From Scott Sumner:
A few years back I got so exasperated reading a Journal of Economic Perspectives piece on income inequality (by Emmanuel Saez and Peter Diamond) that I did a post calling it “propaganda.” I probably shouldn’t have used that term, but I was reminded of my frustration when reading a very good Alphaville post by Cardiff Garcia:
The issue of whether US inequality has climbed since the recession of 2008 has been relitigated this week. A short analysis by Stephen Rose claimed that income inequality had actually fallen, assigning the credit to public policy.
David Leonhardt of the New York Times discussed Rose’s findings, followed by further analyses and critiques from Ben Walsh and Nick Bunker. I’ll present the findings first before adding my own thoughts at the end.
Mainly in response to the heavily cited claim by Emmanuel Saez that 95 per cent of the income gains in this recovery have gone to the top 1 per cent of earners, Rose emphasizes a couple of broad points.
There are two problems with the 95% claim, one has already been discussed by David Henderson, while the other is often overlooked. David pointed out that when evaluating income equality you want to remove cyclical effects, as it’s a long term problem. It’s not unusual for the share of income going to the rich to fall during recessions (as capital gains plunge), and then rise during expansions. It would make more sense to compare 2014 to a year with similar unemployment, say 2004.
The less often discussed problem is that talking about shares of growth can be very misleading, especially when growth is slow.
I would file this under “check your assumptions.” Most people do not look to see exactly what any given data set is really measuring, or how this data might be expected to change under specific circumstances (that is, the “normal” behavior of the data). They just hear the headline name of the data set and start making their own assumptions about its meaning.
It’s not propaganda, either way. It’s just data, and if you want to have an opinion about it that has any basis in reality, you should find out what the data really IS. Check your assumptions.