Thursday, October 14, 2010

Jobless claims revisions

Click on graph to enlarge

Normally, when data is revised, it averages out to some points being revised up if the initial data is too low, and some points being revised down if the initial data is too high. Over time, there is a regression to the mean, which is a neutral value of zero.

But after looking at the chart above of initial and continuing jobless claims, we find that upward revisions occur 75 out of 76 weeks. The odds of this happening by coincidence are statistically impossible. Obviously, there is a persistent bias to publish data which understates the true unemployment picture.

The Bureau of Labor Statistics should rename themselves the Bureau of Lies and Statistics.

No comments:

Post a Comment