The Washington PostDemocracy Dies in Darkness

Piketty’s ‘errors’ aren’t mistakes: They’re questions, and he answered them

French economist Thomas Piketty poses with his book "Capital in the Twenty-First Century." (Charles Platiau/Reuters)
Placeholder while article actions load

Thomas Piketty thinks the Financial Times knows nothing of his technical work.

That's the Cliff Notes version of his 4,400-word response to the criticisms the Financial Times leveled against the data in his best-selling book "Capital in the Twenty-First Century." Piketty says that even though he expects people to improve on his work in the future, he still stands by it today, and the mistakes the Financial Times thinks it's found aren't actually mistakes — and that they'd know this if they'd read the appendices he put online.

Now, to recap, these alleged — emphasis here — errors fall into three categories: 1) transcription mistakes, 2) unexplained data tweaks and 3) in the case of Britain, incorrect data. But the problem with these problems, as I pointed out, is that they aren't ones. They're questions. How did Piketty choose which source to use when they told different stories? How did he adjust them? And how much did his big-picture results depend on these decisions? All good questions — but still just questions.

Well, now that Piketty has answered the FT's criticisms, let's go through them one by one.

1) Skinny finger? The smallest and most straightforward mistake that Chris Giles of the Financial Times says he found is a transcription error. Specifically, it appears that Piketty fell victim to a "fat finger" by accidentally entering Swedish wealth data for 1908 instead of for 1920 like he should have.

But Piketty explains that he was making an adjustment, not a data-entry error. Sweden has estate tax data for before 1908 and wealth tax data for after, which give different estimates for the same amount of inequality. But since these series overlap in 1908, we can see how much they differ, and use that to try to harmonize them. It's rough, but it's reasonable — and it's what Piketty did.

2) Mysterious Excel formulas? Economics is the least exact science, and compiling wealth data might be the least exact part of it. There aren't many sources, and wealth is much harder to measure than income. Not only that, but the few wealth data sources that do exist have their own biases that need to be accounted for. You have to make judgment calls about which ones to use and how to splice them together, when appropriate.

Or how to construct them. See, sometimes there isn't enough data to give a detailed description of wealth inequality, but there is enough to give a rough sketch — which, because we only care about the long-term trend, can still be useful.

Piketty, for example, has numbers for the top 1 percent's wealth in late 1800s Britain, but not for the next 9 percent. But since we know that inequality tends to follow a Pareto distribution — that is, it's heavily skewed towards the top — we can make some educated guesses for the rest. Now, Piketty's book should better indicate where he's interpolating data like this, and better explain his rationales. But as he points out now, most of the Financial Times's problems with his work come down to these kind of technical adjustments.

3) Inequality in the U.K.? The most significant, and, it seemed, on-point criticism is about Piketty's British wealth data the past few decades. But Giles is wrong here for the same reason he is right to raise concerns about the U.S. numbers. Here's why.

It didn't get as much attention, but Giles points out that Piketty's U.S. sources are less reliable than the rest. For the U.S., there isn't as much of the estate tax data that Piketty likes to use, so he has to combine what there is with the results from the Survey of Consumer Finances (SCF) survey.

Now, as Paul Krugman explains, survey data tend to underestimate top-end inequality, because it's hard to get the rich to respond to a voluntary questionnaire about just how rich they are. The SCF, economist Gabriel Zucman told me, is better than most, because it does a better job over-sampling the top to try to correct for this bias, and it has a higher response rate. But it still needs to be adjusted if you're going to combine it with tax data.

Piketty does this, but he admits it has its limitations. He prefers to highlights more recent work by Saez and Zucman that doesn't mix and match tax and survey data, but rather uses capital income flows to construct a consistent picture of U.S. wealth inequality. This is a much more definitive account, and it shows U.S. wealth inequality rising even more than Piketty did.

The problem is Giles does with Britain what he criticizes Piketty for doing with the U.S. Specifically, Giles mixes and matches old tax data with new survey data — but unlike Piketty, he doesn't try to adjust for the survey's downward bias. This all but guarantees that it will look like inequality has been falling recently no matter the reality.

Piketty, in contrast, only uses tax data to generate his British numbers. Now, that data isn't always the best-suited, but it's still better than raw survey results. And it's why Piketty finds that the top 10 percent hold 71 percent of the wealth in Britain, while the survey that Giles touts says it's just 44 percent.

Ask yourself, as Piketty does, whether you think modern-day Britain is "one of the most egalitarian countries in history in terms of wealth distribution." That's what you'd have to believe if you take those survey results at face value.


One final point. I'm all in favor of journalists challenging economists's claims. And I think Giles has done admirable work digging into Piketty's Excel sheets when nobody else did. But throughout this, Giles has assumed the worst without waiting for a full response. For example, he points out that, for the U.S., Piketty takes the top 1 percent's wealth share and adds two to it in 1970. But without knowing anything about why Piketty makes this adjustment, Giles tells us that the numbers "didn't seem to fit what [Piketty] wanted to show, so he just added two to it." That's over the top innuendo.

In the end, he's done what he accused Piketty of: making claims that the data don't support.