Big dataiseverywhere. You probably hear people talking about data on the subway, or the radio, or by the water cooler. You turn on the TV, and the news is all about data. We’re using data to find missing airplanes, become better managers, fight disease, fight crime, fight hunger, fight fires, fight scammers, and fight each other over data. Before you know it, data is going to be all grown up and taking your son or daughter to the prom.
With all of these potential uses for data comes demand for people who can do something with it. Harvard Business Reviewdescribed the emerging job of “data scientist" as “a high-ranking professional with the training and curiosity to make discoveries in the world of big data” and the “sexiest job of the 21st Century.” The distinction between data scientists and data analysts is mostly dependent on the industry: “Scientist” tends to be used for jobs that barely existed a decade ago in tech, start-ups, and social media circles, while analyst is a job that has existed for decades within government, economic analysis, or academia.
Not one of them suggested enrolling at the local university. In fact, the most expensive thing any of them recommended was to buy a book.
Data scientist salaries typically start in the low six figures. Normally, such lucrative careers require years of expensive formal training or some serious connections; past generations of data analysts required access to university-level supercomputers to crunch numbers on the big data scale. But becoming a data scientist is perhaps the most prominent example of a new industry that breaks from the higher education model and allows people to learn the necessary skills without years of classes.
Brian Burke, founder of Advanced NFL Stats, is a former Navy fighter pilot and military contractor turned NFL data scientist. He received four weeks of statistics and econometrics training while receiving a master's in “leadership” from the Navy, but other than that he’s entirely self-taught. Charles Pensig, a senior data analyst at Jawbone, tells me he “doesn’t have much in the way of formal training.” He studied statistics at the University of Pennsylvania’s Wharton School and then “taught myself most of the skills I need.” Carl Bialik, lead news writer at FiveThirtyEight, Nate Silver’s recently-launched “data journalism” site, says his training was “mostly self-taught through Excel.”*
Of course, not every data scientist is an amateur gone pro. Sean Taylor, a research scientist on Facebook’s Data Science Team, says “working with data is all I know.” Likewise, Trey Causey, senior data scientist at Zulily and consultant for an unnamed NFL team (he signed a non-disclosure agreement), has incorporated his statistical analysis interests throughout his formal education, culminating in a minor in quantitative methods in his Ph.D. program.
Despite their different approaches, all the scientists I spoke to had virtually identical advice as to how someone could get started down this lucrative path. Not one of them suggested enrolling at the local university. In fact, the most expensive thing any of them recommended was to buy a book.
FIRST, THEY ALL SAY, stop relying on basic spreadsheets like Excel and learn a programming language. Excel’s formulas offer a variety of tools, but they’re often indirect ways of interacting with the data that can be easily misinterpreted or provide minimal insight. Think of it a bit like the “give a man a fish, he eats for a day” saying: Excel gives you data analysis one point at a time, but programming languages—where you write your own commands—teach you to interact with the data on a more meaningful level, understanding more than just the single formula.
“Break out of a point-and-click habit and get to know a language like R or Python,” Causey suggests. “It can be a slower start than something like Excel or Weka, but it forces you to think about the analyses you want to run and what you expect the output to look like.” It’s always tempting to go back to Excel when you need to do a hit-and-run analysis, but Excel’s ceiling will keep you from running serious analysis and you won’t truly understand the data. “I lean toward recommending R to people because it's had a ton of time to mature and it’s easy to install and get started,” Taylor says. "This is important because you can't really manipulate and understand data without a little bit of light programming. It also paves the way to more sophisticated analyses and fancier plotting.”
Don’t worry: You don’t have to learn these programming languages on your own if you don’t want to. “There is a treasure trove of information available on the Web, most of which is far more gentle, user-friendly, and effective than a grad school course,” Burke says. “Free courses on Coursera or similar sites can be really great sources.” Likewise, Pensig says that “the best classes I took were Coursera's data analysis in R and Codeacademy's python.” For any specific problem, Google is your friend: “You can find answers to just about anything through a well-crafted Web search,” Bialik says.
Once you have the basics down, it’s time to get your hands dirty. “I'd recommend starting some kind of project for fun,” Burke says. The consensus hovered around politics, sports, and movies as vibrant areas for amateur analysis. “Start crunching some numbers from something you're interested in. It forces you to truly understand the concepts and tools.”
Taylor agrees: “You should be curious to learn about the subject and you should have some idea of what the answers should look like in advance” so you can check your work more easily. “Torture [the data] to your heart's content,” Causey says, but don’t get carried away seeking that counter-intuitive finding. On the same point, he provided a sports example: If your model says Mark Sanchez is a better quarterback than Tom Brady, there’s probably something wrong with the model.
Finally, you’ll want to learn how to make your data look pretty. Many websites are increasing their data visualization budgets, including the recently-launched Vox venture with Ezra Klein, in which he calls beautiful data visualizations his version of clickbait. Visualizing data is sometimes referred to as Exploratory Data Analysis, or EDA. “If you can't look at the data, you won't be able to understand the story it's telling you,” Taylor cautioned. To this end, he recommended a free EDA course on Udacity produced by his Facebook colleagues.
This career path also indicates a possible, albeit much more limited, future for online education, a concept that was once hailed as the great democratizer but is now experiencing a bit of a backlash. The recommendations above are very limited in skill set and scope; that is, it’s no liberal arts education. But it also offers a kind of digital apprenticeship that requires no investment (other than time) and an earnings potential far above the mean. One of this generation’s great questions is whether higher education is still worth it. The data scientist might suggest a future where it isn’t.
This isn’t to say everyone should take these steps to become a data scientist, but that anyone with a computer, the Internet, and a basic understanding of math and statistics presumably could. And even if you’re lacking in the latter categories, there are online courses to help with that, too.
*UPDATE — April 21, 2014: We originally wrote that Charles Pensig is a senior data scientist at Jawbone. He is a senior data analyst.