The icon indicates free access to the linked research on JSTOR.

Big data may have helped us address minor issues like climate change and famine, but it has yet to solve two of the most pressing problems of our time: finding true love, and then figuring out what you and your true love can watch together on Friday night. But we can tackle both problems once we recognize that they’re two sides of the same coin.

JSTOR Daily Membership AdJSTOR Daily Membership Ad

Who can tell me what to watch with my honey?

Let’s start with the second problem first, since it’s notionally easier. Lest you think that choosing a Friday night flick is a trivial matter, let me point you to a 1976 article on “Television Watching and Family Tensions,” in which Paul C. Rosenblatt and Michael R. Cunningham note that:

Television watching might be a useful device for dealing with tension between spouses, since American marriage norms seem to value togetherness in a way that makes it difficult for spouses to maintain places in their residence in which they can be apart from each other (Rosenblatt and Budd, 1975). Joint television watching can provide to the couple the appearance of togetherness, and if one or more of the combatants is watching and listening to the television set, there is less opportunity for the highest levels of tense interaction.

At the same time, disputes over what and how to watch can contribute to marital tension and gender inequality, as mapped out by Alexis J. Walker’s 1996 study of how husbands and wives share the remote control, in which she documents that:

two thirds of the women and three fifths of the men reported that there were things about their joint television watching that were frustrating to them…Women complained about their partners’ grazing behavior, both during a show and when they first turned on the television set…Men were more likely than women to say that they usually hold the RCD [remote control device] or have it near them…In half the couples (n=16), according to both women and men, men have the RCD.

If you think that the Netflix era has done away with the petty bickering of the cable TV era, think again. Nowadays you have to choose whose wishlist you’re going to dial up when it comes time to put the Netflix back into Netflix and chill. If you’re in a committed relationship, maybe you’ve even set up a “couple” profile in Netflix so that the service can theoretically learn your collective taste profile and start recommending programs you’ll both enjoy.

But why can’t Netflix do us one better, and recommend joint viewing options based on our separate profiles? If Netflix’s “Top Picks for Alexandra” includes Jeff Dunham’s Relative Disaster, and the same comedy special shows up in my husband’s top picks, then why doesn’t it show up in the list of top picks under our Alex & Rob profile? (Let’s not touch the more worrying question about why a ventriloquist is the only thing Netflix is recommending to me and my husband.)

It’s time for all these services that are aggregating our data for their benefit to start aggregating it for ours. You know what I like, and you know what my husband likes: is it so much to ask that you suggest what we’ll both like?

I’m not just picking on Netflix here. I’d like to make the same request of Spotify (dear God, there must be some middle ground between my show tunes and his 80s bands), Amazon, Goodreads, and every other digital mindreader that makes suggestions based on my past purchases or reviews. Take what you know about us separately, and help us make new discoveries together.

What’s wrong with online dating

Once media and e-commerce companies build up the taste profiles and data sets that allow them to recommend date night movie choices and save couples from marriage-ending fights for the remote, they’ll be ready to tackle an even more lucrative market: matchmaking. Sure, there are plenty of online dating services that already bring data into the matchmaking process: eHarmony, Plenty of Fish, and OKCupid all emphasize the role of data in their ability to find you your perfect mate. But there’s one common problem with all these services: they’re all looking at the wrong data.

In a fascinating, comprehensive, and readable analysis of the research into online dating, Finkel et al. map the manifold failings of online dating as it stands today. If you are looking for some solid evidence you can use as backup whenever your friends try to convince you to try online dating, this article has you covered for your next two hundred arguments.

Unfamiliar with the ins and outs of online dating? Finkel et al. have got that covered, too. They briefly summarize the state of the art as it stood in 2012—the world of online dating before Tinder introduced us to swiping right, but otherwise not too different from today. They note that dating sites offer three key types of service: access, communication, and matching.

Online dating offers the service of access because it gives users “access to a larger number of potential partners than anybody could have access to in the offline world.” Online dating offers the service of communication because services provide different kinds of online interaction ranging from “winks” to instant messaging and web chat. But what’s most relevant for the Netflixes of the world is that online dating sites also promise the service of matching. As Finkel et al. put it,

Matching refers to a site’s use of a mathematical algorithm to identify potential partners, called “matches,” for their users. These matches are presented to the user not as a random selection of potential partners in the local area but rather as potential partners with whom the user will be especially likely to experience positive romantic outcomes. A key assumption underlying matching algorithms is that some pairs of potential partners will ultimately experience better romantic outcomes, in the short term or the long term (or both), than other pairs of potential partners because the individuals are more romantically compatible from the start. Another assumption is that the seeds of this compatibility can be assessed using self-reports or other types of individual-difference measures before two people even become aware of each other’s existence.

Here’s where things start to fall apart for the dating sites of the world, as Finkel et al. subject both their claims and their assumptions to the harsh reality of what we actually know about love and marriage.

Let’s start with the self-reported profile data that drives the date-finding process. As Finkel et al. note, “[t]he information users present in their profiles tends not to be entirely veridical.” That’s psychologist-speak for “bitches be lyin’.” Bitches and bros, it turns out: the authors cite a study that found that four out of five online daters had lied in their profiles about their weight, height, and/or age.

But that’s of less concern than the fundamental problem with matching people based on either similarity or complementarity, whether through their own selection or the site’s recommendations. In the case of sites that present profiles so that users can select their own matches, would-be lovebirds are likely to make terrible choices.

First, the experience of “shopping” from so many options may “undermine users’ willingness to commit, or to remain committed, to a particular partner.” Second, while “[p]eople typically believe they know what they desire in a potential partner,” a variety of studies show that “they may not be especially good at discerning which characteristics will uniquely inspire their attraction and relationship well-being.” (This is psychologist-speak for “why your best friend keeps dating jerks.”)

This is—theoretically—where matchmaking algorithms can improve on human intuition. Just as driverless cars promise to save us from making a dangerous turn on an icy road, matchmaking sites promise to save us from making a wrong turn on the road of love. Before we buy that promise, however, Finkel et al. encourage us to ask whether:

long-term romantic outcomes are predictable at all. Matching sites like eHarmony strongly imply that the likelihood of any two people having a successful intimate relationship is knowable in advance. In their pursuit of this goal, matching sites are standing on the shoulders of giants. Predicting the long-term success or failure of intimate relationships, and marriages in particular, has been called the “Holy Grail” of scientific research on relationships.

And it turns out that this particular Holy Grail has so far remained out of reach. Finkel et al. neatly take down the various efforts at predicting divorce, pointing out their methodological flaws and low replicability. They then move on to analyzing the various studies that attribute relationship success or failure to individual characteristics, interaction quality, or life circumstances. They cite a cross-national study that found that “each partner’s personality accounted for approximately 6% of the variance in his or her own relationship satisfaction and between 1% and 3% of the variance in the other partner’s relationship satisfaction.” They note that “support for the idea that greater similarity of attitudes and values benefits relationships remains weak and inconsistent.”

Compatibility models that focus on complementarity rather than similarity have failed just as spectacularly. As an example, the authors note that “there is no evidence that introverts are uniquely attracted to extroverts.” When it comes to predicting relationship outcomes, the only consistent research finding is that “some people are better at sustaining intimacy than others, regardless of their partner,” so “[a]ssessing these characteristics can potentially act as a screening device.” In other words, dating sites can tell you who is undateable—they just can’t promise to find you the right person for you.

Can digital matchmaking do better?

If it’s so hard to predict relationship outcomes based on the data models that matchmaking sites use today, what makes me think that Netflix, Amazon, or Spotify could do any better? Let’s start with the fact that these sites have access to behavioral data, which is going to be more accurate than self-reported data: that cute woman you’ve been flirting with may tell you she’s a hopeless romantic who makes a ritual of watching Casablanca every year, but Netflix knows she’s actually binge-watching horror flicks.

Next, let’s look at the problem of predicting compatibility from personal traits. It’s a much tougher problem. If Finkel et al. paint a compelling picture of the limitations in predicting relationship outcomes as of 2012, they don’t have much to say about what we can predict in 2018. That’s because the volume of data available to make predictions has exploded in the past five years: a 2014 RAND report on the “data flood” estimated that the digital universe contained five hundred million terabytes of information in 2012, and projected that would grow to ten billion terabytes of information by 2017. Simply put, Netflix and Spotify and Facebook and Amazon know way more about us than the yentas and matchmakers of yore. And they could be using all that data to get us laid.

Here’s how that could work, sticking with the Netflix example. The company introduced the option to have five separate user profiles in August 2013, which means it’s now got more than four years’ worth of viewing data in which at least some households have separated their viewing into his and hers, his and his, or hers and hers. Layer that information with public records on marriages and divorces, transactional data indicating relationship status or (if those options give you the creeps) changes in profile makeup. (Like when “Brad” suddenly disappears from the account, but eight months later, “Chad” appears.) Suddenly you’ve got a massive dataset that can look for commonalities in viewing patterns and match that with relationship duration.

That dataset may well be able to reveal patterns in relationship outcomes that have previously eluded social scientists, even if we aren’t sure why we are seeing what we’re seeing. I’m not going to pretend that a shared love of Star Trek causes happier marriages (though it’s working for me), or that complementary music tastes are going to lead to better sex. What I would argue is that any large-scale correlations, even they’re mysterious, may well outperform all other intuitions about relationship compatibility.

If Netflix can see that lesbians who watch documentary films have longer-lasting relationships when they shack up with women who watch reality TV shows, doesn’t that seem like information worth having? If Amazon can see that people who listen to comedy audiobooks are much less likely to buy The Divorce Survival Guide if they’re sharing an account with someone who eats so much Chunky Soup they need an instant reorder button, doesn’t that seem like it could help all those soup eaters find true love?

As someone who’s been married almost as long as online dating has been around, I can’t pretend to have a dog in this fight: I’m not looking for Spotify to find me a new husband or for Goodreads to find me a nice side piece. But I would love to live in a world in which the dataset that we old married people are compiling on our nights in could be useful in matching up all the single people who are tired of their nights out. Failing that, Netflix, at least you could keep us from fighting over the remote.


JSTOR is a digital library for scholars, researchers, and students. JSTOR Daily readers can access the original research behind our articles for free on JSTOR.

Journal of Marriage and Family , Vol. 38, No. 1 (Feb., 1976), pp. 105-111
National Council on Family Relations
Journal of Marriage and Family, Vol. 58, No. 4 (Nov., 1996), pp. 813-823
National Council on Family Relations
Psychological Science in the Public Interest , Vol. 13, No. 1 (January 2012), pp. 3-66
Sage Publications, Inc. on behalf of the Association for Psychological Science
From the Book Data Flood: Helping the Navy Address the Rising Tide of Sensor Information , 2014
RAND Corporation