I am not anxious to defend Donald Trump's economic performance, but a complaint by Steven Rattner in his New York Times column caught my attention. The column features a graph showing that the ratio of the median wage for blacks has fallen relative to the median wage for white workers since 2016.

This bothered me, because in general the situation for blacks, and other groups who face discrimination, improves in a tight labor market. While it is possible that the plight of blacks has deteriorated due to Trump's open racism, there is another plausible explanation.

The improvement in the labor market has had a considerably larger impact on employment rates for blacks than for whites. If we compare averages for the first six months of 2019 with the 2016 average, the employment to population ratio for blacks has risen from 56.4 percent to 58.3 percent, a rise of 3.3 percent. For whites, the increase has been from 60.2 percent to 60.8 percent, a rise of 1.0 percent.

The increase in black employment has been a very important gain for hundreds of thousands of black families, but it also means that the composition of black workers has changed somewhat. Let's assume that the new workers have less education and experience than most of the people working in 2016.

In the extreme case, that they all fell near the bottom of the wage ladder, the increase in employment by 3.3 percent will mean that the median black worker in 2016 is now the 53.3 percentile worker, whereas the 46.7 percentile worker in 2016 is now the median worker.

When we compare medians from 2019 with 2016, we are looking at workers that were considerably further down the income ladder in 2016. This effect would be much smaller with white workers since their employment rates changed by less. 

One piece of evidence supporting this story is that Rattner's graph shows that drop in the ratio of medians under Trump essentially continues a trend that had been in place since 2012 when the labor market began to tighten substantially. To get the fuller picture, we really need to control for factors like education and experience or use longitudinal data sets that track the pay of individual workers through time, but it's fair to say that Rattner's graph really is not telling us much.