9 August 2020

Everybody Lies – Seth Stephens-Davidowitz

On Google, the top complaint about a marriage is not having sex. Searches for “sexless marriage” are three and a half times more common than “unhappy marriage” and eight times more common than “loveless marriage.” Even unmarried couples complain somewhat frequently about not having sex. Google searches for “sexless relationship” are second only to searches for “abusive relationship.”

Our research suggests that a person is significantly more likely to put the candidate they support first in a search that includes both candidates’ names.

You don’t always need a ton of data to find important insights. You need the right data. A major reason that Google searches are so valuable is not that there are so many of them; it is that people are so honest in them. People lie to friends, lovers, doctors, surveys, and themselves. But on Google they might share embarrassing information, about, among other things, their sexless marriages, their mental health issues, their insecurities, and their animosity toward black people. Most important, to squeeze insights out of Big Data, you have to ask the right questions.

While the methodology of good data science is often intuitive, the results are frequently counterintuitive. Data science takes a natural and intuitive human process — spotting patterns and making sense of them — and injects it with steroids, potentially showing us that the world works in a completely different way from how we thought it did.

When trying to make predictions, you needn’t worry too much about why your models work.

Look for data where others hadn’t looked before, to consider nontraditional sources of data. For a data scientist, a fresh and original perspective can pay off.


A woman is unlikely to be interested when she uses hedge words and phrases such as “probably” or “I guess.”

A woman is likely to be interested when she talks about herself.

A woman is more likely to report a connection if a man laughs at her jokes and keeps the conversation on topics she introduces rather than constantly changing the subject to those he wants to talk about. Women also like men who express support and sympathy.

Physical appearance trumps all else in predicting whether a man reports a connection. That said, there is one word that a woman can use to at least slightly improve the odds a man likes her and it’s one we’ve already discussed: “I.” Men are more likely to report clicking with a woman who talks about herself.

Finally, there is one clear indicator of trouble in a date transcript: a question mark. If there are lots of questions asked on a date, it is less likely that both the man and the woman will report a connection.

This data also has implications for one’s optimal dating strategy. Clearly, one should put oneself out there, get rejected a lot, and not take rejection personally. This process will allow you, eventually, to find the mate who is most attracted to someone like you. Again, no matter what you look like, these people exist. Trust me.


downloaded the text of thousands of books and movie scripts. They could then code how happy or sad each point of the story was.

The computer scientists found that a huge percentage of stories fit into one of six relatively simple structures. They are, borrowing a chart from Reagan’s team:

  • Rags to Riches (rise)
  • Riches to Rags (fall)
  • Man in a Hole (fall, then rise)
  • Icarus (rise, then fall)
  • Cinderella (rise, then fall, then rise)
  • Oedipus (fall, then rise, then fall)

There are a lot of additional questions we might answer. For example, how has the structure of stories changed through time? Have stories gotten more complicated through the years? Do cultures differ in the types of stories they tell? What types of stories do people like most? Do different story structures appeal to men and women? What about people in different countries?

Content is more likely to become viral the more positive it is.

The owners of the American press, instead, are primarily giving the masses what they want so that the owners can become even richer.

The liberal bias is well calibrated to what newspaper readers want. Newspaper readership, on average, tilts a bit left. (They have data on that.) And newspapers, on average, tilt a bit left to give their readers the viewpoints they demand. There is no grand conspiracy. There is just capitalism.


They take pictures of the same locations over and over again. The pictures are sent back to Premise, whose second group of employees—computer scientists—turn the photos into data. The company’s analysts can code everything from the length of lines in gas stations to how many apples are available in a supermarket to the ripeness of these apples to the price listed on the apples’ bin. Based on photographs of all sorts of activity, Premise can begin to put together estimates of economic output and inflation.

In developing countries, long lines in gas stations are a leading indicator of economic trouble. So are unavailable or unripe apples. Premise’s on-the-ground pictures of China helped them discover food inflation there in 2011 and food deflation in 2012, long before the official data came in.

Truth in surveys vs. online searches

What the residents reported to the surveys was very different from the data the researchers had gathered. Even though nobody gave their names, people, in large numbers, exaggerated their voter registration status, voting behavior, and charitable giving.

“About one-third of the time, people lie in real life,” he suggests. “The habits carry over to surveys.” Then there’s that odd habit we sometimes have of lying to ourselves. “There is an unwillingness to admit to yourself that, say, you were a screw-up as a student.”

Lying to oneself may explain why so many people say they are above average. How big is this problem? More than 40 percent of one company’s engineers said they are in the top 5 percent.

Think of Google searches. Remember the conditions that make people more honest. Online? Check. Alone? Check. No person administering a survey? Check. And there’s another huge advantage that Google searches have in getting people to tell the truth: incentives. If you enjoy racist jokes, you have zero incentive to share that un-PC fact with a survey. You do, however, have an incentive to search for the best new racist jokes online.

Adults with children are 3.6 times more likely to tell Google they regret their decision than are adults without children.

How should we interpret this? Does this really imply that boyfriends withhold sex more than girlfriends? Not necessarily. As mentioned earlier, Google searches can be biased in favor of stuff people are uptight talking about. Men may feel more comfortable telling their friends about their girlfriend’s lack of sexual interest than women are telling their friends about their boyfriend’s. Still, even if the Google data does not imply that boyfriends are really twice as likely to avoid sex as girlfriends, it does suggest that boyfriends avoiding sex is more common than people let on. Google data also suggests a reason people may be avoiding sex so frequently: enormous anxiety, with much of it misplaced.

Men conduct more searches for how to make their penises bigger than how to tune a guitar, make an omelet, or change a tire.

Racism and unconscious bias

White Americans may mean well, this theory goes, but they have a subconscious bias, which influences their treatment of black Americans. Academics invented an ingenious way to test for such a bias. It is called the implicit-association test.

The tests have consistently shown that it takes most people milliseconds longer to associate black faces with positive words, such as “good,” than with negative words, such as “awful.” For white faces, the pattern is reversed. The extra time it takes is evidence of someone’s implicit prejudice—a prejudice the person may not even be aware of. There is, though, an alternative explanation for the discrimination that African-Americans feel and whites deny: hidden explicit racism.

Let me put forth the following conjecture, ready to be tested by scholars across a range of fields. The primary explanation for discrimination against African Americans today is not the fact that the people who agree to participate in lab experiments make subconscious associations between negative words and black people; it is the fact that millions of white Americans continue to do things like search for “nigger jokes.”

Boys & girls – seen by parents

Parents are two and a half times more likely to ask “Is my son gifted?” than “Is my daughter gifted?”

In American schools, girls are 9 percent more likely than boys to be in gifted programs. Despite all this, parents looking around the dinner table appear to see more gifted boys than girls.

Parents are about twice as likely to ask how to get their daughters to lose weight as they are to ask how to get their sons to do the same. Just as with giftedness, this gender bias is not grounded in reality. About 28 percent of girls are overweight, while 35 percent of boys are.

Political segregation

People, if left to their own devices, tend to seek out viewpoints that confirm what they believe. Thus, surely, the internet must be creating extreme political segregation. There is one problem with this standard view. The data tells us that it is simply not true.

Chances that two people visiting the same news site have different political views is about 45 percent. In other words, the internet is far closer to perfect desegregation than perfect segregation. Liberals and conservatives are “meeting” each other on the web all the time.

you are more likely to come across someone with opposing views online than you are offline.

The internet news industry is dominated by a few massive sites.

Many people with strong political opinions visit sites of the opposite viewpoint, if only to get angry and argue.

People, on average, have substantially more friends on Facebook than they do offline. And these weak ties facilitated by Facebook are more likely to be people with opposite political views.

DIGITAL TRUTH: Searches, Views, Clicks, Swipes
DIGITAL LIES: Social media posts, Social media likes, Dating profiles

Zuckerberg had learned an important secret: people can claim they’re furious, they can decry something as distasteful, and yet they’ll still click.

Netflix learned a similar lesson early on in its life cycle: don’t trust what people tell you; trust what they do. Originally, the company allowed users to create a queue of movies they wanted to watch in the future but didn’t have time for at the moment.

But days later, when they were reminded of the movies on the queue, they rarely clicked. What was the problem? Ask users what movies they plan to watch in a few days, and they will fill the queue with aspirational, highbrow films, such as black-and-white World War II documentaries or serious foreign films. A few days later, however, they will want to watch the same movies they usually want to watch: lowbrow comedies or romance films.

Faced with this disparity, Netflix stopped asking people to tell them what they wanted to see in the future and started building a model based on millions of clicks and views from similar customers.

When we lecture angry people, the search data implies that their fury can grow. But subtly provoking people’s curiosity, giving new information, and offering new images of the group that is stoking their rage may turn their thoughts in different, more positive directions.

Life expectancy

The first three — religion, environment, and health insurance — do not correlate with longer life spans for the poor. The variable that does matter, according to Chetty and the others who worked on this study? How many rich people live in a city. More rich people in a city means the poor there live longer. Poor people in New York City, for example, live a lot longer than poor people in Detroit.

One hypothesis — and this is speculative — was put forth by David Cutler, one of the authors of the study and one of my advisors. Contagious behavior may be driving some of this. There is a large amount of research showing that habits are contagious. So poor people living near rich people may pick up a lot of their habits. Some of these habits—say, pretentious vocabulary—aren’t likely to affect one’s health. Others—working out—will definitely have a positive impact. Indeed, poor people living near rich people exercise more, smoke less, and are less likely to suffer from obesity.

These are just correlations, but they do suggest that growing up near big ideas is better than growing up with a big backyard.

The greater the percentage of foreign-born residents in an area, the higher the proportion of children born there who go on to notable success.

Spending a lot on education helps kids reach the upper middle class. It does little to help them become a notable writer, artist, or business leader.


Showing a violent movie somehow caused a big drop in crime.

Can you guess why? Think, first, about who is likely to choose to attend a violent movie. It’s young men—particularly young, aggressive men.

Violent movies keep potentially violent people off the streets.

There was one more strange thing in the data. The effects started right when the movies started showing; however, they did not stop after the movie ended and the theater closed. On evenings where violent movies were showing, crime was lower well into the night, from midnight to 6 A.M. Even if crime was lower while the young men were in the movie theater, shouldn’t it rise after they left and were no longer preoccupied? They had just watched a violent movie, which experiments say makes people more angry and aggressive.

Alcohol is a major contributor to crime. The authors had sat in enough movie theaters to know that virtually no theaters in the United States serve liquor.

They could not, for instance, test the months-out, lasting effects — to see how long the drop in crime might last. And it’s still possible that consistent exposure to violent movies ultimately leads to more violence.


Silver searched for players’ doppelgangers. Here’s how it works. Build a database of every Major League Baseball player ever, more than 18,000 men. And include everything you know about those players: their height, age, and position; their home runs, batting average, walks, and strikeouts for each year of their careers. Now, find the twenty ballplayers who look most similar to Ortiz right up until that point in his career—those who played like he did when he was 24, 25, 26, 27, 28, 29, 30, 31, 32, and 33. In other words, find his doppelgangers. Then see how Ortiz’s doppelgangers’ careers progressed. A doppelganger search is another example of zooming in. It zooms in on the small subset of people most similar to a given person. And, as with all zooming in, it gets better the more data you have.

Amazon uses something like a doppelganger search to suggest what books you might like. They see what people similar to you select and base their recommendations on that. Pandora does the same in picking what songs you might want to listen to. And this is how Netflix figures out the movies you might like.

There are major areas of life that could be vastly improved by the kind of personalization these searches allow. Take our health, for instance.

The problem lies with data collection. Most medical reports still exist on paper buried in files, and for those that are computerized, they’re often locked up in incompatible formats.

He created a website, PatientsLikeMe.com, where individuals can report their own information—their conditions, treatments, and side effects. He’s already had a lot of success charting the varying courses diseases can take and how they compare to our common understanding of them. His goal is to recruit enough people, covering enough conditions, so that people can find their health doppelganger.


How, then, can we more accurately establish causality? The gold standard is a randomized, controlled experiment. Here’s how it works. You randomly divide people into two groups. One, the treatment group, is asked to do or take something. The other, the control group, is not. You then see how each group responds. The difference in the outcomes between the two groups is your causal effect.

Offline experiments can cost thousands or hundreds of thousands of dollars and take months or years to conduct.

In the digital world, randomized experiments can be cheap and fast. You don’t need to recruit and pay participants. Instead, you can write a line of code to randomly assign them to a group. You don’t need users to fill out surveys. Instead, you can measure mouse movements and clicks. You don’t need to hand-code and analyze the responses. You can build a program to automatically do that for you. You don’t have to contact anybody. You don’t even have to tell users they are part of an experiment.

Randomized controlled experiments have been renamed “A/B testing.”

Another reason A/B testing is so important is that seemingly small changes can have big effects.

More people in those cities saw the ad. More people in those cities decided to go to the film. One alternative explanation might be that having a team in the Super Bowl makes you more likely to go see movies. However, we tested a group of movies that had similar budgets and were released at similar times but that did not advertise in the Super Bowl. There was no increased attendance in the cities of the Super Bowl teams.

They found a 2.5-to-1 return on investment.


It can be anything from entertaining to self-torture for human beings to play out hypotheticals.

Milan Kundera, the Czech-born writer, has a pithy quote about this in his novel The Unbearable Lightness of Being: “Human life occurs only once, and the reason we cannot determine which of our decisions are good and which bad is that in a given situation we can make only one decision; we are not granted a second, third or fourth life in which to compare various decisions.”

Regression discontinuity

Anytime there is a precise number that divides people into two different groups — a discontinuity — economists can compare — or regress — the outcomes of people very, very close to the cutoff.

Those right above or right below the sharp numerical threshold had virtually identical criminal histories and backgrounds. This one measly point, however, meant a very different prison experience. The result? The economists found that prisoners assigned to harsher conditions were more likely to commit additional crimes once they left. The tough prison conditions, rather than deterring them from crime, hardened them and made them more violent once they returned to the outside world.

Students on either side of the cutoff ended up with indistinguishable AP scores and indistinguishable SAT scores and attended indistinguishably prestigious universities.

Ten years into their careers, the average graduate of Harvard makes $123,000. The average graduate of Penn State makes $87,800. But this correlation does not imply causation.

When two students, from similar backgrounds, both got into Harvard but one chose Penn State, what happened? The researchers’ results were just as stunning as those on Stuyvesant High School. Those students ended up with more or less the same incomes in their careers.


Jawbone scored a huge win using a two-pronged goal. First, ask customers to commit to a not-that-ambitious goal. Send them a message like this: “It looks like you haven’t been sleeping much in the last 3 days. Why don’t you aim to get to bed by 11:30 tonight? We know you normally get up at 8 A.M.” Then the users will have an option to click on “I’m in.” Second, when 10:30 comes, Jawbone will send another message: “We decided you’d aim to sleep at 11:30. It’s 10:30 now. Why not start now?” Jawbone found this strategy led to twenty-three minutes of extra sleep. They didn’t get customers to actually get to bed at 10:30, but they did get them to bed earlier. Of course, every part of this strategy had to be optimized through lots of experimentation. Start the original goal too early—ask users to commit to going to bed by 11 P.M.—and few will play along. Ask users to go to bed by midnight and little will be gained.