of internet accessibility within each Syrian district. And they've been doing this on a scale from one to four. So either it's fully accessible, it's temporarily accessible, it's infrequently accessible or there's no internet access at all. And this is the information that they kindly shared with me that I'm using as a measure of internet access within Syria. And what you can see here is how it varies over time across different Syrian governorates for different types of internet access. Importantly, what we can see here is it's not just that some bomb falls and then all access disappeared. So it's not just a technical problem happening here, but we see obviously that the government is putting work into either providing access or denying access. So what we're really interested in is understanding if the variation in this pattern is correlated to the variation in the pattern of violence used by the Syrian government. So what we see here is the expected proportion of targeted killings given different levels of internet accessibility. So where internet is fully accessible, almost one-quarter of all killings that occurred are targeted. So these are people who've been tortured, these are people who are directly executed or these are people who were previously detained. And as we move down in the level of accessibility, that number decreases. It decreases to about 8% where there's no internet accessibility. Or put in words, in governorates where the internet is fully accessible, one in four are targeted killings. And in governorates where there's no internet access, we see about less than one in ten targeted killings. Now I'm not making any kind of moral or qualitative judgment on any type of killing. All this is saying is that the type of violence used by the government is different in governorates where they have internet access versus ones where there's no internet access. So the higher the internet accessibility is, the more targeted violence the government uses. And conversely, where there's almost no access, there's going to be much more indiscriminate violence used. Okay, so in conclusion, I'm not making a fully causal argument here where I'm saying that the different levels of accessibility directly cause a different type of violence. What I'm trying to say is that the choice of digital response is likely to inform, constrain, or enable different types of repression used by the government. It's likely to be part of a larger repressive campaign. Okay, what we see is that it varies spatially and temporally. So it's not just as if in the beginning of the conflict, everything is selective and then as it moves over time, the government uses only indiscriminate violence. We see a big variation both over time and space. And we see that it not only affects scale, but it actually affects the type or the nature of repression used by governments. I'm happy to talk about all the methods and other details after the talk. And if you have any questions, please email me. Thank you. Thank you very much for this very interesting talk. So we still have a lot of time for questions and answers. So people can gather up at the mics and I will call you. Do we have a question from the IRC or Twitter? Signal Angel. Right now, not really. Okay, then we'll go on with microphone number one, please. Okay, I have actually two questions. One is in regard of the targeted and non-targeted killings. You mentioned children as a target. How do you see in this perspective the recent Pakistani school shootings? I mean, I would say this was targeted, even though not by the government, but by the Taliban, and it affected children. So is it not a... The indicator seems to be not valid in all cases. You can agree. Did you have another question? Yes. Okay, so no, definitely. The measure that I use for targeted and non-targeted violence is definitely something that's situation specific. So in the context of Syria, what we often see is that the circumstantial evidence tells us that these would, for example, kids who were five-year-old, who were with their parents and who were bombed, for example. But we don't see very many cases where children are tortured, for example. So I think in the Syrian case, that makes sense, but you're absolutely right. Depending on the type of conflict we're looking at, or non-conflict, the measure of the type of violence needs to be varied. And my second question is, do you think, or do you have evidence that the higher number of targeted killings during times of full internet access directly correlates to each other? Has it to do with the fact that in times of full internet access, the regime has more and better information to conduct targeted killings? So that would be my hypothesis. This is not something I can directly test, because I don't know which information the government's actually using that it's cleaning from its surveillance techniques. But I would argue that that is one of the reasons why they're probably more likely to directly target people more effectively. Okay, thank you. Thank you. Microphone number two. Just two observations, more comments than anything. Firstly, your construct of government versus civil society or society is kind of simple. Syria is a very large, porous country. You can enter from almost any angle, and you have several religious orientations to the country. So there is fractionalization that's not really reflected. And it's a shifting situation. So if you imagine a war zone in which your town is shelled, you have no electricity, therefore you have no mobile phone, you have no internet. So the outages are spurious. You might be accounting for a cutout in electricity rather than cutout in the network. So two points, right? So one, the government against people. So people against the government is also a form of violence. And we might be measuring the wrong thing. Thanks. Absolutely. I mean, I'm using a theoretical, highly simplified concept. In this specific piece of research, I only looked at the effect of government violence, or basically how government violence works. Obviously Syria is a much more complicated case. But if we only look at the side of government against the opposition, we can still assume that the people who are acting on the side of the government are somewhat unified in that they want to maintain the status quo or go back to the previous status quo. And in terms of measuring the wrong thing, I think there are probably periods within the data that reflect exactly that. But since we see that no governor just goes off the grid, right? So it's not like they're just bombings and then the whole level of accessibility goes down forever. We see that it goes back up again. So obviously it's not like it's just one kind of uniform response. Thank you. Number four, please. Hi. First of all, I couldn't thank you more for putting numbers on this matter because I think it helps tremendously in arguing also at a political level. Two questions. A, I think you demonstrated some very interesting correlations. Still, I have the feeling that the causation is a bit more difficult to explain. Could you elaborate on this and attach to this my second question. Have you made any temporal analysis seeing patterns like internet going off, something happening, internet going on? Yes. So absolutely, as I mentioned at the end of my talk, I'm not using the word causation in the social science sense in this case. What I'm saying is that these two phenomena, internet control and violence, are likely to be part of the same campaign used by governments. When they use which campaign is a very different question? So the first thing I wanted to establish with this was to see, do they even co-vary? Do we see any kind of correlation here? The second question is, when do governments use what type? I refer to that a little bit in the theoretical part. So we're likely to see governments, or in the case of the Syrian government, we're likely to see them using targeted violence in areas where they haven't fully given up hope of control. So where they're still fighting to maintain control, where they still have some faint hope that it'll be theirs at some point again. So in a sense, leaving the internet on there is a form of providing infrastructure for the people who are still supporting them, and spying on the people who might not be. And in places where they've completely given up hope, where they say, okay, we're going to punish people here for their involvement with any of the opposition groups, that's probably where we're going to see more indiscriminate violence. In terms of temporal analysis, yes, I've written a different research project on temporal analysis, and what we there basically see is that during full network outages, as they occurred a couple of times in Syria, the level of violence, the scale of violence, increases tremendously and significantly. So there's definitely some kind of larger strategic idea going on behind that. I'd love to talk to someone from the Syrian government, but I haven't been able to up until now. Thank you. Thank you. So question from the internet? Yeah. Is any work planned or already done regarding what is happening in Ukraine? I haven't worked on the Ukrainian case. I'm aware of other researchers working on the Ukrainian case, but I think that's also going to be a very interesting example of how governments can strategically use the internet. And even more there, I mean, what I've been looking at here is kind of very crudely looking at internet access as a form of information control, but actual information manipulation might be something that would be really interesting in looking at the Ukrainian case. Then microphone number one, please. Hi. Thanks for the talk first. Could that be that in areas with less or no internet access, there's just less information getting out that could identify a death as a targeted killing so that the correlation isn't as strong between those and there's a little bit of correlation between no internet access and less information that identifies killings as such. No, we don't find that. That's a very good point and I think what we do see is that where there's no internet access, we know less about the number of people killed, but the information that we do have on recorded killings is just as good. So we have information on that the people were, for example, killed in bombings. So the quality of the information stays roughly the same, it's just less. The quality stays the same, the quantity sometimes varies, yes. Thank you very much. Another question from the internet. Yeah, there was a question about the source categories. Somebody wants to know if you can name explicitly some of these sources. So yeah, the sources we use are human rights groups working in Syria. So if you have a look at the report we wrote... Where is it? Here. So the report names the groups in detail and they're for human rights groups that have been working actively on the ground and they've been doing absolutely amazing work. So when I talk about holes in the documentation, this doesn't mean that these human rights groups aren't doing great work, they're doing amazing work, it's just that some areas are almost inaccessible for them. So yeah. Okay, then one last question from Mike Tu. Hello. My question is about the methodology, how you determine if there is internet access and another question you partly answered already. I think the proportion between targeted killings and non-targeted killings regarding in relation to internet access is like a proportion, but have you also looked at the absolute figures, how they change with having internet access or maybe having some internet access? Yeah. So in terms of the methodology, as I say, the data I use on internet access wasn't something I collected myself, but as far as I understand, they partly had people helping them collect information on the ground within each district. So they had people who were telling them every month, we've been able to access the internet via our phone, we've been able to access the internet via DSL and so on. Like a survey. Yes, like a survey. I'd be very interested in hearing other ways to measure this maybe on a more discreet scale. So if people have ideas on that, I'd be very interested to hear about that. I haven't looked at the differences in scale of violence when it comes to different levels of network accessibility. What I have done is looked at the difference in the scale of violence between days where the internet is off and days where the internet is on. And there we do see a significant difference. So on days where the internet is off, we see a much higher level of violence. And this is even controlling for documentation patterns and everything. So it's not just an artifact in the data. Thank you. Thank you very much for your questions. If you still have questions, you can just after the talk, walk up to Anita and ask her. And then I would like to thank you again for your presentation. Thank you.