 Hello everyone and welcome. My name is Mike Fandrick and I'll be your host this afternoon. Thank you for joining us for this edition of iSchool Insights which is a virtual webinar series from Syracuse University School of Information Studies. The iSchool Insights series highlights trending topics in the information field and features lectures, presentations, and opportunities for engagement with the iSchools faculty, leadership, alumni, and current students. Before we get started today I would first like to acknowledge with respect the honored organization, firekeepers of the Haudenosaunee, the indigenous people on whose ancestral lands Syracuse University now stands. I also would like to point out to everyone in attendance today that this session is being recorded and barring any sort of technical difficulties the recording will be made available on the iSchools YouTube channel maybe a day or so after we finish recording. We're very fortunate today to be joined by our featured speaker Dr. Jennifer Stromer-Galley. Dr. Stromer-Galley wears many hats here at Syracuse University and at the iSchool specifically. So in addition to being a professor at the iSchool she is also senior associate dean for academic and faculty affairs. She's the director of the Center for Computational and Data Science and is also an affiliated faculty member with the Department of Communication and Rhetorical Studies as well as the Department of Political Science. Dr. Stromer-Galley will be presenting today on a very timely and relevant topic especially considering the presidential debate last night here in the United States. Her presentation today will touch on her extensive research in social media, strategic communication and online interactions in the political field and will specifically focus on the social media use in the 2020 U.S. presidential elections. So Jenny today will be giving us some insights on how the Trump and Biden campaigns are using social media advertising to influence voters and then ultimately to try to turn them out on election day. In terms of an agenda, Dr. Stromer-Galley will present for approximately 40 or 45 minutes and then we're going to use the remainder of our hour for a question and answer session with our attendees. Please note that the Zoom chat function is going to remain disabled for the duration of the presentation and then I will turn it on at the end so that you can submit your questions via the chat function to me as the host and I will relay relay them to Dr. Stromer-Galley. So without any further ado, I would like to again thank Dr. Stromer-Galley for her time today and for joining us and I'll pass over the mic now to Dr. Jennifer Stromer-Galley. Thank you. Thank you so much, Mike, for that wonderful introduction and thanks to everybody for joining today in the middle of your workday to talk politics with me for an hour. I'm going to share my screen. Give me one second here to make sure I push the right buttons. Mike, can you confirm we're good to go? Yes, we are. Also. Okay, great. Thanks. So Mike has already introduced me well. So I will just mention that I will be talking a fair amount about a project that I am involved in. It's the illuminating 2020 project. And the purpose of the project is to help journalists and the public to better understand the advertisements that the presidential candidates are running on Facebook and Instagram. So I'll mention this project as we progress through, but that this project in particular is very close to my heart. So if you're here today, then you probably are like me in that you are currently being inundated with advertisements on your social media accounts. In fact, I saw last night after the debate that the Biden campaign purchased advertisements on a number of news and other platforms for advertising on slot. The focus of my conversation with you today is on the social media advertisements. This image here I took from my Facebook wall yesterday. You might wonder, you know, what it is and why campaigns are spending money on Facebook advertisements. Our analysis from the illuminating project suggests that the campaigns are spending at least through September 13 70 million dollars of their hard earned word chest on Facebook and Instagram ads. It's worth noting that that 70 million is only about 30% of all of the advertising money that the campaigns spend in a presidential campaign cycle. Of course, they run television advertisements, they do mailers and those kinds of things. So proportionally, the advertisements spend on Facebook is lower than other outlets like television. But there are some distinct differences in the costs of Facebook versus television that are worth noting. So Facebook advertisements are roughly a dollar to roughly $100 for a television ad. In other words, Facebook advertisements are cheap for campaigns. The other thing to know is that the campaigns run many advertisements on Facebook. We estimate on the illuminating project that since June, the Biden and Trump campaigns have run over 200,000 advertisements on that platform. So why? Why are campaigns spending this money and this energy to put advertisements in front of you as you just wish to zone out and check in with your friends on your favorite social media account? They want data about you. Let me unpack that. President Trump is, as you know, running for office this election cycle again, running for reelection as a number of things, including the law and order president. And he wants to try to find people like me. That is people who are using social media, who are actively scrolling my Facebook account and who might have interests that align with his campaign and its agenda. So for example, Donald Trump might want to know if I'm interested in politics, if I'm likely a Republican, if I'm interested in women's rights, and perhaps if I'm interested in gun rights. Facebook provides a wealth of data that helps campaigns to answer those questions to find people like me, who have the demographics, the interests, the background and the partisanship to get involved with his campaign. So Facebook collects a variety of different sources of data to help with that matchmaking, if you will. They collect information about your interests and your hobbies, your friend network. Anytime you post on Facebook, the comments, the topics, the words that you use all help to build a profile that Facebook then makes available for advertisers like Donald Trump to speak to. So based on my habits, activities, friend network on Facebook, the Facebook predictive algorithms can answer some of these questions. Am I interested in politics? Yeah. Am I likely Republican? Maybe. Am I interested in women's issues? Yes. Am I interested in gun rights? Yes. Why does that profiling matter? You might recall in the 2016 presidential election that Donald Trump won by a very narrow margin on the electoral college vote. Hillary Clinton actually won the popular vote by about 3 million votes, but Trump won the electoral college. There's some speculation and unfortunately it's only speculation because Facebook does not make the kind of data available to researchers like me to really know if there was actually an effect on voter turnout. But the researchers and the journalists speculate that Facebook and some of this modeling work of predicting people like me and my interests helped the Trump campaign to basically find a needle in a haystack. So, Brad Pascals, who was Trump's digital director in 2016 and recently was his campaign manager for the 2020 campaign, said that what was beautiful about Facebook is that it allowed them to really hone in and identify Republicans in otherwise Democrat areas. I'll take Syracuse as an example. The city of Syracuse, the preponderance of voters in Syracuse, identify with the Democratic Party. If a candidate like Donald Trump wants to run advertisements in Syracuse, those television advertisements blanket the entire region. And that means that most of his TV advertising is actually communicating to Democrats who aren't going to vote for him anyway. But Facebook allows the Trump campaign to find those Republican voters that live in Syracuse. And there are some, they do live here. And so that targeting is the power that Facebook provides to campaigns that needle in a haystack. So let me dig into that a little bit more with you. Taking this theme that Donald Trump has been campaigning on for the last six months or so on law and order. I want to impact for you some of the different ways and mechanisms that the campaigns use Facebook to try to find that needle in a haystack. To do that, I need to unpack very briefly where campaigns in the first place start to build up a set of people that they want to target in the first place. I could spend a lot of time on this slide and in my classes I do, but I'm just going to touch on a couple of highlights for this. On the far left of your screen, there is the voter registration column. It's really important to recognize that if you are registered to vote, that data is the most valuable data for a political campaign because if you're registered to vote, that means they can actually get you to vote on election day. Campaigns don't want to spend time talking to voters, sorry, talking to people who aren't voters because that's just a waste of energy and money. So they need to know that you're registered to vote. Some states, not all, but some states also ask on voter registration forms, what political party you are registered or affiliated with. New York State is one of those states. And so you can choose to register with a political party or not. But if you are registered with, say, the Republican Party, then the Trump campaign knows not only are you a registered voter, but you're a Republican. And that is gold because those are the voters that the campaign wants to be talking to on Facebook to mobilize them and get them to the polls on election day. The other thing I want to note is the public records column that's the next one over from voter registration. Your public records are things like your occupation if you are in an occupation that has to be licensed. So for example, if you're a police officer, or a nurse, a doctor, a teacher, all of those occupations need to be licensed with the state. That information can be requested by the political parties and is used then to provide additional layers of information on top of that voter registration data. So for a campaign to know that you are registered to vote, and that you are a licensed police officer, might increase the probability that they predict that you're a Republican if they don't know already that you're registered with a Republican party. Similarly, if you have a gun license, or have a hunting license, all of also available to in the public records further adds information for the campaigns about your interests and hobbies that would signal perhaps your affiliation with a political party. So now I want to talk about the different ways that Facebook and the political campaigns in effect kind of work together to be able to help the campaigns find that needle in a haystack of say Republican voters in areas that are otherwise hard to target or find. The first is called custom audiences. So as I mentioned, campaigns build up databases from a variety of different sources of information. And then from those sources of information, they make predictions, if they don't know that you're aligned with the Republican party. And again, I'm going to use the Trump example here to kind of simplify things. So Trump wants to find potential Trump supporters. And so to do that, they take all of that data, and they begin to identify people in their campaign lists that they know are Republican voters. So I'm going to give you a concrete example. Let's say that the Trump campaign wants especially to talk on Facebook advertisements to men who are licensed as police officers. Maybe they think that that's a key demographic that they can really activate police officers who are aligned with the Republican party that will help shore up the base, maybe in swing states like Ohio or Michigan, where the vote is going to be close. So the campaign hands over those lists with personally identifying information like your email address, which is the coin of the realm, the key that opens the kingdom for campaigns to Facebook. Then they pass those lists to Facebook. And then Facebook matches the lists from the campaigns to their own knowledge of the users on Facebook, any matches on personally identifying information, then get served the ad like the one that I'm showing you here, which is focused on one of Trump's themes of attack around the idea of defunding the police. But that's not the only way. Facebook also makes available to political campaigns, what they call interest audiences. So as I mentioned, my profile, my behavior about my behavior as I use Facebook builds up a set of profiles about me of my interests. So let's say that the Trump campaign wanted to target people who are interested in gun rights, because again, they're thinking if they can mobilize people who care and are passionate about gun rights, that that might then get them to turn out to vote on Election Day. In this case, basically, all the campaign does is say, Hey, Facebook, I want to target people that you think are interested in gun rights. And Facebook has predictive algorithms that puts people into those categories of interested in gun rights. And those audiences are then served the ad that the Facebook campaign wants. I should note, in both of these cases, Facebook never shares the actual information, the names, the contact information of the actual Facebook users back to the Trump campaign. Facebook will provide percentages of hits or basically how many people were potentially targeted with an advertisement. But just rest assured, Facebook is not giving your data back to the campaigns. Then there's look alike audiences. This is kind of interesting feature of Facebook and was used heavily in the 2016 campaign. So again, let's say that the Trump campaign wants to target people who look like law enforcement. So in this case, the political campaign knows that they've got a set of people who they have already identified are Trump supporters aligned with the Republican Party that fall into the domain of law enforcement security, police officers, corrections, etc. They then pass those names and the personally identifying information to Facebook. And then Facebook has proprietary algorithm that basically predicts based on the characteristics from the campaign list, which Facebook users look like those Trump supporters who are law enforcement affiliated. And so my hypothetical data scenario here, you can see that there are matches that are accurate people who do in fact align with or are in the occupation of law enforcement. But with all predictive algorithms, there are some inaccuracies. So there's some unemployed folks and students, a professor who may not be affiliated with law enforcement. And all of those people then get the ad. Okay, so those three different mechanisms for how it is that campaigns use Facebook to target advertisements is part of the reason why I'm getting served a Donald Trump advertisement that is highlighting Judge Amy Coney Barrett for Supreme Court, because I match a certain set of predictive algorithms that suggest I look like I am a Trump supporter. For the record, I am neither a Trump nor a Biden supporter in this context. But this is for purposes of the research. Alright, so what do campaigns and do with all that data? What is it the campaigns are doing on social media? And I guess the thing to really understand is that campaigns are using the Facebook advertisements to try to find and mobilize their base of supporters, again, finding that one person or that household in an otherwise demographic or political affiliation that isn't like the rest. Now I want to dive into the illuminating project in a bit more detail and some of the analysis that we have been doing, looking at the advertisements on Facebook. And again, when I say Facebook, what I really mean is that company Facebook and the platforms that Facebook owns, specifically Facebook, the platform and Instagram. So when I say Facebook, I really mean Facebook and Instagram. Also, I should note that this project uses and is directly accessing data that Facebook makes available to researchers and journalists. They have what's called an application programming interface, basically a door in between their databases and our databases here at Syracuse, where certain rules allow us to gain access to certain amounts of information in those databases. Specifically for us, we're allowed to access the campaign advertisements and some related data on those advertisements such as gender, gender targets of the advertisement. In other words, when a campaign runs an ad, they do specify particular demographic characteristics, gender, age, location. And so that information we are able to access. We also have the amount of money that the campaigns are spending on the ads. And then as a team, we have built a set of algorithms that predict the content of the messages in the advertisements. So specifically, we're interested in the type of message that the campaigns are running on Facebook. And I'll explain that a bit more in a minute. We're also interested in the tone of the advertisements, how civil or connecting to the modicum of decorum, which is a big fancy word that says, are people nice to each other? Are they talking respectfully? Or are they being uncivil or nasty to opponents or other members of government in their advertisements? We're also really interested in the topics in the advertisements and what the campaigns are emphasizing as the major issues that their campaign is is advocating for. Okay, also, you should know that the website, the backup, the data that I'll be talking about and sharing with you is all available. So if you, not that I want to distract your attention from the presentation, but if you want to, you can go to illuminating.ischool.syr.edu and sort of play along because all of the presentations I'm going to share with you, you can look yourself and see what the data shows, because the website and this dashboard that we've built are meant to help you and journalists to follow along, make sense and hold, ideally, more transparent what the campaigns are saying and doing on Facebook. As I mentioned, when there are 220,000 ads that the campaigns are running just since June, it's really hard to get your arms around what the campaigns are saying and doing. And that's really the purpose of this project. Okay, so some observations so far of 2020. I'll just note that all of this presentation that's coming is based on the illuminating project. The dates for this portion are June 1 of this summer to September 13. So through the convention period in the start of the general election phase. So let's talk a little bit about message types. That is the types of advertisements that the campaigns are running. So in looking at this time period, and what the campaigns are saying in their advertisements, you might notice that the most common type of message in Facebook advertisement is calls to action. A call to action ad is any ad that says, you know, get involved, contribute, sign the petition, watch the video, you might have noticed in the example ads that I shared with you that they all have some kind of linker button that takes you back to the campaign website. The purpose is so that the campaign can learn more about you and also to get more and as a result to get more data about you, but then also to hopefully convert you into a more active supporter of the campaign. So calls to action are the most common type of message. I should note that ads do often have multiple types of messages in them. So it's also the case that in Donald Trump's case, he is emphasizing issues. Both campaigns are talking about their personality, their characteristics to lead that's what the persona category is. And Joe Biden and his advertisements are he's much more likely to attack Trump than Trump is to attack Biden in this time period. Next, I want to talk about tone. So here's a question for you. You may have watched the debate last night, you may be following the presidential election. If you are overseas, I know that there is a fair amount of news coverage happening about the US campaign right now. So thinking about what you've heard and what you've seen about Biden and Trump and their campaigns. Who do you think is more in civil in their advertising? I'll give you a second to think about that. So our analysis of civility and I should just mention what we mean by civil here. Again, we are very concerned about the decorum, the expectation that our political leaders should be held to a higher standard of civility and kind of appropriate discourse so that when political differences arise, they can be deliberated in a rational and appropriate mode. Our analysis so far of Trump and Biden in terms of the civility of their advertisements suggests that Donald Trump is compared to Joe Biden has more ads that are uncivil in their tone than Joe Biden does. Much of that has to do with Donald Trump's tendency and his advertisements as well as in his public speaking to put names on people. So for example, he talks about Joe Biden as sleepy Joe. Or last night during the debate he mentioned Hillary Clinton and talked about her as cricket Hillary. Those kinds of references we classify as uncivil because they are demeaning and derogatory to in this case the opponent in the debate. Here's an example of an uncivil ad. This is one that has been running continues to run by the Trump campaign and you'll see some a language here Democrats have lost their minds. Joe Biden supporters are pushing to defund the police and violent crime has exploded. We can't stand by all these anarchists destroy our country. So calling the other party the Democratic Party anarchists that we consider that to cross a line of the modicum of decorum. Then there is the learn more about why sleepy Joe is dangerous for America. So for our classifiers, this would be categorized as uncivil topics. I'll just pause here for a minute and note that we have different approaches to the algorithms that we've built. The in civility algorithm and the message type algorithm are based on supervised machine learning, which basically means that humans spend time looking at large numbers of these advertisements categorizing them. And then those categorized ads are fed into software that basically looks at the message features of those ads that have been categorized by the humans. So for example, on attack, for example, we're actually even better calls to action call to action ads tend to have very distinct features to them. They ask people to take some action. So there's some regularized speech, if you will, in those ads making it very easy to classify calls to action. Topics are used a bit differently or developed a bit differently rather. For message topics, we have built what are called lexicons, basically large lists of words that are indicative of the topic. So again, thinking about Donald Trump and Joe Biden and the topics they might be emphasizing. Imagine for a minute, what topics would you expect that Donald Trump might be talking about as compared to Biden. And I'll give you a sec to think about that. So in our analysis, what we have found, and again, this is June through September, is that Donald Trump emphasizes safety and the economy as the two most common topics that he includes in his advertisements, while Joe Biden emphasizes social and cultural issues and foreign policy as the most common topics in his advertisements. I'll note that the third category for Donald Trump is governance governance is a category that includes discussion about our institutions, processes, the ways that we go about governing ourselves. So for example, the discussion about whether or not the Senate and Donald Trump should appoint a new Supreme Court justice to replace justice Ruth Bader Ginsburg. That is a governance discussion. As well, the debate happening or it's not a debate, the discussion happening about mail and balloting, also as a governance discussion around how it is that the public should vote. Those are governance topics. And as you can see, Donald Trump is emphasizing those in his advertisements. You'll also note that Joe Biden is emphasizing COVID in a portion of his advertisements while Donald Trump barely talks about COVID at all in his. Now I want to briefly talk about gender. So this is data that again, that comes to us from Facebook, and it's based on the requests, if you will, from the campaign for who they want to target in their advertisements. Is Trump more likely to target men or to target women in his advertisements? Our results suggest that Donald Trump actually is targeting men and women about equally in his advertisements. But what's noteworthy is that Joe Biden is emphasizing or targeting women more actively in his campaign than men. And I'll talk more about that here in a few minutes. And then age, another data point that comes to us from Facebook. So thinking about Trump and Biden and their communication and their advertisements, you especially given that they're advertising on Facebook, you might think indeed I might think that the campaigns would target young voters more than older voters on Facebook and Instagram. But I would be wrong. As it turns out, Donald Trump and Joe Biden both emphasize or target in their advertisements, the older demographics of voters in the United States. It is worth noting, however, that the Biden, sorry, that the Biden campaign does spend more on the 35 to 44 age bracket than does Donald Trump. And you'll note that neither campaign is spending as aggressively, if you will, on the 25 to 18 year old age bracket. So even though you might imagine that Facebook and Instagram users are younger, the campaign is more interested in older voters. And I'll talk more about that one in a second too. So does this all matter? Does it make a difference that campaigns are doing and running all of these ads on Facebook? I sigh here because I don't know. Which is really hard to say because I really, really wish I could say to you, we know that when a campaign runs an advertisement on Facebook, that it has x% increase of vote, or it increases the number of dollars that they raise. We don't know. Facebook doesn't make transparent the level of detail of information that we would need to be able to answer some of those really important questions. For appropriate reasons, Facebook doesn't want to share personally identifying information with researchers like me, and I imagine you as potential Facebook or social media users also really don't want the platforms to share that kind of information with us. Similarly, the political campaigns don't share the data with researchers about what works and what doesn't as they run their campaigns. So really, all we have to go on at this juncture are surveys, which some folks are doing. I have as a team member have to rely on some of the other survey work. But for the most part, this project, we're looking at what the campaigns are doing. Their behavior, their strategies help suggest what they think is working or at least who they're trying to mobilize and activate in their campaign work. So that's what I want to talk about last. So our analysis suggests that about a third of the Facebook advertisements that the campaigns are running are targeted at battleground states. Battleground states are those it's currently 12 states that are very close. That is, within a margin of error, the Trump and the Biden campaigns are basically tying even in public opinion polls. So the states are a toss up, they could go to Biden, they could go to Trump. And those states tend to be swing states, most election cycles in kind of recent history. So there are states like faith, I almost said Facebook, that's funny, states like Florida, Pennsylvania, Michigan, Wisconsin, Nevada, those states bubble up to the top as battleground states. And they are disproportionately being targeted in the advertisements from Facebook. So I've done a little bit more analysis looking at the conventions, so the conventions, the Democratic and Republican conventions happen at the end of August, where the two candidates and became the official nominees for their party to run against each other in the general election phase. So I wanted to look at the analysis of what the campaigns are doing in those 12 battleground states through from August 3rd, basically the beginning of that convention period through September 20th. On topic, there's some interesting differences that crop up. So while Joe Biden was emphasizing COVID, generally, but to a lesser extent than other topics, like social and cultural issues in the battle ground states, Joe Biden is emphasizing the economy and COVID and health, which intersect as topics at a disproportionate rate compared to what he's doing in his generic or kind of overall advertising. Donald Trump's strategy is a little bit different as well from his mass targeting, if you will, or the strategy across the 50 states. In the battleground states, Trump is focusing on the economy on law and order or safety issues, as well as immigration, which again, strategically, he's not emphasizing as much writ large. But in the battleground states, that's a key theme that he is reinforcing. In tone, one of the things that stands out for me is the distinction, the distinctiveness on civility. So you might recall that Joe Biden did have a portion of his advertisements classified as uncivil. But that that margin gets even narrower, smaller in the battleground states. So Joe Biden's tone overall is quite civil in the battleground states. While Donald Trump's tone in the battleground states is it's not quite 50 50, it's about 55 45% on civil ads, and then gender. So the distribution in terms of gender targeting for Trump is no different than his general strategy in the battleground states, there's a little bit more emphasis in the Biden campaign to target women than in the strategy writ large across the 50 states. So to sum up the data that you give to social media companies about yourself feeds into predictive algorithms about your interests. And then campaigns work with Facebook to try them to find supporters like me, and mobilize them to get involved and to vote. So again, in this example here, that arrow is actually a click on link. And if you click on that, it'll take you to the campaign website so that you can donate money to the Trump campaign. That mobilization activity is important for the campaigns because it then gives them additional information about you that they then feedback into the data loop of the data's data patterns they're trying to build up to make more predictions about people that they can target in their advertisements. And then last and what I want to go back to from the slides I shared before, is when I look at what the campaigns are doing in terms of their strategy, what I see is that party demographics are driving their campaign strategies, both kind of generally but also in those battleground states. So for example, the Republican Party skews older than the Democratic Party currently. That what I'm mean by that is Republican voters tend to be a bit older than Democratic voters. And so it makes sense then that the Trump campaign would be emphasizing older voters in their Facebook advertisements. Then while the Biden campaign would also be emphasizing that 35 to 45 age bracket quite intensely in their advertisement targeting because that reflects the party demographics. Same with gender. Women are more likely to self identify with a Democratic Party than they are with Republican Party. And you can see that reflected as well in the strategy of the Biden campaign. Biden is emphasizing and targeting women at a much higher rate than he is targeting men, which speaks to party demographics. So I have now walked you through a whole bunch of information related to Facebook and campaign data and how that data then drives the campaign strategies that then lead to you getting inundated with these Facebook ads on your Facebook wall, just when you really want to check out and check in with your friends on social. I do have to quickly make some acknowledgments because I cannot do this project without my team. The illuminating team is a large group of people. I especially want to give some thanks to Dr. Jeff Hemsley who is here at Syracuse University with me as well as Dr. Brian McCurnan and Dr. Patricia Rossini who's at the University of Liverpool but was a postdoctoral researcher with me and then a PhD student before that. As well I need to acknowledge Sarah Bolden and Anya Korsanska. They both are doctoral students here in the School of Information Studies and I could not do this project without them. In addition, I work with about a dozen undergraduate students and I work with about five master students who help us with the machine learning and with the analysis of the data that I presented to you today. I also have to acknowledge Iitan Hirsch. He wrote a book called Hacking the Electric and that table that I shared with you of all the different sources of data that campaigns get that actually comes from his research. And so with that I would like to open it up to questions and I will note that if you don't have a question now and you do end up playing with the illuminating website. I've included the URL here on this slide. Feel free to play with the data and if you've got questions or ideas don't hesitate to reach out to me on my email address or to reach out to me on Twitter. I'm an active Twitter user. I should also mention that the illuminating project does have a blog and a newsletter so if you would like to learn more and stay abreast of the work that we're doing through the election period and the postmortems that we will surely be doing once the election is complete feel free to stay connected those ways. Okay Mike I will turn it back to you. I believe that folks if you want to ask a question you should type it into the chat field which I think is now available to you to do so. Mike will moderate the questions and then Mike should I turn off the slideshare do you think? Yes please yeah please please turn off the slideshare and yes as Dr. Schroemer-Galey just mentioned you should now have the ability to send a message to me in the chat if you have a question that you'd like me to relay to Dr. Schroemer-Galey please send it to me Mike Fandrick in the chat and you should be able to do that now I just changed that setting. In the meantime Jenny I think I'll maybe I'll kick off with a question while we wait for our attendees to type some in. Sure. Your presentation focused mainly on targeting candidates using social media to target their supporters I'm also curious if you're in your research if you find candidates using social media to target the other guys supporters and maybe an effort to suppress voter voter turnout rather than to turn their their own voters out. Actually Mike that's a really really important question and it is one of the reasons why we launched the project in the first place. The 2016 presidential election there was reporting that came out of Bloomberg News that suggested that the Trump campaign was actively targeting Hillary Clinton supporters in an effort to demobilize them in other words to send advertisements targeted to particular sorts of Clinton supporters with really negative ads that were attacking Hillary Clinton with the idea being that those voters if they were to receive that ad might go oh you know maybe I just won't vote I just I she's not that great a candidate I don't really I'm not that excited by her after all so yeah I just won't vote and that there is some evidence that the that the Trump campaign was actively working that angle this election cycle. There's a feature on our website on the the illuminating project website that is called unique ads and one of the things that we'll be doing a deeper dive on here in the next two weeks is looking at those unique ads specifically I'm interested in those smaller ad buys I have to back up and explain one thing so when an ad I mentioned to you that there are 220 000 ads the campaigns are running that's not 220 000 different distinct unique ads that's 220 000 ad buys often the campaign runs as an ad that has a very similar or the same creative content that is the image the text but they buy it multiple times targeting different demographics or different regions of the country and they they do that because that helps them get some data about which ads are working where with whom so that goes back to the the data that they learn about us when we engage those ads and so in order for us to really see which of those ad buys are working our targeting in particular ways we need to look at some of those unique ad buys look at the creative content and see if it's especially nasty or especially potentially demobilizing and then look to see where they're being targeted so the unique ads feature basically aggregates the set of ad buys on a given ads creative that's the same so that we can begin to see whether or not there are some efforts to demobilize and so that's coming because i do think it's an important topic and we need to be watching that excellent thank you very much we have plenty of questions coming in now in the chat i will apologize in advance to our attendees we're not going to be able to hit all of them but i'm going to do my best to sort of consolidate them so we can hit as many as we can the first first question that came in is a follow-up on how you measure civility when you're when you're assessing the various social media content specific question is it binary is it just name calling and are you coding this automatically or individual one by one right great question so civility is coded with an algorithm so we have built a algorithm that uses what's called supervised or human supervised machine learning which means that we have students actually undergraduates who look at hundreds and hundreds and hundreds of ads and they categorize them based on some rules and in fact the rules are described on the illuminating website it's important for us that you know where and how we're doing this classification work so we do explain in detail what counts as being civil or uncivil so we have some rules and as i mentioned for us a message is uncivil if it is failing to meet the norms of appropriate behavior or decorum of our elected officials it has historically been the case that we hold politicians to a higher standard of decorum in their public discourse and so that's the the the concept that we're applying here in terms of civility and so it's not a word based approach it's it's features based so it's a set of things that includes the words that the campaigns are using in their advertisements as well as the words in relationship to other words as an indicator that it's an uncivil ad so some of it is name calling like sleepy joe or crooked clinton or crooked hillary and it's other things like exaggerated language that attacks particular targets but it's more complex than that but hopefully that helps and every ad is categorized on a binary of civil and uncivil believe me i get it you might be saying yeah but you know there's there's a gradation there but the challenge is that machine learning is not necessarily the most nuanced and we are trying to have high accuracy in our machine learning work so everything you see on the website is at least 75 percent accurate what that means is that we look at some of the machine coded messages ads and we compare them to what we call gold labeled or what we say is the truth of the category of that ad and we look to make sure that it's at least the case that 75 percent of the time that category classification by the machine and the humans agree thanks mic next question yeah so this one i'll try to consolidate a couple different questions and it relates to how Facebook compares to other social media platforms both in the way that it handles its user data as well as how the campaigns may be trying to utilize or exploit that data so the specific question relates a couple different ways a few questions we receive to twitter and whether basically we see the same strategies being used on twitter that we do on facebook right um so fun fact twitter banned advertisements by the political campaigns and parties and they did that about a year ago don't quote me on that it's been a while my memory is not great so you might notice that there isn't any political advertising by the presidential candidates on twitter specifically um so uh in terms of other platforms youtube is interesting there's a ton of advertisements being run and targeted on youtube um also google when you do a search on google or the banner ads that run on websites all of that is is important data it's much much harder for us to work with and gain access to it each platform has different policies and rules about how to access that data and also what they make available around the data so in other words knowing um facebook provides things like gender and age uh the this amount of money that's being spent uh google and youtube it's a different structure to the data and for a variety of reasons for the purposes of this project we're just focused on facebook um but it is an important question because facebook is one of an ecosystem of social media and kind of internet targeted advertising more generally but it is worth noting that even in that context facebook advertisement is uh at least from our analysis is the single largest digital spend for the political campaigns based on our analysis but we'll see once the election's done and others get a handle on other um data sources whether or not that holds great i'll use another question as a follow-up for that one um this attendee states that facebook has a mean age of 11 years higher than what other social media platforms have so are you concerned in any way that there may be some kind of age bias or something going on by using facebook data versus other data from other social media platforms or data from those larger populations in general absolutely so it's important to note that i only talk about facebook and instagram because that's only the data we have so there would be so the claims that i'm making about the strategies by the political campaigns are focused only on facebook and instagram so there's no bias or skew in that situation um because i'm i not i don't want to make claims about social media writ large it is worth noting that while facebook skews um older instagram skews a bit younger and facebook advertisements as i mentioned are both facebook and instagram in this analysis here's another kind of combination of questions and it relates sort of how if you have any tips as to how kind of consumers of social media content can be maybe more critical about what they're seeing and maybe some tips or suggestions as to how to how to be a good or responsible consumer of social media content well that's a great question um in this challenging right because there is another reason why we started this project was we were concerned and and continue to be very concerned about misinformation incorrect information um i care a lot that people vote and that they have full access to the ballot there is a lot of misinformation flowing around right now regarding mail-in ballots absentee ballots and um i think it's further challenging and raising questions for voters about how best to actually exercise their vote um and i guess what i would recommend is to consume as many different sources of information as possible there is research that suggests that that people who rely on on kind of a single source of news whether that's fox news msnbc the new york times if you only really consume one source of news then you may only be getting kind of one perspective about things and so it's important to have a you know just as uh you want to have a healthy diet you know when you eat you want to have more than junk food you want to be eating your carrots and broccoli and getting your proteins and calcium those kinds of things the same holds with information you need a broad and healthy diet of a variety of sources of information one of the challenges in using social media and it is absolutely the case that people use social media platforms like facebook as their news source you are getting shares of news stories from your friend network when you're consuming facebook uh as again part of your routine life the challenge with that is your friend network is like you and so if you tend to lean or are a strong democrat then the information ecosystem that you get on facebook likely reinforces that perspective creating an echo chamber or a kind of an information bubble and that potentially means that you could get exposed to incorrect and misleading information that's accidentally shared by your friends and because it's shared by your friends it tends to um we tend to not be a skeptical or critical of that information when it comes by us well if our friend shared it it must be true we take it as true but there are risks there that actually it's not so a broad you know being a herbivore if you will or an omnivore that's the one being an omnivore of information helps to be less susceptible to misinformation and falling prey to voting against your own interests excellent thank thank you again um again i apologize or our attendees i'm certainly not going to be able to get through all of our questions we have maybe three or so minutes left but i'll just keep plugging along um this question relates to how the the candidates and their teams um how do they measure efficacy of the various topics that that they so if the trump administration is focusing on law and order or the biden biden campaign is focusing on covet or other social issues how exactly do you measure the efficacy of those particular issues in a social media ad yeah great great great question um so and i i don't think i did a very good job of articulating this piece but part of what campaigns are doing when they have that button so when they run in advertisements on facebook or instagram there is always a button to click on to link back to the campaign in some fashion or other when you click on that link you're engaging with the ad and if you engage with the ad that tells the campaign something about you and it tells the campaign something about the efficacy of that advertisement so one of the things we know from reporting out of 2016 and prior elections is the campaigns run multiple versions of ads they do what's called ab testing so they test out different topics different layout styles different images to see which ones you're more likely to engage with so for example going back to the the example i had with you of targeting law enforcement so if i wanted to see which advertisement i've created of maybe a hundred different sorts would be most effective at getting policemen to and i mean policemen because again my example was men who are police officers um to actually click on that ad would be to run many versions trying different topics to see which ones though that particular demographic is engaging with so the campaigns basically get a bit of feedback or a feedback loop on the engagement metrics based on the topics so you can surmise if the trump campaign is emphasizing safety and law and order in the campaign they're getting a feedback loop on facebook and instagram that suggests that is something that's resonating with the target demographics they're trying to reach which then leads them to run more ads focused on that topic i hope that makes sense it's a bit complicated i'll have to make a better slide next time for that one so i'll combine a couple questions now to close it out it relates to how maybe strategies have changed between 2016 and 2020 and specifically we have a question about how in 2016 how data from cambridge analytica may be played apart and how i think particularly the trump campaign targeted voters a lot can be said about cambridge analytica and a lot has been said about cambridge analytica the one one thing i'll say about cambridge is that cambridge analytica was trying to convince the trump campaign that they could help not only to find those demographic characteristics but could really appeal to the interests of voters and run ads that especially spoke to the personality characteristics and interests of the targeted voters i um i wrote a book presidential campaigning in the internet age and i've interviewed campaign staff since 1996 trying to understand how it is that they're using digital media and in talking with the digital directors from the 2016 campaign about cambridge analytica and how effective it was the one consistent message i heard from the campaign directors especially the digital directors is that cambridge analytica actually was not effective at being able to do the kind of personality profiling with any accuracy so in the end while cambridge analytica did work with the trump campaign what they ultimately provided was more kind of traditional data wrangling and data analysis on the target demographics based on what we know about the party and party profiles already so the skewed towards older men in the republican party for example so it's so it's kind of an unknown how much of an effect cambridge analytica really had on the trump campaign and their effectiveness but again the campaign directors were a bit unimpressed with the actual capacities that the cambridge analytica team brought to the trump campaign in terms of differences i have to tell you i actually predicted coming out of 2016 that the digital ad expenditures would be on par with the television ad expenditures in other words i was thinking that facebook and the the suite of digital um platforms and portfolios websites youtube etc would be more like 50 percent and the research that we're seeing so far suggests that it's not actually the digital is not going to quite meet television advertisements in terms of cost and ultimately i think that's that is because tv ads are just so much more expensive than facebook ads in terms of the strategy it looks the same i have to say from 2016 is compared with 2020 the one caveat in all of this is facebook did not make available in 2016 data about facebook advertisements i have no way to know what the advertisements looked like in 2016 and so when facebook made this facebook ad library and api available in 2018 i knew i wanted to do this project because it would at least allow us this election cycle to try to be and help journalists to uncover potentially some of the problems that we heard about in the 2016 election so we'll see we're still a month left to analyze the campaign and see how things end up but that's my sense so far excellent well thank you very much again we've we've got a little bit over time so i again apologize to the questions that we we weren't able to to get to um jenny maybe could you plug the the web page for the illuminating project again maybe if people want to submit questions i can in fact what i'll do is just bring the slide back up if you're one of the visual sorts so again if you've got questions or you'd like to play with the site you can go to illuminating.ischool.syr.edu or send an email to me or find me on twitter at profjsg and i will happily answer your questions and if you have story ideas or things you're curious about we would love to hear them maybe we can write a blog post about it excellent well thank you very much for that and then thank you very much to Dr. Jennifer Stromer-Galey for her time today for giving a great presentation and for answering all of our questions and also thank everyone for for joining us today i really appreciate you taking the time to to view our presentation as well as ask some questions here in the chat i do want to say again that we did record this session so we'll be making it available on the iSchools YouTube page so please look out for that and also please keep a look keep an eye out for our future additions or issues of iSchool insights we'll be posting these sessions pretty frequently they'll be on various different topics in the information field and more information on topics and dates and other details will be made available on the iSchool calendar the Syracuse University calendar we'll also be sending out various emails to different newsletters and notifications so i again want to thank Dr. Stromer-Galey for her time today and thank you everyone for joining us i hope you all have a good rest of your day thank you mike it was my pleasure take care everybody