Dr. Liesel Sharabi studies the data science of love, including the ways that algorithms and artificial intelligence (AI) help to facilitate intimate relationships. She has written about matchmaking algorithms for the Harvard Data Science Review and discussed their use in online dating with media outlets like TIME Magazine, WIRED, and The Wall Street Journal. She is currently an associate professor in the Hugh Downs School of Human Communication and director of the Relationships and Technology Lab at Arizona State University.
Episode Description
According to the Pew Research Center, three in ten US adults say they've used a dating app, with Tinder, Match and Bumble being the apps most likely to have been tried. Pew's research has also found that one in 10 partnered adults in the US met their significant other on a dating app or site. Dating app success is a focus of this episode of Stats and Stories with guest Dr. Liesel Sharabi.
+Full Transcript
Rosemary Pennington
According to the Pew Research Center, three in ten US adults say they've used a dating app, with Tinder, Match and Bumble being the apps most likely to have been tried. Pew's research has also found that one in 10 partnered adults in the US met their significant other on a dating app or site. Dating app success is a 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 Miami University's departments of statistics and media, journalism and film, as well as the American Statistical Association. Joining me is regular panelist John Bailer, emeritus professor of statistics at Miami University. Our guest today is Liesel Sharabi. Sharabi studies the Data Science of Love, including the ways that algorithms and artificial intelligence help to facilitate intimate relationships. She's written about matchmaking algorithms for the Harvard Data Science Review, and discussed their use and online dating with media outlets like Time, Wired and the Wall Street Journal. Sharabi is currently an associate professor in the Hugh Downs School of Human Communication, and director of the Relationships and Technology Lab at Arizona State University. Thank you so much for joining us today.
Liesel Sharabi
It's wonderful to be here.
Rosemary Pennington
How did you get started in studying dating apps?
Liesel Sharabi
Yeah, I mean, I think that there's been this shift that's happened over the past couple of decades, where, if you're interested in romantic relationships, and interested in people's decisions in terms of how they're selecting partners, how they're forming relationships, it's only natural that you're going to gravitate towards the dating app space, because it's just how so many people are doing these things today.
John Bailer
Liesel, you recently contributed an article Finding Love On a First Date, matching algorithms and online dating that appeared in the Harvard Data Science Review. What kind of was the purpose of this piece? What inspired you to do this?
Liesel Sharabi
Yeah, so in this piece, I wanted to look a little bit at the history of matchmaking, online dating and how some of these matchmaking systems have progressed over the years. And so looking at how far we've come, starting from some of the first online dating sites that were very much oriented around profiles, and people going through and finding partners they were interested in and having to narrow the pool themselves, all the way up until what we have now, which are dating apps, where you have algorithms that are trying to figure out people's tastes and make recommendations based on their swiping behaviors.
John Bailer
You know, one thing that struck me, and I confess to not using this dating app, and if I did any other confession, my wife would kill me. So, you know, as I was looking at this, I saw that some of the apps have a self-selection component, and others have algorithmic selection. When you're thinking about pairings and partnerings, and you know, I guess maybe I knew that, but I didn't really know that. So could you talk about that difference between self-selection versus algorithmic selection?
Liesel Sharabi
Yeah, so I think it helps to look a little bit at the history of online dating and where we've been in sort of how we've arrived at this point. So historically, the way a lot of these algorithms worked was that they very much relied on self report data. So having the user, you know, tell the platform, “this is what I'm looking for, this is who I am,” and then they would go out and try to find it. But the problem with that is that we know that people aren't always such good judges of what they're going to find attractive in person. So if you do actually deliver on what it is they're looking for, it's no guarantee that that's actually going to work out for them. So more recently, just within the past 10 years, and the shift to mobile dating, a lot of these apps have started looking more at ways that they can infer what people are looking for from their behavior. And so instead of having them self report their preferences, they're trying to arrive at that judgment based on how people are swiping through profiles.
Rosemary Pennington
So John, unlike you, I actually have experience with dating apps when I was in college. After a very nasty breakup, I attempted to try Match, and went on two very unsuccessful dates because they were individuals that seemed great until I showed up in person, and they were not great. And then I did not ever use them again. But when I was in grad school, I remember being at a conference, and the young woman across from me at a poster session was talking with us about Tinder. I think she had done some research on Tinder. And my co-author and I were asking her all of these questions about it. And we're fascinated by the way she was just sort of flipping through and suddenly that could be a match that will pop up. And I'm like, how did that happen? Like, it's what is happening with these algorithms like that's what I'm so fascinated by like, what is it? What is going on with these algorithms that it's sort of you guys, you're swiping through like something is digging, and like, you're matching.
Liesel Sharabi
I mean, part of it is that you're ahead of a lot of people in the sense that you know that there's an algorithm that's determining what recommendations you're getting. And I think some people think that especially on dating apps, it's location based, so you're seeing everybody in your vicinity. And that's not necessarily true. So there are algorithms working behind the scenes to try to narrow that dating pool down. And so the way that a lot of dating apps are doing this is using something called collaborative filtering, which is essentially making recommendations based on users who appear to have similar tastes. So if you have two people, person A and person B, and they're swiping on the same types of profiles, you know, the next person that person A swipes on might be recommended to person B as a possible match. Because it's essentially saying, Okay, you seem to like the same types of partners. And so whoever they swipe on next could be a good recommendation for you. And so this is the same kind of technology that you see on other platforms. I mean, it's how Amazon recommends products, it's how Netflix recommends movies, where it doesn't necessarily have to know a lot about you, in order to start making inferences about the types of partners that you might be attracted to.
John Bailer You know, one thing that is, we'll start talking a little bit about the study and some of the work that you've been doing, and this, this idea that whether you think it's going to work may matter about whether it works. And I, you know, coming from a background that had a lot of bio statistical background, talking a lot about the placebo effect. I thought, I wonder if this is like that, that later on in your paper, you describe that. So could you sort of help set the stage for the kinds of questions that you wanted to answer with some of the work that you've done?
Liesel Sharabi
Yeah. And so these dating apps, online dating sites really vary in terms of how central the algorithm is to what they're doing. And on some platforms, if you're using it, it would be really hard to not know that there was an algorithm making recommendations, because it's part of their marketing. And it's part of their advertising. Like I think of the eHarmony algorithm, for example, or OKCupid, where it's essentially what people sign up for, it's what they pay more money for. And so they're very mindful of the fact that it is part of what is causing them to see certain partners show up as recommendations. And so in doing that, and in making the algorithm so central to what some of these platforms do, they're essentially creating high expectations. And some people, I think, really buy into that idea that an algorithm is going to be a better judge of compatibility than they are. And some of my research has shown that, to the extent that you do believe that you tend to have better outcomes. So the belief system behind it can be incredibly powerful.
John Bailer
Yeah, so you're talking about some of these things, and the one that really cracked me up was the ELO scores. Yes, I'd heard about this from chess, and rankings of chess, but the idea that there could be the equivalent for matchmaking, based on you know, I don't know, a grandmaster of love. What is that?
Liesel Sharabi
Yeah, so in chess, these scores are essentially you're getting a score based on your wins and losses based on how skilled your opponents are, and dating apps have started incorporating some of that scoring into their own recommendation system. So when someone swipes right on you to indicate that they're interested, that is essentially, you know, helping boost your score, when they swipe left to reject, then, you know, that would be decreasing it. And it also depends on who's doing the swiping. So somebody who has a higher rating, who swipes right to indicate interest would also do more to enhance your score. So essentially, what some of these platforms have been trying to do is figure out what your desirability level is, in order to use that to then make recommendations. And the reason that they're doing this is because they're also interested in reciprocity, it's not enough to know that I'm interested in someone, we also have to have some indication that that person would be interested in me in return. And so they're trying to find different ways of going about gathering that information.
Rosemary Pennington
As I was listening to you, it's someone who is on the site who is sort of on the market. If you get matched more often, do you get a higher ranking or you're like a five star rating like an Uber or something? I don't know how desirability is measured in these spaces? Yeah.
Liesel Sharabi
And I mean, it's a tough question to answer because with all of this, the algorithms are proprietary, and platforms tend to be incredibly secretive about how they're actually going about matching people. And some of that is because of pushback they've received in the past, where people don't necessarily like the idea that they're getting a desirability rating; that doesn't sit well with a lot of people. And so we can, we can speculate about what they're doing. But it's hard to know for sure, with the exception of platforms that have been more public about how they actually go about this process. But to your point, I mean, I think it stands to reason that they're looking at how many swipes you're getting. And they're also looking at who's doing that swiping to try to get a sense of where you fit into their dating pool, and the types of partners that they should be showing you.
John Bailer So I want to follow up on one of the points that Rosemary was making about just trying to measure things. I mean, just trying to measure attraction, anticipated future interactions, disclosures, partner selections, uncertainty, you know, what, can you talk a little bit about? How do you decide on what are some of the important predictors? And then ultimately, given some of that decision and identification? How might you go about actually measuring them?
Liesel Sharabi
This is a really great question. And the way a lot of platforms are doing this is by looking at matching rates and looking at messaging rates. So how many matches are you getting? How many messages are you exchanging, and then they're looking at the number of messages exchanged as an indication of whether you were interested enough to actually have a conversation. So when it comes to the platforms, their metrics of success are primarily based on what you're actually doing on the platform, because that's the data that they have access to. I think another important marker of success is what happens after you actually meet a person. And that's where their data becomes more limited, because they don't necessarily know. And so, in my research, things that I've been looking at are, you know, how attracted were you to someone after you actually met them face to face? Did you want to see them for a second date? Were you interested in pursuing a relationship? So getting some self report data from them beyond just knowing you know how many times they message somebody on the platform, because a lot of times those messaging exchanges never actually lead to a date. And if they do, they don't necessarily lead to a relationship. And so I think that, you know, people have different motivations for using online dating. Of course, not everyone is looking for something serious or long term. But I think that for the most part, people are looking to have a good first date to meet somebody face to face, who they're interested in. And so I've been trying to look more at those outcomes as a metric of success.
Rosemary Pennington
You're listening to Stats and Stories. And today we're talking with Liesel Sharabi about the data science of online love. We've been sort of talking around this, but actually, what are you doing when you're studying? And if the algorithms are black boxed? In many ways? How can you really go about studying how successful they are?
Liesel Sharabi
Yeah. And so that's part of what's led me to look at people's perceptions. I mean, without a platform telling you specifically how their algorithm works, it's very difficult to actually test its effectiveness. And so there are scholars who have been trying to determine whether it's even possible that an algorithm would be able to predict compatibility based on the inputs that we're assuming they have. But at the end of the day, it's really difficult to know for sure, without knowing what they're doing, but something that we can look at would be people's perceptions of the process, people's expectations of what compatibility matching is capable of. And then looking at how that affects their behavior and some of the long term relationship outcomes.
John Bailer One of the things that I've found pretty interesting is things like housing at University has changed. There's compatibility matching now among roommates. You know, that there's often this kind of algorithmic matching or suggestions for matching among students going to campuses that haven't identified in advance someone that they'd want to share a space with. I wonder if there's data that I would think would be really interesting about whether or not matchings worked, you know, at the end of the year, you know, I despise this person, I wanted to throttle them at least halfway through, I guess that that algorithm prediction wasn't so hot. I mean, there seems like there's a way to do a little bit of validation, that without the same kind of, protection, of proprietary information that you see, in some of the dating apps. Do you see any possible kind of external validation in this other kind of matching context? It's not romance, but it's certainly congeniality.
Liesel Sharabi
Yeah, definitely. And something else that I think is so interesting about studying recommendation systems and online dating is that they do have implications for all of these other situations where matchmaking is happening. So it's happening with roommates on college campuses, it's happening. I mean, I've read papers where they're using the OKCupid algorithm to do rare disease matching. So they're using it in medical settings to try to identify people with similar types of symptoms. I've heard about the Hermes algorithm being used in job searches and trying to match prospective employees with employers. So there are all of these other situations where you have a large pool of people, and you're trying to narrow it down, and you're trying to, you know, deliver recommendations even beyond the dating context. So, I mean, we all want to find love. We all want relationships. And so I think that in itself is important. But also if we can get the matchmaking part of this right, I do think there are implications for so many other areas where these kinds of processes play out.
Rosemary Pennington
These companies often have their own internal data and studies that they're doing. I wonder, has there been any response from these companies to the work that you and others have been doing on how successful they are?
Liesel Sharabi
Yeah, I mean, not that I'm aware of. I think a lot of platforms try to avoid these types of things completely. There’s no upside now. And I mean, part of it too, is that the more actual users are aware of how the matchmaking system is operating, the better position they are to try to game the system. And you hear about people doing this as well, where they try to figure out how these recommendations are being made. And then they do things to try to manipulate the system. And when you have people doing that, that also creates problems, both for them, and also just for the system more broadly. And so there are some valid reasons as well as why they wouldn't want to be incredibly public about how they're going about this.
John Bailer So as you've been doing this work, I imagine there have been some things that really surprised you that, you know, you I didn't expect. Maybe there's some things that validated what you had expected going in, but I'm curious about the surprises that have possibly emerged in doing this work.
Liesel Sharabi
Yeah, I mean, I think one of the biggest surprises, and what really led me down this path and studying recommendation systems is how important people's perceptions are to their experiences in online dating. And I think this is significant. When you think about the impact that these algorithms are having on people's lives, like the decision, you know, the choice of romantic partners, one of the most important decisions arguably a person would ever make, that's going to be a life partner, someone you know, who's going to be really central to their experiences moving forward, and you have algorithms that are essentially driving some of those decisions. And I've had participants in some of my studies tell me about how much importance they've placed on those recommendations in terms of whether they decided to pursue a relationship with someone or not. And I think that's really noteworthy, especially when we're having this conversation about whether an algorithm, you know, can even make those kinds of recommendations effectively. But when people think that it can, it really does shape their lives in significant ways.
John Bailer Are there certain characteristics for people of why they select certain online dating tools? I mean, so yeah. So can you talk a little about how people select certain tools or what maybe it's not, it's not completely exclusive. But there's sort of why people gravitate to one towards one versus another.
Liesel Sharabi
I mean, this is also part of what makes it so difficult to study how effective their algorithms are. Because there's also the selection effects that happen, where people gravitate towards some platforms, as opposed to others because they're looking for certain types of relationships. So there are some platforms that are very much marketed towards people that are looking for long term commitment. So you can imagine that the people that are signing up to use them are also going to be more motivated to make relationships work. So is it that the algorithm works? Or is it that those people were just more motivated to find a relationship in the first place. So that's also a tricky part of all of this. And then to add to that a lot of people also aren't just using one platform, they are on multiple platforms simultaneously. And so sometimes they're seeing the same types of partners. And so that adds another layer of complexity. But I kind of think about dating apps, it's just a modern version of a singles bar, where each one has its own culture, it attracts its own certain type of crowd. And so people kind of hop from one to another based on what they're in the mood for and what they're looking for.
Rosemary Pennington
This is coming out right before Valentine's Day. I wonder if we are going to be able to turn on the TV or open the newspaper and see countless articles about love and relationships. And I'm sure there'll be some of these apps? I wonder what advice you would have for people who are going to come across these news stories about dating apps? Like, what should they be careful of what they should be watching for, to be able to judge whether the the news story is sort of legit, or it's or the reporter has done their due diligence and not just sort of cheerleading for another app?
Liesel Sharabi
Yeah, I mean, I think that there is a large literature around online dating, there's a lot of research that's been done on these platforms. And so I think we're at a point where there is some scientific evidence driving some of the claims that you're seeing in news stories about what's effective for people on what's not, the biggest piece of advice that I would have for people is, you know, if you want to improve your recommendations, if you don't like the matches that you're getting, one of the best things that you can do is pay attention to your profile, because that's really what it all comes down to, especially when we're talking about apps where they are making inferences based on how people are swiping. Because the profile just drives everything; it drives who is swiping on you, it drives how many swipes you're getting. And so I think some people think, okay, I need to game the system, I need to figure out what they're doing. And I need to try to beat it. That's going to be a tough game to win without really, you know, having insight or knowledge. So I think a better approach is to try to figure out how I can adjust my profile to attract the types of partners that I think I would be interested in? And also being mindful of like, is this a good representation of me, if I were to actually meet somebody face to face, because there's also this tendency to kind of exaggerate sometimes, and that might get you lots of matches, but it's not going to get you a lot of second dates.
John Bailer So I'm curious, how do you study online dating apps? What are some examples, if you could have a strategy, or as a type of study that you've done, where you're trying to evaluate the kind of performance of such systems?
Liesel Sharabi
Yeah, I mean, one way that you can study them is obviously by asking people about their experiences, which can be a little bit biased, of course, another way that you can study them is by actually looking at people's behaviors on the platform. And I think that's something else that's really interesting about the online dating context is that these platforms have so much data. And increasingly, people are having the ability to gain access to some of that data. So on some of these apps, you can download your data, you're able to access it. So that information can be used in research, also messaging exchanges. I mean, we have a record now of how people are initiating relationships from the moment they first make contact with someone all the way up to the point where they've actually formed a relationship. And it's just sitting there waiting for people like me to analyze it. And so this, this digital trace data that people are leaving behind every time they interact on the platform, I think that has huge potential for changing how we study relationships, and for helping us gain insight into what attracts people to each other.
John Bailer
Alright, as I was thinking about this, I was imagining kind of different trajectories for people that have sort of different characteristics coming in, you know, sort of, depending on the age in which you first encountered the system or start using a system, whether you had been, you know, your marital status prior to this, maybe you had been married before, maybe not, maybe divorced, maybe widowed? You know, what, what kind of see that there's a lot of potential straight cohorts that might interact with these online systems differently. Have you gotten any insights about whether some of these online systems work well, for certain subgroups of the population, but maybe not as well, for others? Is it based on some of the definitions of how you are working well, like a successful first date, ongoing second date, or whatever.
Liesel Sharabi
I mean, I think at minimum, they work differently. So when you look at people who are under the age of 30, about half of them use dating apps. So dating apps are incredibly popular among young people. And something that's different with them is that they've grown up with the technology, it's always been part of their dating experience, which, for me, at least, is kind of crazy to think about that we have people in their early 20s, who have maybe never had a relationship with someone that they did not meet through a dating app or through social media, because this is just always how things have been done. And then you compare that to somebody who's a little bit older, who maybe has had relationships before that didn't work out with people that they met face to face, and now they're trying to jump back into the dating pool, and everything has changed and you know, online dating, it's a whole new thing for them, and they have to reorient to how people are using these platforms and some of the norms and expectations. So I think that can be difficult for people and it can lead to differences, potentially, in how they're using these platforms. So yeah, I don't necessarily know if that means that one group would be more effective, but I do think that we could expect that there would be differences in use for sure.
Rosemary Pennington
These technologies have certainly changed. I mean, since,you know, I have been an adult and thinking of matches, essentially like replication of dating ads and newspapers right online with just photos. See like eHarmony is a matching algorithm thing, which I remember being like this big, big thing, right? Oh, suddenly, this is why we're different. And then to like things like Tinder and Grindr, which is just like swiping really quickly, given the work that you've been doing. I wonder, what do you think of the kind of the next big tech development that's going to be coming down the pipe when it comes to dating technologies?
Liesel Sharabi
Yeah, I mean, the last big shift happened about 10 years ago. And that was when we went from online dating sites to multiple dating apps. And not a lot has changed since then, up until now. And I think that now we're at this really pivotal moment in online dating because of AI. And I think that has the potential to change everything from the matchmaking systems, to how people actually engage with these platforms. And there's some startups that are doing some really interesting things with AI. There's one in particular, that is having people train AI chatbots, to represent them. So training them on their personality, and then sending them out to interact with other chat bots. And essentially go on a bunch of virtual first dates, and then come back and tell you if Hey, these are the people who might be interested in so instead of us sitting there swiping through profiles, you have your AI out there going on dates with other chat bots. And I mean, I think that kind of thing, it raises a lot of questions with one of the biggest being: how do you actually train AI to represent you? And is that going to be an accurate simulation? But I also think there's a lot of possibility there for changing the experience and making it so you're spending less time swiping and more time actually going out there and meeting people in person. Like, it seems like the more technology advances, the more it kind of takes us back in some ways to face to face interaction. And I'm hoping that's the direction that that would lead. But it's interesting to see some of the things that they're doing with AI currently.
Rosemary Pennington
I met my husband in a bar at a concert, and I can't imagine having a chatbot going out. That's just so incredible to me. Well, that's all the time we have for this episode of Stats and Stories. Liesel, thank you so much for being here today. Stats and Stories is a partnership between Miami University's Department of Statistics and media journalism and film, and the American Statistical Association. You can follow us on Twitter @StatsAndStories, Apple podcast, or other places you can find podcasts. If you'd like to share your thoughts on the program, send your email to statsandstories@Miamioh.edu or check us out at statsandstories.net, and be sure to listen for future editions of Stats and Stories, where we discuss the statistics behind the stories and the stories behind the statistics.