Creator Of Chart Of Doom Defends Chart Badly And Casts Aspersions

If you are going to accuse someone of cherry-picking data, you want to make sure your case is pretty solid. In particular, you don’t want to criticize them for being anti-empirical when your underlying case is based on an elementary methodological error related to basic math. Like Pavlina Tcherneva has done.

Tcherneva is the creator of the much-heralded Chart of Doom showing that the share of income growth going to the top has (nearly) steadily risen since the mid-twentieth century. I argued that the chart conveyed that impression only because of several methodological shortcomings. I acknowledged that inequality has risen but showed that the rate at which it has risen hasn’t steadily grown. That is a point that none of my critics has acknowledged as my point (even one who received an email response from me ahead of his response explaining that that was my point). It is a policy relevant point, which is why I spent the better part of three days crunching numbers and writing my original piece.

Tcherneva has now weighed in with her response to me. She starts off making the same mistake that others have–she says that my argument is “that income distribution [sic] has been improving since 1990.” Actually I find that the rate at which the gains have been going disproportionately to the top has been slowing, which is to say the gains are still going disproportionately to the top and, as I stated at the beginning of my essay, the share of income going to the top is (probably) increasing. My argument was not about whether inequality is growing or not, it was about whether, as Tcherneva says in her latest post, “with virtually every postwar expansion, a greater and greater share of the average income growth has gone to the wealthy [sic] 10% of families.”

But her most egregious error comes in her discussion of what one finds looking at business cycles instead of economic expansions. I had argued that by looking only at expansions, Tcherneva obscures the fact that when one group gets a disproportionate share of gains in an expansion, it can get a disproportionate share of losses in the subsequent recession. If all outsized gains are eroded by subsequent outsized losses, then over a business cycle, it’s a wash. What we care about are business cycles. Tcherneva now doesn’t seem to disagree that it is better to look at business cycles. Indeed, she thinks doing so reinforces her original chart because the way I looked at business cycles was wrong. But that’s not all–she says I did it wrong to mislead the reader: “he’s calculating his business cycles incorrectly (and I am forced to conclude, based on what you are about to read, deliberately so because, he doesn’t like the outcomes that derive from correct business cyclical calculations…I can’t think of any other possible motive)!”

Tcherneva is wrong on both counts–I did it right and my motive was to ensure that public policy isn’t based on charts that fail to convey history correctly.

According to Tcherneva, I “double report” the years that mark the peak of business cycles. This is a technical argument, but that she is wrong and I am right is fairly easy to convey. Imagine that from 2014 to 2018 mean income for the bottom 90% is as follows:

2014: 30,000

2015: 32,000

2016: 30,000

2017: 32,000

2018: 30,000

Imagine that mean income for the top 10% looks like this:

2014: 300,000

2015: 350,000

2016: 300,000

2017: 350,000

2018: 300,000

Assume that 2015 and 2017 are business cycle peaks (they are clearly peak years for income, but let’s also assume there is no disagreement that they are business cycle peaks by any other criterion). What I did in my essay is exactly what Emmanuel Saez did in his much-cited note finding that the top 1% received 95% of the income gains from 2009 to 2012. It’s what anyone would do if they thought about it. We want to know what share of income gains between 2015 and 2017 went to the top. It’s clear from this construction that there were no gains–both top and bottom end up with the same income in 2017 as in 2015. (By the way, the formula to estimate the share of gains going to the top would blow up in this case because the relevant denominator would be 0.)

What Tcherneva says I should have done (and actually does) is to instead look at the share of gains going to the top between 2016 and 2017. That is, she would look at the share of gains going to the top through 2015, then she would look at the share of gains going to the top from 2016 to 2017. Doing what I and anyone else who thought about it would do–looking at the share of gains going to the top through 2015 then from 2015 to 2017–she calls “double reporting” 2015. So she re-does the business cycle chart without “double reporting” and gets results….well, that are still very different from her original chart but also very different from my charts. But she is estimating the wrong thing here. What is relevant in the example I have constructed is that the top’s outsized gains from 2016 to 2017 were preceded by outsized losses from 2015 to 2016, and they ended up where they started. Tcherneva would ignore 2015 to 2016 and say the top got an outsized share of gains during the “2016-17 business cycle,” which isn’t a business cycle at all because it excludes the initial income losses. It starts from a below-peak point and ends at a peak. It is “like” only including expansions in that each business cycle it drops the first bad year.

I challenge any economist reading this to defend her approach. I will offer you this space to do it.

I will also offer this space to Tcherneva to apologize for accusing me of trying to cherry-pick my results.

Tcherneva also thinks I somehow gamed the results by the years I used to define business cycles. This is also wrong-headed. All I did was use the peak years she selected for her original chart to define the start and end points for business cycles. She showed an expansion running from 1949 to 1953 and then another running from 1954 to 1957. That means 1953 was a peak (the end of an expansion, by her definition) and 1957 was a peak. My “business cycle” runs from 1953 to 1957. I did this not because I think the “peaks” she picked are the best ones or the ones I would use, but because I wanted to be as consistent with her methods as possible. Imagine if I had changed the years how much criticism I’d have gotten for gaming things!

The one exception is that I did combine the 1969-73 and 1973-79 business cycles into one, and I regret that I did not explain why I did so. The reason is that the best scholars of income trends know that the 1973 peak was a manufactured one. Richard Nixon’s price controls were timed to keep prices low before the 1972 election and subsequently lifted. What would have been gradually increasing prices ended up being sharply rising prices after 1973, so that 1973 was a peak for inflation-adjusted income. We don’t know when a real peak would have come, and there’s no good way of dealing with the issue except by simply comparing the previous peak (1969) to the next one (1979). I will add an explanatory note to my original post laying all this out.

But there is no funny business here. Despite what Tcherneva says, I’m not convinced that keeping the two business cycles separate would alter my results in any way that would help her. I will conduct the analyses myself tonight and add them to this essay for transparency. If the results alter the conclusion of my original essay, I will edit the essay and add a note.

(As a side note, Tcherneva says that “peak years” should be taken from the Piketty-Saez data, but you get different ones if you use the Census Bureau income data, as most people do. You also occasionally get different ones with different income definitions in the CBO data. Remember, the Piketty-Saez data says that the bottom 90% was poorer in 2012 than in 1979.)

If it’s not clear, let me say it explicitly: the figures Tcherneva provides in her latest post are junk.

God, what else….so she thinks the Piketty and Saez figures may understate inequality. She really wants to believe this because she tried to make the argument on Twitter last week that their data exclude people too poor to file taxes. That’s not true–they impute incomes to these people, who are a very small group. So now she’s citing a paper that imputes income from wealth to people and finds that inequality is higher in any given year with the imputations.

Of course what matters for this debate is whether the Piketty-Saez figures understate or overstate the rise in inequality. The Wolff-Zacharias paper she cites says that including the imputed income from wealth, “shows about the same change over the 1982-2000 period” as when you don’t add income from wealth. So, nothing-burger.

More: Tcherneva says that the appropriate way to deal with the downward pull on pre-tax and -transfer incomes exerted by the elderly is “not to remove them from the data but to account for all forms of income. Except Winship wants us to include the transfer and tax effect but exclude other sources of income that tend to benefit the wealthy, e.g., capital gains. You can’t have it both ways.” Well, I actually kept capital gains and “all forms of income” in all my charts that include the elderly, so Tcherneva looks pretty out of line here too. Incidentally, I did that so I would not be accused of cherry-picking. So much for that–here’s Tcherneva again, “Why remove the elderly from the picture or only account for transfers, while excluding capital gains? By definition that would make the income distribution picture rosier, wouldn’t it? It is a perfect sleight of hand.”

Finally, Tcherneva ends with a long discussion of capital gains that fails to address any of the methodological points I carefully lay out for why they are badly measured. That was my stated reason for leaving them out of my final two charts, not any contention that they shouldn’t be counted ideally.

Whatever, all that matters is that inequality has risen, and well, they’ve really shown me, haven’t they?

This piece originally appeared in Forbes

This piece originally appeared in Forbes