Out of My League: A Professor Looks at Dating’s ‘Matching Hypothesis’

Berkeley I School Professor Coye Cheshire

You’ve undoubtedly heard it before: don’t date someone who’s “out of your league.” Whether or not this is good advice, it’s a commonly accepted fact that people tend to gravitate toward partners of a similar social worth. There’s even a theory that says just that, called “the matching hypothesis,” which you probably remember from your Psych 101 class. People tend to seek out partners of a similar level of social desirability, not just in terms of physical attractiveness but also in terms of other qualities, like intelligence and personality.

The matching hypothesis is almost conventional wisdom, but large-scale online dating data gave four UC Berkeley researchers a new way to evaluate its claims.

In the mid-2000s, UC Berkeley School of Information professor Coye Cheshire, former Ph.D. student Andrew T. Fiore, along with Lindsay Shaw Taylor and G.A. Mendelsohn from the UC Berkeley Department of Psychology began to use large-scale data to investigate a variety of questions about romantic relationship formation in online settings. As they began to accumulate enormous amounts of data, the emerging field of data science gave them the ability to test a variety of different research questions—including the long-held tenets of the matching hypothesis. With the advent of online dating sites, researchers suddenly had a wealth of relationship data at their fingertips, and data science offered them the tools to look at this large-scale data with a critical eye.

There was certainly a lot to look at. For starters, it’s a common misconception that the matching hypothesis is about people pairing off based on their physical attractiveness. This isn’t actually the case; instead, Walster et al. (1966) posited that individuals are likely to partner up based on similar levels of self-assessed self-worth, asking the specific question of whether people select partners of “similar social worth.”

Since inherent self-worth is tricky to measure, a reductionist view of the matching hypothesis has led physical attractiveness to stand in for that self-perceived self-worth over the years. In fact, the attractiveness quotient is what most people tend to think of now when they hear the term “s/he’s out of your league.” Due to these misconceptions and the complexity of their research questions, Cheshire and his team opted to break the problem into four experiments:

  1. EXPERIMENT ONE: Are one’s feelings of self-worth correlated with the social desirability of target partners?
  2. EXPERIMENT TWO: Does a person’s physical attractiveness correlate with the physical attractiveness of the people they contact?
  3. EXPERIMENT THREE: Does the popularity of online dating site members (as measured by unsolicited messages received) correlate with how desirable they judge their partners to be? Does their popularity correlate with their partner’s popularity? Do one’s feelings of self-worth correlate with those of people s/he communicates with?
  4. EXPERIMENT FOUR: Do more popular individuals select others whose popularity matches their own? Are they selected by this group as well?

What was the end result? As it turns out, humans are apt to date “out of our league”…or at least attempt to. Think of the online dating site population as a virtual bar that spans the entire United States; as you might guess from your own experience, an initiator’s physical attractiveness is not directly correlated to the attractiveness of those they choose to contact. Instead, users tend to contact people who are more attractive than themselves. However, other portions of this experiment showed that individuals voluntarily selected similarly desirable partners from the very beginning of the dating process, demonstrating that part of the traditional matching hypothesis (partnering based on self-worth) does hold true. Different ways of assessing social value led to differing conclusions for these researchers.

The design of this experiment helped to measure a broader conception of self-worth and social worth on multiple dimensions, extending beyond just physical attractiveness. This is something that has been overly simplified in the field of psychology, and data science techniques applied to online dating data presented a unique way to use large-scale analyses to go back and reassess a long-held truth.

This was a complex, multi-level study, which could only be made possible by a collection of large-scale data and flexible research methodologies. Thanks to the volume of data and the variety of tools at their disposal, researchers have the ability to combine methodologies to tackle a problem from different angles, as the UC Berkeley team did upon discovering that many equate worth with attractiveness.

The results of the UC Berkeley team’s experiments are interesting, but they hold an even deeper meaning for prospective data scientists. With the massive amounts of data and tools we currently have at our disposal, it’s becoming apparent that researchers now have the ability to go back and test fundamental assumptions in academic fields like psychology.

What does this mean? Even those data scientists who don’t plan to work in academia now have the ability to add something to the public dialogue. Testing the matching hypothesis was a boon to both industry and academia; by partnering with an online dating site, Cheshire and his fellow researchers were able to challenge long-held truths while at the same time working to understand some of the underlying social mechanics of relationship formation in a thriving business. The benefits of this research are twofold: it can help with future designs in online dating systems, while the data collection reveals different things of great interest to academic researchers.

Data science presents an interesting crossroads for social research. While the aforementioned research scholars are not necessarily the ones at work designing systems in the private sector to collect data, data scientists themselves are able to get right in the thick of things to build, collect, and analyze data, all while redirecting research to answer new questions that arise in the course of an experiment.

This is exactly why collaborations between industry and academia are important—research centers like Walmart Labs and Target labs are eager to work with academic researchers who can bring the tools and knowledge of data science and complex social systems to bear on industrial experiments. By collecting data for practical, pragmatic purposes, the two industries can then review standard assumptions, giving back more to society than just an increase in Click-Through Rate (CTR) to any one company. Instead, alliances between academia and industry help researchers understand fundamental social processes, leaving everyone better off.

To find out more about this study, view Taylor, Fiore, Mendelsohn, and Cheshire’s original paper: “‘Out of My League’: A Real-World Test of the Matching Hypothesis.” (PDF, 533kb)