Dataclysm: Who We Are (When We Think No One's Looking)

by Christian Rudder

  Below I’ve plotted new messages received per week, by the recipient’s physical attractiveness:

  The sharp rise out at the right smashes down the rest of the curve, so its true nature is a bit obscured, but from the lowest percentile up, this is roughly an exponential function. That is, it obeys the same math seismologists use to measure the energy released by earthquakes: beauty operates on a Richter scale. In terms of its effect, there is little noticeable difference between, say, a 1.0 and 2.0—these cause tremors that vary only in degree of imperceptibility. But at the high end, a small difference has cataclysmic impact. A 9.0 is intense, but a 10.0 can rupture the world. Or launch a thousand ships.

  What you definitely can’t see in the chart above, because I aggregated the data to obscure it, is that men and women experience beauty unequally. Here is that OkCupid message density, split out by gender, with the aggregates as the dotted line in the middle.

  It’s hard for me to convey how much attention the upper-right corner of this curve entails, short of tracking you down and screaming in your face about my hobbies. Especially in larger cities, where the message flow is 50 percent higher than even what you see above, a woman at the top of the scale has something like a term paper’s worth of hey-what’s-up-do-you-like-motorcycles-because-I-like-motorcycles waiting for her every time she comes to the site. A dudeclysm, if you will. However, neither beauty’s effects, nor the male/female split, are confined to the sexual realm.

  Here is data for interview requests on Shiftgig, a job-search site for hourly and service workers:1

  And for friend counts on Facebook:

  Success and beauty are correlated for both sexes, but you can see that the slope of the red line is always steeper. On Facebook, every percentile of attractiveness gives a man two new friends. It gives a woman three. On Shiftgig, the curves aren’t even comparable in this way. The female curve is exponential and the male is linear. Moreover, they hold whether the hiring manager, the person doing the interviewing, is a man or a woman. In either case, the male candidates’ curves are a flat line—a man’s looks have no effect on his prospects—and the female graphs are exponential. So these women are treated as if they’re on OkCupid, even though they’re applying for a job. Male HR reps weigh the female applicants’ beauty as they would in a romantic setting—which is either depressing or very, very exciting, depending on whether you’re a lawyer with a litigation practice. And female employers view it through the same (seemingly sexualized) lens, despite there (typically) being no romantic intent.

  It is hardly fresh intellectual ground that beauty matters, and that it matters more for women. For example, a foundational paper of social psychology is called “What Is Beautiful Is Good.” It was the first in a now long line of research to establish that good-looking people are seen as more intelligent, more competent, and more trustworthy than the rest of us. More attractive people get better jobs. They are also acquitted more often in court, and, failing that, they get lighter sentences. As Robert Sapolsky notes in the Wall Street Journal, two Duke neuropsychologists are working on why: “The medial orbitofrontal cortex of the brain is involved in rating both the beauty of a face and the goodness of a behavior, and the level of activity in that region during one of those tasks predicts the level during the other. In other words, the brain … assumes that cheekbones tell you something about minds and hearts.” On a neurological level, the brain registers that ping of sexual attraction—Ooh, she’s hot—and everything else seems to be splash damage.

  To my second point, that beauty affects women in particular, Naomi Wolf’s bestseller The Beauty Myth showed that better than I ever could. In short, my raw findings here are not new. What is new is our ability to test ideas, established ones, famous ones even, against the atomized actions of millions. That granularity gives strength and nuance to previous work, and even suggests ways to build on it.

  The paper “What Is Beautiful” was based on a research sample of only 60 subjects—barely adequate to prove the effect, let alone its many facets.2 But now we can go from “What Is Beautiful Is Good” to asking “How Good?” and in what contexts. In sex, beauty is very good. In friendship, it’s only somewhat good, and when you’re looking for a job, the effect really depends on your gender. As for Wolf’s seminal work, we can confirm the truth behind her broad observation that “today’s woman has become her ‘beauty’ ”—three robust research sets agree that the correlation is strong. And, better, we can extend some of her most cogent arguments about beauty being a means of social control. Think about how the Shiftgig data changes our understanding of women’s perceived workplace performance. They are evidently being sought out (and exponentially so) for a trait that has nothing to do with their ability to do a job well. Meanwhile, men have no such selection imposed. It is therefore simple probability that women’s failure rate, as a whole, will be higher. And, crucially, the criteria are to blame, not the people. Imagine if men, no matter the job, were hired for their physical strength. You would, by design, end up with strong men facing challenges that strength has nothing to do with. In the same way, to hire women based on their looks is to (statistically) guarantee poor performance. It’s either that or you limit their opportunities. Thus Ms. Wolf: “The beauty myth is always actually prescribing behavior and not appearance.” She was speaking primarily in a sexual context, but here, we see how it plays out, with mathematical equivalence, in the workplace.

  As I’ve mentioned before, I have a young daughter, and in our rare downtime, Reshma and I will speculate about her and her life and where it might lead. All parents do this—give them a quiet moment and it’s inevitable, just like two drunks in a bar will always argue. Every family must have their own particular flights of fancy, but ours go more or less like most, I imagine. My wife or I will start, it doesn’t really matter who: Our little girl’s going to be so smart. Oh yes, we’ll teach her everything we can. She’ll be so gentle, so good-hearted. These things are very important to a good life, we agree. And of course, look at that skin, like chai, those eyes, she’ll be so pretty. I mean, wow. Yeah, we’ll have to put locks on the doors when she’s a teenager. And there the conversation takes a little turn. But not too pretty, right? Yeah, we wouldn’t want that. We both sit back, and the conversation moves on to something else. This is what it comes down to: I can’t imagine anyone wishing limits on a son.

  Unfortunately, it’s a problem the Internet is surely making worse: for The Beauty Myth, social media signals Judgment Day. Your picture is attached to practically everything, certainly every résumé, every application, every byline. If people care about what you are doing, they will find out what you look like. Not because they should, but because they can—Facebook and LinkedIn have essentially extended OkCupid’s Love Is Blind problem to everything. Even just ten years ago, it was almost impossible to tie the average person’s name to her photograph; now you just Google the words—everyone does—and up pops a thumbnail from a social network. We’ve all had to pick through snapshots for that “best” one. Choose wisely, friends, because it defines you in a way it never has before. There’s a momentum to the trend that might not be obvious to people who work outside the industry. The new design standard of the last two or three years, more open and more photocentric—what I think of as “Pinteresty”—is making not just pictures, but beauty specifically more important. OkCupid recently made a change for some photo displays, going from the size of the black box to that of the red, below:

  The designers just wanted the page to look more modern. What they didn’t anticipate (and later had to mitigate) was the following: all those extra pixels allowed the pretty faces to outshine the others all the more. The rich got richer. It was the web-design equivalent of American domestic policy.

  Given this pressure it’s no wonder that body-image blogs are so prevalent. And that posts tagged like #thinspiration #thinspo #loseweight #keeplosing #proana #thighgap became so common that both Tumblr and Pinterest (independent of each other) had
to alter their Terms of Service to ban this kind of content. If you’re wondering what the last two hashtags are, #proana is short for “pro anorexia”—people in favor of starvation as a weight-loss technique. Meanwhile, #thighgap refers to having thighs so thin that they do not touch when you stand with your feet and knees together. It’s a trait fetishized by teenage girls. Quite apart from the questionable desirability, it’s biologically impossible for most of them. The full depravity of the phenomenon can’t hit you until you search for these tags yourself and are confronted with an unending page of broken bodies tilting at the camera—not only are the “inspiring” women deathly thin, they are also frequently in lingerie, bikinis, underwear. The blogs, created by women, are truly the epitome of the male gaze—and I say this as a person reflexively skeptical of the language of the academic left.

  Tumblr and Pinterest banning the content didn’t solve anything, of course, least of all their users’ body-image issues, so the sites are now taking another approach. Because these blogs are tagged, they are able to intervene algorithmically—search for thighgap on Tumblr and the screen goes blank, an overlay appearing:

  “if you or someone you know is dealing with an eating disorder …”

  A link to help and resources follows. It is a small measure, but before the behavior was digitized, there was practically no way to get directly at this problem, at least not until visible damage had already occurred. There was only rumor—an ear at the bathroom door, perhaps a parent’s sad suspicion. Data is about how we’re really feeling—feeling about one another, yes, but also about ourselves. If it finds divides in our culture, our politics, our habits, our tribes, it finds divides within us, too. And that’s a hopeful thought, because for anything to be made whole, the first step is to know what’s missing.

  1 I foreground trend lines here because the data is slightly sparser and therefore more noisy than usual. This sample is ≈5,000 people.

  2 The study of beauty by traditional methods is especially susceptible to the problem of insufficiency. If your research topic is, say, wealth, you can very easily get a measure of someone’s net worth or income and then move on to the dependent trait you want to look at. But to study beauty, first you have to determine how good-looking your subjects are, which is a resource-intensive process. Beauty being so wildly subjective (as opposed to, say, hair color, where if you crowdsourced it, you might get slight variations—brown, brunette, chestnut—that are essentially synonymous), you get wide swings in opinion that can only be absorbed by sampling a large, diverse research set. As we’ve seen with WEIRDness earlier, that has not been a strength of past academic research.

  8.

  It’s What’s Inside That Counts

  There used to be two ways to figure out what a person really thinks. One, you caught her in an unguarded moment. You snooped around, you provoked, you constructed some pretext in a laboratory, you did whatever you could do to get your subject to forget she was being watched. Research like that was probably a lot of fun—a lab coat, a hidden camera … who knows, a fake mustache—but on a large scale, it was impossible. So for data en masse, you had only option two: to ask a question and hope for an honest answer. That’s been the popular standard since Gallup formed the American Institute of Public Opinion in 1935.

  Unfortunately, surveys have historically been unable to uncover true attitudes on topics such as race, sexual behavior, drug use, and even bodily functions, because respondents edit their answers. Observed behavioral data is very useful, as we’ve already seen. But there are some things—thoughts, beliefs—that don’t entail an explicit action. And often the ugliest, most divisive, attitudes remain behind a veil of ego and cultural norms that is almost impossible to draw back, at least through direct questioning. It’s a social scientist’s curse—what you most want to get at is exactly what your subjects are most eager to hide. This tendency is called social desirability bias, and it’s well documented: the world over, respondents answer questions in ways that make them look good. The most famous case was the so-called Bradley effect: in 1982, California voters told exit pollsters they had elected a black governor, Tom Bradley, by a significant margin, but in the privacy of the ballot box they had actually given his white opponent a narrow victory. Throughout the ’80s and ’90s, black candidates often received more support in polls than in actual elections. Problems beyond racism, like depression and addiction, are similarly difficult to diagnose at a societal level because people can’t be honest about them. Even on OkCupid’s match questions—which are by and large unseen by anyone but the answerer—the users are just unwilling to own up to certain attitudes, even ones they in fact act upon elsewhere on the site. The mere act of asking elicits self-censoring. Almost every site that registers opinions or collects descriptive data has the same problem. But there is one place that doesn’t need to ask for anything, and so the data is set free: With search, there is no ask. You just tell.

  Google’s only prompt is that famously open page, with its lone entry form—that slim rectangle of emptiness, cursor parked and ready, just waiting for your thoughts. The company’s business is to help people find stuff in the vast thicket of the Internet, and it’s done that spectacularly. But almost as an afterthought to its world-beating success, as users enter each new desire into the database, Google has become a repository for humanity’s collective id. It hears our confessions, our concerns, our secrets. It’s doctor, priest, psychiatrist, confidante, and above all, Google doesn’t have to ask us for a thing, because the question is always implied in the blank space of the interface: Hey, what’s on your mind? Ahab and his whale, Arthur and his grail. What a person searches for often gives you the person himself. The trick till now has been, How can we see the search?

  Since 2008, Google has provided that insight with its Google Trends tool. It allows anyone to query their aggregated search database, and with the right phrasing and a little cross-tabulation, you can use it to extract an excellent sample of the private mind, of the internal workings that have until now remained off-limits to research since research began. Since the service launched, scientists have used Google Trends to predict the stock market, uncover drivers of economic productivity (richer countries are more concerned with the future than the past), and most famously, track epidemics of flu and dengue fever in real time—and thereby stanch them as quickly. When people are getting sick, they search for symptoms and remedies. Google Flu senses what’s afoot and alerts the CDC.

  The site also records other kinds of virulence. Because there is no asking, and unlike on social sites, no other person on the other end of the line, people unleash their vilest impulses into Google. “Nigger,” for example, is a common search term—included in 7 million searches a year. In the United States, the search volume is highest where you might expect—West Virginia—but it’s steady throughout the country. Brooklyn has few things in common with the town I grew up in, Little Rock, but this is one—“nigger” is as common in New York City as it is in central Arkansas, and as common in Chicago as it is in Fresno.1 Judging by search volume, the word is literally more American than “apple pie”—by 30 percent. And, tellingly, it appears much more often in Google than it does in a more public venue for the psyche, Twitter. Using “nigga” as a control, since it’s similar in meaning but lacks the baggage, “nigger” appears about 30 times more often in search than in social media.

  Unlike the acute cycles of disease, racism runs a slow, grinding course—working at the generational, not the metabolic level—and it’s one of the few places where we can begin to see data’s broad longitudinal possibilities. Further, tying the ebb and flow in searches to real-world events allows us to unlock some of the emotional shading behind the data. For example, if you plot searches for the word “nigger” over the 2008 campaign cycle, you can watch the country come to grips with the prospect of a black president.

  Working through the six red peaks, from left to right you see: Super Tuesday on February 5, followed by the bitterly
contested Pennsylvania primary on April 22. On June 6, searches hit a new high. Hillary suspended her campaign, and Obama won the nomination. On July 15, complicating the data (and indeed the moral discussion), Nas released an album whose unofficial title was Nigger, and it went to number one. But even in the wake of that confounding event, overall search volume plummeted as the fact of Obama’s ascendancy settled in. Racial and even political tension dissipated while the nominees, neither yet official, positioned themselves for the fall. In fact, the volume of racially charged searches reached its lowest point over the whole campaign the week of the Republican National Convention in early September.2

  Having hit a minimum there, however, animus built back quickly to the norm, then exploded on election night itself, when searches for “nigger” hit a level never since equaled. The next day, when America woke up to the confirmed reality of a black president, roughly 1 in 100 searches for “Obama” also included the epithet or “KKK” in the query string. But almost immediately afterward the volume of racially charged searches dropped sharply, and except for one last gasp of anger at the inauguration, that lower level (25 percent below the pre-Obama status quo) has held. You hear a lot about our “national conversation” on race; when you look at the data, you see it’s really more a series of national convulsions. But you also see that for all the failed promise of his famous byword, Obama did change the course of our nation’s favorite epithet: