The data are from Piketty and Saez. Curiously, these authors don’t analyze the income shares of the bottom 90th percentile. They say this is because before 1945 most people were exempt from filing tax returns except top earners. But I suspect they think these data are a bad measure of bottom 90th percentile even after that because in the working paper, the published paper, the comment and the “summary for the broader public” the shares of the bottom 90th percentile are never reported. In any case, this means whoever at the WSJ that created the chart extrapolated from Piketty and Saez’s data on top income shares and didn’t take the data directly from them.
Tax data have a number of problems, all addressed in their paper. There’s evasion, exemptions, and income shifting. Also, because most taxes are filed by family instead of by individual, tax data can only be used to look at family incomes. This means some of the trends will be the result of demographic changes (i.e. families differentially getting smaller in each income percentile).
But Piketty and Saez use tax data, instead of survey data (e.g. Census), to analyze top income shares because given there’s no random sample, the top of the top have a proper representation in the data set. In a random survey, its unlikely you’ll end up surveying one of the 14,836 families that make up the top 0.01% earners. They’re using the right data for the job, but the job isn’t to analyze shares of the bottom.
That’s why we have survey data. And I happen to have survey data, the 2000 Census and the 2006 CPS, sitting on my desktop. By my figuring, wages of the bottom 90th percentile went up by 5% between those years.
So, we should fight over what data is better, but its at least important to know that survey data gives a different answer than tax records. Its a bit premature to conclude, as Setser did, “Most Americans didn’t benefit from the expansion of the past few years.” That said, a 5% increase is meager over 6 years. I won’t be throwing any parties.