Beware the Noisy Data

Some disquiet on Twitter about the household survey indication that the unemployment rate for black men took a sizeable upward leap this month. I don’t want to dismiss those concerns, but due note that the same survey indicates that the population of white people in the United States peaked in December and that there are nearly 1 million fewer of us around than there were a few months back. That’s not actually happening, that’s just the survey exhibiting measurement error. One million people is a lot of people, but it’s less than one percent of the overall stock of white people in this country so these things happen. When you add in the fact that the unemployment rate is a ratio and that the black male sub-sample is relatively small, there’s lots of room for error.

I think our generic assumption should be that if overall employment levels are rising, then black male employment levels are also rising. If you see several months in a row of divergence, then it’s time to start looking into things. But a one month fluctuation can very likely be one of several measurement problems.