An Interview with Andrew Flowers, Quantitative Editor at FiveThirtyEight

Andrew Flowers serves as the Quantitative Editor at FiveThirtyEight, a data journalism website owned by ESPN. Before joining FiveThirtyEight in 2014, Flowers served as an economic research analyst at the Federal Reserve Bank of Atlanta. Flowers applies his background in economics, data analysis, and computer programming to his writing, approaching political, sports, and economic topics from a numbersstandpoint.

The Politic: What are the roles of a quantitative editor? Describe the process of quantitative editing.

Andrew Flowers: FiveThirtyEight is a data journalism site, and like any journalism site, we need editors. We have several senior editors who handle features, news, and everything in between. They edit the prose, the reporting, and the argument of the piece. I’m the quantitative editor. So I edit the methodology and the data analysis in a piece, or whether the approach a writer takes in analyzing the data is sound. I check that the specific statistical tests they are running, such as regressions, are done in the right way; they must be robust to different specifications, meaning the statistical model does not change dramatically if you add or tweak a different type of variable in the analysis. Basically, my role as a quantitative editor is to “stress test”the data analysis within the pieces our writers submit.

The Politic: What previous work or education experience prepared you for this position?

Andrew Flowers: My education experience includes an economics degree from the University of Chicago. This really prepared me because the core economics curriculum at Chicago was very mathematical and rigorous in its critical thinking aspects, not simply memorizing facts and theories in economics. It was more focused on applying them. But more so than the core economics curriculum at Chicago, I took some empirical economics classes where I got my hands dirty with data sets and played with data and different kinds of statistical programming languages. More importantly, recently I used to work at the Federal Reserve Bank in Atlanta, where I was an economic analyst from 2008 until I took this job at FiveThirtyEight in early 2014. My role at the Atlanta Fed was very interesting because it changed a lot. Right when I got there, the financial crisis and recession ignited, and while it was unfortunate for the US economy, it was exciting to be there as an economics student and directly applying what I had just learned. My day-to-day roles at the Atlanta Fed included analyzing incoming data, whether it was GDP, inflation, the jobs reports, financial market data, or data from Europe, and distill it for other senior economists and policy-makers. I would break it down and say: “I’ve analyzed it in all these different ways, here’s my interpretation.”I thought that prepared me for my job at FiveThirtyEight, since it essentially reflects the mission of the site in general. We write about news stories and fuse data with journalism in a smart, legitimate way.

The Politic: You graduated with a degree in economics and previously served as a Senior Economic Research Analyst for the Federal Reserve Bank of Atlanta. Did you always see yourself working in journalism?

Andrew Flowers: I always saw myself as a writer. However, I did not really see myself going into journalism. The route I was thinking of was to be an academic. I definitely wanted to write for a popular audience, I just didn’t know it would be from the nitty-gritty world of journalism rather than the “Ivory Tower high-horse”position, so to speak. I think FiveThirtyEight has really bridged the gap between those two. We like to say data journalism is social science on deadline. So I think I’ve kind of melded these two worlds. I still have this kind of academic flavor in my work life where we are rigorous but we also don’t take ourselves too seriously, inject voice in our writing, and report, which is something academics rarely do. All the traditional characteristics of journalism I’ve been learning and adopting because they’re crucial.

The Politic: FiveThirtyEight is known for its use of data analysis and mathematical models. What do these elements add to a piece of writing?  Why does FiveThirtyEight support this style of journalism?

Andrew Flowers: I’ll answer those in reverse. FiveThirtyEight supports using data in journalism because data is one aspect that is often underrepresented in traditional journalism. But we’re not just data. We’re more than that. We value reporting and non-empirical facts tremendously. But we add to it. The precedent that Nate Silver, our editor-in-chief, set with his blog and his tenure at the New York Times showed that journalism needed this. Why did it need this? Journalists and pundits were making arguments, interjecting theories about the world that simply were not backed up by simple empirical analysis. FiveThirtyEight wants to fill this void and fulfill that mission. We can do all the elements of traditional journalism, and we value them, such as reporting, but we’re going to add to this with an emphasis on statistics, programming, and data visualization. These are skill sets that were underrepresented in journalism before. When you add that all together, you get a journalism product that is still faithful to reporting and traditional journalism, but also brings the empirical rigor that was not there before.

The Politic: Is FiveThirtyEight changing the way issues are reported? Are more news sources moving towards the use of “big data”?

Andrew Flowers: Absolutely. Many major news sources that can afford to invest in it are being deliberate in hiring and cultivating journalists on staff who have skills in statistics, data analysis, and data visualization. In some ways this is a competitive market response, because they think readers want something more “rigorous”, because Nate had success with it and FiveThirtyEight continues to have success with it. But I want to point out, and Nate has said this before, that data journalism didn’t start in the last few years. NICAR (National Institute for Computer-Assisted Reporting) has been around for decades, and there are journalists who have been doing data journalism as we conceive of it now for a long time. It has just gotten popular recently. Also, you mentioned the words “big data”. At FiveThirtyEight, we’re actually a little skeptical of the hype around that idea. We appreciate the movement we’re living through right now, because if you interpret the words “big data”to mean “an availability of data sets in an easily accessible manner through the internet, with increasing amounts of data on just about everything and tools for data analysis which make it easier to use”, then yes, that’s a good thing. But that doesn’t mean we’re going to have an easily inferred sense of the world just because we have data. Data analysis is actually really hard, in that more data doesn’t always help you answer the questions more accurately. Sometimes a lot of data just adds noise to a problem. Don’t get me wrong, we’re all for data. But we don’t want to over-promise the public, and by “we”I mean those of us who tout “big data”.

The Politic: Is FiveThirtyEight changing the way consumers form opinions and come to conclusions? Are readers more likely to trust an article with statistics rather than one without them?

Andrew Flowers: I think there are two separate issues here, and it is a really interesting question. One is whether or not a data journalism article done really well is going to garner more respect from an intelligent reader than a traditionally reported but non-data journalism article would. I think yes, it will. Again, it has to have an equal level of reporting and other non-data journalism elements to earn and receive that respect. But yes, I think it can powerfully alter how people view the news. The second, separate issue, which is deeper and philosophical, is if a data journalism article can change someone’s views. And that is more profound, so I don’t know if I can answer that. If a reader is empirically open-minded and willing to have his or her mind changed, then yes, data can help. But from reading books like The Righteous Mind by Jonathan Haidt and other findings in cognitive psychology, people have found that more information doesn’t make people more circumspect about how they view the world. Sometimes the reader who seeks out the most information suffers from confirmation bias and will only remember the data journalism articles that back up his views, and will quickly forget those that challenge his views.

The Politic: As a writer, is it easier to persuade readers with statistics and analysis or with prose?

Andrew Flowers: Stories are the most powerful device we have to convey information. Stories tend to be the most successful when they have characters. The traditional news story or feature magazine-style piece that has real people or events or places in them is not going away and is not losing its power. But I think this can be combined with data analysis that is accessible. A story that has both of these elements, traditional elements such as characters along with data analysis that is sound and rigorous, is the most powerful way to catch and change the minds of readers.

It is very difficult to combine these elements. It sounds easy, but we try hard at it every day. Nate has set a great example for us. My other colleagues that you’ve interviewed such as Ben Casselman and Carl Bialik, along with other journalists with great data skills, have the talent to weave in traditional story elements that illuminate the problem for the reader as well as temper any conclusions they have with sound data arguments. I think this will garner the greatest amount of respect from readers.

The Politic: What “signs”should a reader look out for to identify bias in an article filled with so many statistics?

Andrew Flowers: Personally, I worry about tone more than I worry about bias. I think having a strong view on something, when you have enough subject matter expertise as a reporter to adjudicate a problem, is not always bad. Some people might call that bias but I think there are some scenarios where this is legitimate. What I worry about more than bias, or at least what is easier to detect from a reader’s point of view, is tone. Specifically regarding data, when a tone is withholding information and masking, then I worry. I’m not saying you need to walk the reader through every single calculation in a mundane way, but you need to be transparent. These are actual procedural failures on the part of a data journalist if you don’t do those things, such as explaining your methodology or releasing your data. When a journalist is being flippant with data by signaling to the reader: “Just take my word for it, the data says x”, and they don’t discuss how they arrived at x, that way of withholding the process and not letting the reader make up his or her own mind is more worrisome than any sort of bias about x or y.

It comes down to treating the reader with respect. These are complex issues, and we’re not going to hide the complexity. When I have a judgment I’m rendering, not as the expert in the field but as the journalist who has really covered this and backed it up with data, I’m going to present it with a tone that let’s you decide.  It should be concise rather than plodded with methodology, as well as transparent rather than hurried and withholding.

The Politic: Walk me through the process of writing an article for FiveThirtyEight. Do you make an observation and find data to support it? Or do you start with the data and draw conclusions from it?

Andrew Flowers: The story matters above all else. Only in extreme, rare circumstances do you find data that evolves into a story or argument out of the blue. Even though we’re data oriented and we want to make sure our analysis is correct, the first hook of a story pitch, when we talk to our editors, is whether or not the angle is interesting. Is it going to provide a good avenue of analysis for the reader? If it is, then we do the analysis and see where it takes us. We never “data mine”until we find an interesting quirk and then reverse engineer a story from that. In rare circumstances however, and sports is a good example, the story discusses a data record being set or a threshold being met. In this case, the data marks the story for the reader. But usually, for a complex problem such as long-term unemployment or police department racial diversity, you go look for the data. To summarize, at FiveThirtyEight the pitching process and ideas that get debated in the brainstorming phase are focused on the story and argument first. While the data is absolutely crucial, it usually does not drive the conversation.

The Politic: One of your articles discusses a debate surrounding the development of a “value-added”measure for teacher effectiveness based on standardized-test scores relative to the expectations of those test scores. Now that this measure is gaining support and credibility, what changes might teachers make? As a result, how will students be affected?

Andrew Flowers: I have found the issue of value-added models, or VAM’s, to be incredibly contentious. The story I wrote covered an academic or empirical debate that I felt was beginning to tilt in one direction because of the consistent reproducibility of these estimates in different contexts. If that debate is settled, and I’m not saying for sure it will be, it doesn’t mean that VAM’s are going to be implemented in public school systems throughout the nation, nor should they be. That’s an important point. Even advocates of these models, such as Chetty, Friedman, Rockoff and the academics who did the research that seems to show that VAM’s could work, would be circumspect, I would bet, about expanding VAM’s nationwide in a piecemeal approach. One issue is an academic empirical debate, which I think is beginning to resolve itself. The second issue is a much more complex and political issue. It is going to be resisted by certain entrenched interest groups, and it is going to be overextended and applied in crude ways by other interest groups. Even if you believed from my reporting that VAM models are sound, you should still be skeptical that VAM’s are anywhere near being implemented nationwide.

The Politic: Another article comments on recent research that found the cost of child care to be not quite as burdensome as it is made out to be. What should families who are worried about the costs of child-care take away from this article?

Andrew Flowers: What I want my readers to learn is that quick, take-away data points put out by government reports should be treated with a large grain of salt. This is not because they are fake, but because they can be overly crude. The specific paper discussed in my article was criticizing a measure of child-care costs as put out by the US Bureau of Labor Statistics in a regular report. The paper tempers the Bureau’s characterization of child-care costs as exploding. This doesn’t mean that the cost of child-care isn’t an important issue, and I hope my readers come away with that. The research does not indicate that child-care costs are falling, but that they might not be rising as much as they are made out to be. The market has bifurcated. Thirty years ago, people who did consume child-care usually consumed, roughly speaking, a common set of services. But today, they don’t. There are less expensive child-care services, or those that are provided by the government such as Head Start. On the other end, there are very high-end expensive services, such as enrichment programs and private preschools that are rising in cost dramatically.

The Politic: Are there ways to statistically determine the success of politicians or businesses compared to their peers? What are the comparable “sabermetrics”to politicians?

Andrew Flowers: For politicians, there are many domains in which you can analyze them. Obviously it is important for campaigns and elections, and I think Nate, my colleague Harry Enten, and others in the data journalism world have been at the forefront of analyzing polling data and providing insight about that element of politics. In terms of legislation and policy, however, we are much more behind in bringing data to the table. Honestly, the Congressional Budget Office, think tanks, and other policy academics don’t get enough credit for, on a regular basis, providing analysis on legislation. Often this is simply undersold and obfuscated. Either interest groups want to obfuscate them or it is just not sold well since it is not done by a journalist. Beyond the CBO and think tank work done on policy, where is the cutting edge in terms of data and politics? I really am curious by ideological scores, which FiveThirtyEight has begun to increasingly use to analyze candidates. These scores come from legislators’and governors’actual track records in office. Beyond that, if they’ve never actually held political office, it can come from the language that they use or their fundraising. There are many interesting ideological scores done with fundraising and speech that I think are some really interesting data work when it comes to politics.

The Politic: Your latest article looks at recently obtained data that determines that Uber serves New York’s Outer Boroughs more than yellow and green taxis. How do you think Uber will use this data to defend its position?

Andrew Flowers: In a narrow sense I don’t care. And I don’t mean that to dismiss your question. I would imagine they are going to use this to justify their current operating scheme. They can sell to the public the fact that they service the outer boroughs really well. How they are going to use it beyond that I’m not sure nor do I really care. There’s lot’s of data analysis still to come and we’re still looking at it in terms of other demographic effects. These include whether or not demographics plays an important role in driving Uber’s market share in NYC, or whether it is simply geography. And it may just be that the reason Uber is so popular in outer boroughs is that subway access and yellow and green taxi service are simply not as good. Uber and the taxis may be complements rather than substitutes. This is an impression I received from a first glance at the data.

 

 

Leave a Reply

Your email address will not be published. Required fields are marked *