Phillip Arceneaux is an assistant professor of strategic communication in Miami University’s Department of Media, Journalism & Film, where he specializes in political communication, public diplomacy, and public relations. Arceneaux also leads the Diplomacy Lab at Miami University and has consulted on public relations and marketing for organizations such as the Department of State, Naval Academy, and Central Intelligence Agency.
Josh Anderson is an assistant professor in the School of Journalism at the University of Arizona, where his work focuses on studying science communication from an ecological perspective. Anderson’s research is situated within agenda-building and social identity scholarship. He’s a mixed-methods researcher who uses survey and experimental methods, interview-based qualitative methods, and computational methods to answer his research questions.
Episode Description
There's a lot of concern about how bots and algorithms shape our experiences, both online and off. Increasingly, those concerns are focused on our politics. Can bots fuel political polarization? Can algorithms shape political narratives? A recent study seeks to understand the agenda-building capabilities of social bots in a political context, and that is a focus of this episode of Stats and Stories, with Phillip Arceneaux and Josh Anderson.
Transcript
Rosemary Pennington
There's a lot of concern about how bots and algorithms shape our experiences, both online and off. Increasingly, those concerns are focused on our politics. Can bots fuel political polarization? Can algorithms shape political narratives? A recent study seeks to understand the agenda-building capabilities of social bots in a political context, and that's the focus of this episode of Stats and Stories, where we explore the statistics behind the stories and the stories behind the statistics. I'm Rosemary Pennington. Stats and Stories is a production of the American Statistical Association in partnership with Miami University's Departments of Statistics and Media, Journalism, and Film. Joining me, as always, is regular panelist John Bailer, Emeritus Professor of Statistics at Miami University. We have two guests joining us today on the show. The first is Philip Arsenault. He's an assistant professor of strategic communication in Miami University's Department of Media, Journalism, and Film, where he specializes in political communication, public diplomacy, and public relations. Arsenault also leads the Diplomacy Lab at Miami University and has consulted on public relations and marketing for organizations such as the Department of State, Naval Academy, and Central Intelligence Agency. Josh Anderson is an assistant professor in the School of Journalism at the University of Arizona, where his work focuses on studying science communication from an ecological perspective. Anderson's research is situated within agenda building and social identity scholarship. He's a mixed methods scholar who uses survey and experimental methods, interview-based qualitative methods, and computational methods to answer his research questions. Phil and Josh are two of the authors on a paper with the Journal of Public Relations Research that focuses on the role of social bots, what how they played in shaping the. I'm going to start that sentence over again because I totally flip flopped the two sentences when I was reading that in my brain, three, two. Phil and Josh are two of the authors on a paper out with the Journal of Public Relations Research. It focuses on the role of social bots in shaping campaign communication during the 2022 Ohio midterms. Josh and Phil, thanks so much for joining us.
Phillip Arceneaux
Thank you so much for having us.
Josh Anderson
Yeah, happy to be here.
Rosemary Pennington
I'm just going to ask you first and foremost, what is a social bot, and how do you identify them?
Phillip Arceneaux
I'll get us started off on that. So, a social bot, very simply, is a computer robot. So, it is a robot that is programmed to act in certain ways in a digital environment. It is algorithmically governed, and we use them in a variety of different capacities. So a bot can be assigned to any sort of repetitive task online. So chat bots. So when you contact your airline because you're unhappy, or you want to make a complaint, or your baggage is lost, and you hit talk to an agent, you're not talking to a human being. You're talking to a chat bot. So that is a a database of information that is programmed to try and resolve your issue in a certain set of ways. So that's one way that we see bots. Another way are social bots. So what it's the same concept, but they exist in online social environments where we interact with people socially. So with our parents, with our friends, with our significant others. And here I'm talking very much like social media, social networking platforms, Facebook, Instagram, X, all these platforms. Their job is to mimic human behavior, so they are programmed to act in certain ways where they adopt personalities to comment on certain posts or content, but also to like, reshare in certain ways that promote certain types of agendas. And so, what we look at in this study is trying to understand how are bots used as a public relations strategy to try and promote certain agendas of discourse, social discourse on political and public policy issues.
John Bailer
I was really intrigued at the idea of of kind of these personas that social bots may assume. You know, so so what's what's sort of the the thinking that goes on on behind the scenes for trying to identify and formulate and frame the persona that is assumed by such a bot,
Phillip Arceneaux
definitely, I will take a stab at that. So what I would say it comes down to a lot of like online consumer psychology. So trying to understand there's this you know in social science we talk about this idea of the two flow step of communication. I'm much more likely to believe something or to find it credible if it comes from someone, even if I don't know them, someone who seems like me, someone that I think is comparative to me in some way, an everyday person on the internet who cares about whatever group we're following or whatever issue we're talking about, so I'm much more likely to think what they say is credible. So bots, understanding that, adopt personality sets that try and mimic that. So they embed themselves in social discourse and social behavior by adopting these personalities. If I knew that this is a bot who's just being promoted by some company to talk about their product, I'm unlikely to find that credible. I'm unlikely to be moved to try and consider that buying that product or trying their service. But if it comes from someone who is more like me, someone who I think is a person who you know, this is the whole idea of Yelp, right? I want consumer reviews that are from you know by people for people. It's the same concept that we see play online.
Rosemary Pennington
So I guess I wonder how you identify them, right? Because sometimes it can be really hard to unmask who's a bot and who's not.
Josh Anderson
Yeah, I can speak to that one. So excuse me. So it is it is a little bit of a tricky. Situation to do for a lot of those social bots because you know as Phil was saying there there's an incentive for the people who design create these and deploy them to make them appear real in a lot of cases I mean you know in some cases you can see bot accounts that clear right clearly identify themselves as bots so you know a lot of the early literature on talking about bots on Twitter and now X was about bots that were curating news, so it was finding a lot of news that was published somewhere on on the website and then bringing it all together. And those were clearly labeled as a bot in most cases. They were never trying to sell themselves as a person. Now a lot of bots now, especially when we talk about propaganda and targeted political communication, there is a strong incentive to to hide that fact, so at a really detailed level, to be very very confident about it, it can take years. I mean, it can take a very long intensive schedule of how you do that. You might have to, you know, Sam Woolley, who is now at at Pitt University. So something that he did a lot of his career in was doing actually in-depth interviews to identify bot accounts, so that's that's not something that we can do for every project. I mean that's that's a significant amount of resources. So there are some systems that are often that often live on whatever platform you're looking at. In our case, for the study that we were looking at, we used a system developed by Mike Kearney, who had a computer system that would go in and take a look at the account for any account that you feed it, and it would give some degree of confidence that that account was a bot or not. So some of the things it would look at are like posting behavior. I mean, is it an account that's making a new like 2000 new posts a day? Well, that tells you to some degree that's probably there's some kind of automated process going on there because most people don't make posts of that degree of magnitude.
John Bailer
So one thing that you all were doing that in this project we were trying to connect these social bots to the idea of agenda building. Can you talk a little bit about the framework for how you conceive of agenda building and what group is is making such agenda?
Phillip Arceneaux
Absolutely. So agenda building starts all the way back a long time ago with its kind of predecessor theory of agenda setting. So, '67, I believe, Max McCombs and Shaw published a paper talking about the agenda-setting function of the press. The idea being that what things the press deem newsworthy that they cover, that tends to be what voters think are pressing public policy issues. So, we were just speaking earlier. If we're talking about Venezuela and we see Venezuela covered in the news, if all of a sudden we start seeing Iran covered in the news, people think Venezuela is no longer as important of a public policy issue as Iran. So that's the agenda setting theory. So I come from a public relations background. So the idea of agenda building is if the agenda, if the press set the agenda for the public, who sets the press's agenda? So that comes from a public relations perspective of how can we use media relations, use press releases, use direct interviews, press conferences, all these kind of tools that we can think about from the political campaign side. How can we use those to try and influence what the press cover, how they cover it, how they talk about it, and then from that distill that down to the public to try and influence how voters and constituents think about public policy issues.
Rosemary Pennington
So you all were looking at how these social bots could contribute to agenda building or not around the Ohio midterms in 2022. I guess I just am curious what drew you to this to begin with. I mean, I know Phil, you're a public relations researcher. Why, why that, why that political moment? And and again, why, what was it about social bots and their potential agenda building function that you found interesting enough to kind of dive into?
Phillip Arceneaux
I'll start, and Josh might have some different thoughts on this. For myself, first off, I had recently moved to Ohio, so this was the first time I was actually sitting in Ohio for the entirety of an election cycle. So I was very interested in thinking about this model that we've used. We use very regularly on both midterm and national elections. So we replicate this, looking at different kind of small variables and contexts across states, across different elections, that all those components. So we're always looking for something new. What's going to contribute? What's going to add theoretically, and we had been talking about you know since 2016 there was this idea of fake news, and then we kind of moved into the disinformation and misinformation rhetoric. But we finally starting to get into the you know post COVID years, early 2020s, talking about this actual concept of social bots. That is what is actually making the fake news or making the misinformation and disinformation happen so much. So we had this concept, and we had an election, and we're like, great. Why don't we actually look at this in an electoral context to try and understand? You know, campaigns can use bots, super PACs, and Democrat and political parties, and all these groups can anyone who has funding in a relatively small amount of funding can use these bot networks to promote their issues and promote their agendas. So it's not unique to just politics, or just the the campaigns themselves. But we wanted to know: Do the campaigns one does it help to extend their agenda? So we use the concept of algorithmic algorithmic amplification. Can basically the campaigns increase the reach and output of their messaging through these algorithmic means via social bots, and also trying to understand. This in misinformation and disinformation ecosystem is that actually are bots actually hurting an electoral campaign's ability to promote their own agenda and get that communicate that to constituents, which whether whichever party you're looking at, whatever campaign or candidate you are for or against, every in a democracy, every candidate and every campaign has a right to communicate their perspective on the issues to the voters.
Josh Anderson
Yeah, I think that you know the what really roped me in as far as like why this project was so interesting for for me to work on and contribute to was that one it really it really speaks to this kind of moment that I think we're having in the social sciences of a lot of interest in AI and its applications, which you know, social bots-it's not AI in the sense of they-they aren't typically AI in the sense of like generative AI that we think of. They're often a lot more you know simplistic processes than that, but really just writ large. I mean, it's hard not to talk about AI at this point in some way. And so this was, I thought, kind of strong application that tells us a lot about the social world. You know, it was an it was a fairly high covered election. I mean, you know, J.D. Vance was running in this election, just put in perspective, and it matched up well with the timeline of you know a lot of this research that was coming out about AI and what we could contribute to there. So that was something that I figured would be a good contribution of how do we try to determine what kind of role they might play. And again, what Phil said about algorithmic amplification-that's something we're specifically looking at because that's what other people have found, especially in a lot of the these interview studies of people who make and run these bot accounts-is that the people who make them very clearly see them as a tool for amplifying political messages, and that's documented in quite a few political contexts, both in the U.S. and places like Hungary and India, Bangladesh. And you know what we what was really interesting was can we see this in action? Like, can we actually see this in in the data that there is this kind of amplification effect or anything else that might be there?
Rosemary Pennington
You're listening to Stats and Stories, and we're talking with Phil Arseno and Josh Anderson about how social bots can shape political communication.
John Bailer
So before we dive into the the data and the analysis, I just had one one kind of last clarifying question for me, and that is, you know, what who are the source of the bots? So who's who is producing them, and how does it how do the bots relate then to campaigns and other other players in this in this marketplace.
Phillip Arceneaux
So I'll take a first stab at that. Anyone can use these social bots, and they are basically guns for hire. They are mercenaries who can be employed by anyone who has the funding to do it. And one of the unique things about social bots is their return on investment. They are incredibly cheap in terms of what it costs to buy impressions, what it costs to buy exposure for your issue. It's actually very cost efficient to be able to push that out. And again, because of that two-step flow of communication, that psychological component-that if a campaign tells me that that candidate's going to be the best for this, I don't know if I'll believe that. But if someone who seems like me tells me that campaign that candidate's going to be the best. Okay, I might start to think about that a little bit more. So I think that is one of those one of the reasons that we can think about why this is so important, and it can be deployed in the political sphere. Campaigns can can pay for these bots, super PACs, political parties, you name it. But this can also be used in the private sector. So this can be companies who don't want to talk about an issue or really want to talk about an issue can use that. I think again, from a public relations perspective, it's very useful for smaller companies that don't have that kind of exposure. So, like nonprofits, if you have a nonprofit that's really passionate about a political public policy issue, and they don't have the means or the brand exposure to be able to get that national coverage, they for a relatively small cost, they can employ some bots, be able to push that in the right social circles and micro-target that content to certain circles, whether public policy spheres or news journalists and news editors, the right spheres to get that in front of people's eyes to then again make make it disseminate further than they could on their own.
Josh Anderson
Yeah, I mean, just to put in perspective, the cost element is again this is deeper at a deeper level than we we did with this study. But reading some of the ethnographies that people like Sam Way put out, I I think some of the most interesting cases are people who are just people, just people who have political opinions and want to have a wider reach to spread those. So maybe their own personal social media account they have, you know, maybe a few 100 followers. They have very little presence themselves. They're not a political actor, in that sense. But they do have political opinions that they want to influence the conversation about. So they spend, you know, probably a few $100. Maybe it's gotten more. It's probably gotten more expensive as computing costs have gone up. But you know, still a very small amount of money, of you know a hobby of scale, and they can create 5000, 10,000 bot accounts. It's not hard. Technically, it's really not very challenging. It's something that you can teach yourself pretty well, and so you know it's attractive to certain people who want to have that political amplification.
Rosemary Pennington
So I'm curious, what exactly did you guys do to try to figure this out?
Josh Anderson
This being the yeah yeah
Rosemary Pennington
so the yeah so this this work that you were interested in how these bots were impacting this communication during this midterm what did you actually do as your study
Josh Anderson
yeah absolutely I can I can walk through that is that in the broader platform of how we run these agenda building projects one of the first steps that we take are singling out different columns of communicators. So we might find a way of, if we're doing this on, say, X or Twitter, for example, we might have a list of accounts that we expect to be communicating about it. In this case, we did. So we had the campaign accounts, like so, that was just lists of specific accounts that we knew would be communicating about it. We might have news accounts, so that we already identified, and then we might collect data, and in this case again, we did of everybody else who was tweeting about the election based on the keywords they were using. So it was out of that last bracket that our task was to pull out which were the bot accounts from that versus which were the ones that we figured were not bot accounts. So what we used was actually a package on a statistical software called R that was and the name is always fun on these R packages. So it's tweet bot or not, and of course R is in it because people who write an R love to include R and and their package names, and Mike Kearney is the one who developed this, which I believe is still not functional because it was still using the old method of data collection from Twitter that had changed actually around the time we were doing a study, and but we did get it in before that change. So what it did was again this is that system that looked back at the account for every tweet that was that for every account that you fitted, it looked back at its history, seeing things like its posting speed, looking at how similar each post was to another, and it used inferential tests to give a degree of confidence that that was a bot account, just based on the account history.
John Bailer
So, so now you have these different groups, whether they're they're campaigns, newspapers, the public, the public accounts being classified as to whether or not you think they're a bot or a social bot or not. So now you've got just a boatload of communication that's that's been that you're you're collapsing into some pool for analysis. So you know, I think you had something like you know how many accounts, almost almost a million accounts that you're looking at, or
Phillip Arceneaux
I think, and correct me, Josh. I think we had around 900,000 tweets was the corpus, and then we had 34 election campaigns. We had 47 newspapers that represented each one of not only Ohio's major, or one of 47 newspapers that represented Ohio's major media markets, but also at least one representing every congressional district as well, and then again, I think in terms of users on X, I think we had around 300,000 individual users, and from that we extrapolated around 2000 were identified as likely to be bots.
Josh Anderson
Yeah, that sounds about right to me.
John Bailer
So, so then what?
Phillip Arceneaux
So then, and again, Josh, feel free to correct me or not. So from then we have to say how do we actually identify the content. So if we're thinking about what are they talking about, we now know who we want to be looking at, but we want to understand what are they talking about and how can we track that across different groups. So we started to develop keyword dictionaries. So we had to identify what could the what could we tell the computer if this word is present, this tweet needs to be coded as this policy issue is present or is not present, so in a binary fashion. So we started. We actually use the data itself because we've run studies like this numerous times. We have in a variety of electoral context. We have a pretty reliable sense of the kind of main categories of public policy issues that we look at: criminal justice, healthcare, energy, civil rights, these kind of categories, and so what we did was we started to look at the actual content itself, and we started with the campaigns because if we're trying to figure out how is the campaign, how are the campaigns driving the discourse, we want to know what language they are used, and is that being reflected by other communicators online. So we basically did some initial inductive analysis of the of the tweets. It helped having an Ohio native who was part of this study because I was actually living out this election, so it was very easy to identify the main things that a lot of the larger campaigns were actually talking about, and so we started to develop rudimentary keyword lists. So okay, we we know we they talked about this, we know we talked about this. We could start to put certain words under certain categories, and then we used another method where we took the 1000 most common words. So we did a little bit of data cleaning. We removed a lot of the noise, and then we say, what are the 1000 most common words that we see from the political campaigns? So we then took that list and said, okay, how do we start to categorize these within these different primary category groups? From that, and I will check everything to make sure I'm giving you the right the right information. So we identified off that initial analysis, we identified 13 primary policy domains or keyword dictionaries for 13 different groups: civil liberties, economy, education, energy, environment, executive affairs, foreign policy, healthcare, immigration, judicial affairs, and legislative affairs. So that was kind of the bulk of like if the. Talking about public policy, largely most terminology fit within those contexts. We did have some other words that, to us and our interpretation, were relevant to the election, but they didn't quite fit under those. So we inductively created four more groups, and those were campaign rhetoric. So get out and vote. You know, donate. These kind of words that we see campaigns elections tend to use very often. Campaigning rhetoric, election rhetoric. We also looked at political stakeholders. So, an important thing: we don't just look at what issues are being talked about. We looked at what stakeholder groups are we referencing in relation to those issues, and that could be individuals such as the candidates or politicians. But it could also be organizations. So, you know, if we're talking about civil rights or civil liberties in the context of a Republican campaign, who's talking about Second Amendment gun rights, the NRA will likely be referenced. So, how can we track what stakeholders are being linked to these public policy issues that fell into that category? And last, another one that we've we've started using in some other studies are also cities. So, cities are stakeholders. So, how often is like Cincinnati being talked about versus Columbus versus Cleveland, and what issues are being talked about relative to those specific cities. So, in all, we identified, I believe, it was 17 keyword dictionaries that we used for analysis.
John Bailer
And so, so one aspect of doing this is that now you're going through this giant classification scheme, then, and you're trying to to populate how how those 17 categories are being used by different groups, including the bots and looking at that over time, so you're trying to you're indexing this over time. So, so what kind of insights were you hoping to glean from looking at these patterns over time?
Josh Anderson
Yeah, so I think the first way to answer that is to talk about how you transform these into being time-based data, because you know, as as you start off with, it's not there's not necessarily a time component to just categorizing their huge corpus of tweets, which I think even started out into the millions of tweets. It was quite large, right? But you know, you're slicing out of these different categories. How do you digest it into time? Well, one is that you think about the time frame that you're collecting from. So, and again, they'll correct me if I'm wrong here, but I think it was about 72 days.
Phillip Arceneaux
That sounds
Josh Anderson
correct. Yeah. So for each day, then that becomes, if you think about it in terms of data table, that becomes your new row. Is that each day is telling you across, you know, all of the different accounts and some collection of accounts, these are how many times that among all of those accounts, they talked about this number of issues. So maybe on the first day there were, you know, I don't, I'm not speaking from data here, but just you know, 200 criminal justice tweets on day one from the Republicans. That might be something that that you do to turn it into what we call a time series. So then for everything that we were categorizing there, and for every communicator group that we were breaking out, there was another time. There was a time series made of that. So then, what the inferential tests that we ran were called Granger causality tests. Now, I want to talk about that because causality is kind of one of those incredibly loaded words when you talk about statistics, and it's important to say that we are not actually saying these are truly causal relationships. That's the important point: is that the Granger there is implying something different from true causality. But what it tells us in the basic scheme is that it's a test to build in a time lag between two different time series. In this case, we were comparing a time series for the same issue, but across each pair of communicators we're looking at. So for every one of those, we ran a test building in an inferred time lag to see if one, if a if communicator A creating more tweets about some issue, is associated down the road, with communicator B creating those tweets as well. So how we make sense of that is that in that case, if we see that relationship, we can say that for that issue, communicator A is leading communicator B. That's how we really would describe it: is that them talking about it more or less is associated later on with another communicator talking about it in the same direction.
Rosemary Pennington
So, what did you find? How influential were the social thoughts?
Josh Anderson
Phil, do you want to take that, or you want me to?
Phillip Arceneaux
Yeah. So, we-I'll add a little bit more context. So, we actually looking at the gender buildable theory. We did this at three different levels. So, the first level, which is what we largely just talked about, is what we call object salience. So, we looked at very simply in a binary fashion: was the issue present or was the issue not present? And that's what we we used in this initial level of analysis. Second level agenda building, which is what we call attribute salience. We look at the context of how that issue, if it was present, how it was mentioned. So not just is healthcare mentioned, but is it mentioned in a positive sense or in a negative sense? We used Luke 22, which gave us some sentiment scores, so we we operationalize that in terms of a sentiment score. So we were able to again attract was the was at time one was communicator A later on influencing the sentiment, which which this issue was mentioned by communicator B, and then third level agenda building is what we call network salience. So we look at co-occurring network. So, are there statistical patterns in which we see certain issues mentioned, certain stakeholders mentioned, and certain sentiment of mentions? Are there are there intentional networks that we see forming that are consistent across time? So, we looked at all three of these levels, and overall, we found that at first level agenda building, the social bots were effective to a certain degree at influencing what issues were being talked about, it was most prevalent among the campaigns. They were actually driving the campaigns more than any other actor group that we identified. They did influence to a certain extent the public, but the public also had. We looked at bi-directional influence. The public influenced them. We think that's actually very nascent, or that that's a very expected finding because they're trying to hide themselves. They're trying to obfuscate within public discourse. They don't want to be very different from it. So we we found that to be an expected finding. But it was really the fact that they were able to drive the the campaigns so much. The greatest extent of influence that we found from the bots wasn't that what issues they were pushing. It was the salience or the sent sorry the sentiment with which they were pushing the the bots were the most effective group group across all of the categories that we looked at in driving negative discourse of public policy issues. So not only were they influencing certain categories, but a larger bard of categories, they were really driving a negative political rhetoric that we saw online about these policy issues.
John Bailer
Okay, you know what? As I was listening, this is a huge amount of work. I mean, you know, kudos to to your to the team that worked on this. There's there's a lot of nuance to it, and there's there's a lot of challenges when doing something like this. I mean, you mentioned that you're looking at all pairs of communicators, but that's also for all the dictionaries that you're doing, right? So you're looking at so so you've got a lot of things you're testing, and you know, one of the questions that that comes up is is just, do you think this is going to be reproducible when others look at this in other contexts, or even if you were to look at this in a future election? I mean, assuming you can get the data. I mean, as you know, there's some there are challenges as systems and access to data on systems changes. So, so my big question is is your sense of kind of what would be next for you, and and how might that tie into reproducibility?
Phillip Arceneaux
Do you want to take the first stab, Josh?
Josh Anderson
Yeah, absolutely. Yeah, I mean, just to get it out of the way, yeah, there's a reason that we talk about. I talk about this as a process of how you do this kind of research, and not as a model, because in actuality, I think that the total number of models we ended up running were about 2500 separate models. I mean, that's that is a huge number of statistical tests, so what's important about that is that they're not. It's not necessarily what what what adding multiple comparisons to to answering one question because really what it is is we're asking you know about that many different questions, so there's not necessarily an expectation that for every issue there would be like the campaigns would be leading the press. We'd probably see it for some issues and not for others. Why? Well, you know that could depend on a lot of just the nature, the contextual nature of the of the race that, in essence, wouldn't necessarily be reproducible to other races. Now, that's actually probably the biggest issue in reproducibility has to do with data availability. So the time that we did this work is so important to understand like how we were even able to do it. Is that this was done before changes to the Twitter slash X API came after Elon Musk's takeover of the platform, that really largely made it much, much, much more difficult to collect data. I mean, you know, we collected initially. I think it was before it got narrowed down to due to retweets and other things that happened in data cleaning. It ended up being about 10 million tweets. I think. Oh my lord! Initially, yeah. So it was it was a huge, huge number of data that took you know many days to just collect, even working through the more permissible API rules at the time. I mean, now that to collect a data set of that size, it would cost it would just cost an enormous amount of funds to to collect that. Even worse for reproducibility, though, is that the system we used for the bot detection, that that confidence testing software, that I still don't. I'm not aware that that's even functional at this point, because that requires querying the API that was run through actually just the creator's personal API account. I mean, it was when you ran that software that went through somebody else's account who was gracious enough to let other researchers use that, and you know that's something to run that system. Now I I don't know if anybody who would be able to that would be an enormous use of resources to be able to do something like that.
Rosemary Pennington
Well, that's all the time we have for this episode of Stats and Stories. Phil and Josh, thank you so much for joining us today. Thank you.
Phillip Arceneaux
Thank you for having us.
John Bailer
Thank you.
Rosemary Pennington
Stats and Stories is a partnership between the American Statistical Association and Miami University's Departments of statistics, and media, journalism, and film. You can follow us on Spotify, Apple Podcasts, or other places where you find podcasts. If you'd like to share your thoughts on the program, send your email to statstories@amstat.org, or check us out at statsandstories.net. And be sure to listen for future editions of Stats and Stories, where we discuss the statistics behind the stories and the stories behind the statistics.