Indeed, Ashenfelter, despite not knowing exactly why his regression worked exactly as it did, used it to purchase wines. According to him, “It worked out great.” The quality of the wines he drank noticeably improved.
If your goal is to predict the future—what wine will taste good, what products will sell, which horses will run fast—you do not need to worry too much about why your model works exactly as it does. Just get the numbers right. That is the second lesson of Jeff Seder’s horse story.
The final lesson to be learned from Seder’s successful attempt to predict a potential Triple Crown winner is that you have to be open and flexible in determining what counts as data. It is not as if the old-time horse agents were oblivious to data before Seder came along. They scrutinized race times and pedigree charts. Seder’s genius was to 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.
WORDS AS DATA
One day in 2004, two young economists with an expertise in media, then Ph.D. students at Harvard, were reading about a recent court decision in Massachusetts legalizing gay marriage.
The economists, Matt Gentzkow and Jesse Shapiro, noticed something interesting: two newspapers employed strikingly different language to report the same story. The Washington Times, which has a reputation for being conservative, headlined the story: “Homosexuals ‘Marry’ in Massachusetts.” The Washington Post, which has a reputation for being liberal, reported that there had been a victory for “same-sex couples.”
It’s no surprise that different news organizations can tilt in different directions, that newspapers can cover the same story with a different focus. For years, in fact, Gentzkow and Shapiro had been pondering if they might use their economics training to help understand media bias. Why do some news organizations seem to take a more liberal view and others a more conservative one?
But Gentzkow and Shapiro didn’t really have any ideas on how they might tackle this question; they couldn’t figure out how they could systematically and objectively measure media subjectivity.
What Gentzkow and Shapiro found interesting, then, about the gay marriage story was not that news organizations differed in their coverage; it was how the newspapers’ coverage differed—it came down to a distinct shift in word choice. In 2004, “homosexuals,” as used by the Washington Times, was an old-fashioned and disparaging way to describe gay people, whereas “same-sex couples,” as used by the Washington Post, emphasized that gay relationships were just another form of romance.
The scholars wondered whether language might be the key to understanding bias. Did liberals and conservatives consistently use different phrases? Could the words that newspapers use in stories be turned into data? What might this reveal about the American press? Could we figure out whether the press was liberal or conservative? And could we figure out why? In 2004, these weren’t idle questions. The billions of words in American newspapers were no longer trapped on newsprint or microfilm. Certain websites now recorded every word included in every story for nearly every newspaper in the United States. Gentzkow and Shapiro could scrape these sites and quickly test the extent to which language could measure newspaper bias. And, by doing this, they could sharpen our understanding of how the news media works.
But, before describing what they found, let’s leave for a moment the story of Gentzkow and Shapiro and their attempt to quantify the language in newspapers, and discuss how scholars, across a wide range of fields, have utilized this new type of data—words—to better understand human nature.
Language has, of course, always been a topic of interest to social scientists. However, studying language generally required the close reading of texts, and turning huge swaths of text into data wasn’t feasible. Now, with computers and digitization, tabulating words across massive sets of documents is easy. Language has thus become subject to Big Data analysis. The links that Google utilized were composed of words. So are the Google searches that I study. Words feature frequently in this book. But language is so important to the Big Data revolution, it deserves its own section. In fact, it is being used so much now that there is an entire field devoted to it: “text as data.”
A major development in this field is Google Ngrams. A few years ago, two young biologists, Erez Aiden and Jean-Baptiste Michel, had their research assistants counting words one by one in old, dusty texts to try to find new insights on how certain usages of words spread. One day, Aiden and Michel heard about a new project by Google to digitize a large portion of the world’s books. Almost immediately, the biologists grasped that this would be a much easier way to understand the history of language.
“We realized our methods were so hopelessly obsolete,” Aiden told Discover magazine. “It was clear that you couldn’t compete with this juggernaut of digitization.” So they decided to collaborate with the search company. With the help of Google engineers, they created a service that searches through the millions of digitized books for a particular word or phrase. It then will tell researchers how frequently that word or phrase appeared in every year, from 1800 to 2010.
So what can we learn from the frequency with which words or phrases appear in books in different years? For one thing, we learn about the slow growth in popularity of sausage and the relatively recent and rapid growth in popularity of pizza.
But there are lessons far more profound than that. For instance, Google Ngrams can teach us how national identity formed. One fascinating example is presented in Aiden and Michel’s book, Uncharted.
First, a quick question. Do you think the United States is currently a united or a divided country? If you are like most people, you would say the United States is divided these days due to the high level of political polarization. You might even say the country is about as divided as it has ever been. America, after all, is now color-coded: red states are Republican; blue states are Democratic. But, in Uncharted, Aiden and Michel note one fascinating data point that reveals just how much more divided the United States once was. The data point is the language people use to talk about the country.
Note the words I used in the previous paragraph when I discussed how divided the country is. I wrote, “The United States is divided.” I referred to the United States as a singular noun. This is natural; it is proper grammar and standard usage. I am sure you didn’t even notice.
However, Americans didn’t always speak this way. In the early days of the country, Americans referred to the United States using the plural form. For example, John Adams, in his 1799 State of the Union address, referred to “the United States in their treaties with his Britanic Majesty.” If my book were written in 1800, I would have said, “The United States are divided.” This little usage difference has long been a fascination for historians, since it suggests there was a point when America stopped thinking of itself as a collection of states and started thinking of itself as one nation.
So when did this happen? Historians, Uncharted informs us, have never been sure, as there has been no systematic way to test it. But many have long suspected the cause was the Civil War. In fact, James McPherson, former president of the American Historical Association and a Pulitzer Prize winner, noted bluntly: “The war marked a transition of the United States to a singular noun.”
But it turns out McPherson was wrong. Google Ngrams gave Aiden and Michel a systematic way to check this. They could see how frequently American books used the phrase “The United States are . . .” versus “The United States is . . .” for every year in the country’s history. The transformation was more gradual and didn’t accelerate until well after the Civil War ended.
Fifteen years after the Civil War, there were still more uses of “The United States are . . .” than “The United States is . . . ,” showing the country was still divided linguistically. Military victories happen quicker than changes in mindsets.
So much for how a country unites. How do a man and woman unite? Words can help here, too.<
br />
For example, we can predict whether a man and woman will go on a second date based on how they speak on the first date.
This was shown by an interdisciplinary team of Stanford and Northwestern scientists: Daniel McFarland, Dan Jurafsky, and Craig Rawlings. They studied hundreds of heterosexual speed daters and tried to determine what predicts whether they will feel a connection and want a second date.
They first used traditional data. They asked daters for their height, weight, and hobbies and tested how these factors correlated with someone reporting a spark of romantic interest. Women, on average, prefer men who are taller and share their hobbies; men, on average, prefer women who are skinnier and share their hobbies. Nothing new there.
But the scientists also collected a new type of data. They instructed the daters to take tape recorders with them. The recordings of the dates were then digitized. The scientists were thus able to code the words used, the presence of laughter, and the tone of voice. They could test both how men and women signaled they were interested and how partners earned that interest.
So what did the linguistic data tell us? First, how a man or woman conveys that he or she is interested. One of the ways a man signals that he is attracted is obvious: he laughs at a woman’s jokes. Another is less obvious: when speaking, he limits the range of his pitch. There is research that suggests a monotone voice is often seen by women as masculine, which implies that men, perhaps subconsciously, exaggerate their masculinity when they like a woman.
The scientists found that a woman signals her interest by varying her pitch, speaking more softly, and taking shorter turns talking. There are also major clues about a woman’s interest based on the particular words she uses. A woman is unlikely to be interested when she uses hedge words and phrases such as “probably” or “I guess.”
Fellas, if a woman is hedging her statements on any topic—if she “sorta” likes her drink or “kinda” feels chilly or “probably” will have another hors d’oeuvre—you can bet that she is “sorta” “kinda” “probably” not into you.
A woman is likely to be interested when she talks about herself. It turns out that, for a man looking to connect, the most beautiful word you can hear from a woman’s mouth may be “I”: it’s a sign she is feeling comfortable. A woman also is likely to be interested if she uses self-marking phrases such as “Ya know?” and “I mean.” Why? The scientists noted that these phrases invite the listener’s attention. They are friendly and warm and suggest a person is looking to connect, ya know what I mean?
Now, how can men and women communicate in order to get a date interested in them? The data tells us that there are plenty of ways a man can talk to raise the chances a woman likes him. Women like men who follow their lead. Perhaps not surprisingly, 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. If a man says, “That’s awesome!” or “That’s really cool,” a woman is significantly more likely to report a connection. Likewise if he uses phrases such as “That’s tough” or “You must be sad.”
For women, there is some bad news here, as the data seems to confirm a distasteful truth about men. Conversation plays only a small role in how they respond to women. 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. And as previously noted, a woman is also more likely to report a connection after a date where she talks about herself. Thus it is a great sign, on a first date, if there is substantial discussion about the woman. The woman signals her comfort and probably appreciates that the man is not hogging the conversation. And the man likes that the woman is opening up. A second date is likely.
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 seems counterintuitive; you might think that questions are a sign of interest. But not so on a first date. On a first date, most questions are signs of boredom. “What are your hobbies?” “How many brothers and sisters do you have?” These are the kinds of things people say when the conversation stalls. A great first date may include a single question at the end: “Will you go out with me again?” If this is the only question on the date, the answer is likely to be “Yes.”
And men and women don’t just talk differently when they’re trying to woo each other. They talk differently in general.
A team of psychologists analyzed the words used in hundreds of thousands of Facebook posts. They measured how frequently every word is used by men and women. They could then declare which are the most masculine and most feminine words in the English language.
Many of these word preferences, alas, were obvious. For example, women talk about “shopping” and “my hair” much more frequently than men do. Men talk about “football” and “Xbox” much more frequently than women do. You probably didn’t need a team of psychologists analyzing Big Data to tell you that.
Some of the findings, however, were more interesting. Women use the word “tomorrow” far more often than men do, perhaps because men aren’t so great at thinking ahead. Adding the letter “o” to the word “so” is one of the most feminine linguistic traits. Among the words most disproportionately used by women are “soo,” “sooo,” “soooo,” “sooooo,” and “soooooo.”
Maybe it was my childhood exposure to women who weren’t afraid to throw the occasional f-bomb. But I always thought cursing was an equal-opportunity trait. Not so. Among the words used much more frequently by men than women are “fuck,” “shit,” “fucks,” “bullshit,” “fucking,” and “fuckers.”
Here are word clouds showing words used mostly by men and those used mostly by women. The larger a word appears, the more that word’s use tilts toward that gender.
Males
Females
What I like about this study is the new data informs us of patterns that have long existed but we hadn’t necessarily been aware of. Men and women have always spoken in different ways. But, for tens of thousands of years, this data disappeared as soon as the sound waves faded in space. Now this data is preserved on computers and can be analyzed by computers.
Or perhaps what I should have said, given my gender: “The words used to fucking disappear. Now we can take a break from watching football and playing Xbox and learn this shit. That is, if anyone gives a fuck.”
It isn’t just men and women who speak differently. People use different words as they age. This might even give us some clues as to how the aging process plays out. Here, from the same study, are the words most disproportionately used by people of different ages on Facebook. I call this graphic “Drink. Work. Pray.” In people’s teens, they’re drinking. In their twenties, they are working. In their thirties and onward, they are praying.
DRINK.WORK.PRAY
19- to 22-year-olds
23- to 29-year-olds
30- to 65-year-olds
A powerful new tool for analyzing text is something called sentiment analysis. Scientists can now estimate how happy or sad a particular passage of text is.
How? Teams of scientists have asked large numbers of people to code tens of thousands of words in the English language as positive or negative. The most positive words, according to this methodology, include “happy,” “love,” and “awesome.” The most negative words include “sad,” “death,” and “depression.” They thus have built an index of the mood of a huge set of words.
Using this index, they can measure the average mood of words in a passage of text. If someone writes “I am happy and in love and feeling awesome,” sentiment analysis would code that as extremely happ
y text. If someone writes “I am sad thinking about all the world’s death and depression,” sentiment analysis would code that as extremely sad text. Other pieces of text would be somewhere in between.
So what can you learn when you code the mood of text? Facebook data scientists have shown one exciting possibility. They can estimate a country’s Gross National Happiness every day. If people’s status messages tend to be positive, the country is assumed happy for the day. If they tend to be negative, the country is assumed sad for the day.
Among the Facebook data scientists’ findings: Christmas is one of the happiest days of the year. Now, I was skeptical of this analysis—and am a bit skeptical of this whole project. Generally, I think many people are secretly sad on Christmas because they are lonely or fighting with their family. More generally, I tend not to trust Facebook status updates, for reasons that I will discuss in the next chapter—namely, our propensity to lie about our lives on social media.
If you are alone and miserable on Christmas, do you really want to bother all of your friends by posting about how unhappy you are? I suspect there are many people spending a joyless Christmas who still post on Facebook about how grateful they are for their “wonderful, awesome, amazing, happy life.” They then get coded as substantially raising America’s Gross National Happiness. If we are going to really code Gross National Happiness, we should use more sources than just Facebook status updates.
That said, the finding that Christmas is, on balance, a joyous occasion does seem legitimately to be true. Google searches for depression and Gallup surveys also tell us that Christmas is among the happiest days of the year. And, contrary to an urban myth, suicides drop around the holidays. Even if there are some sad and lonely people on Christmas, there are many more merry ones.
Everybody Lies Page 7