He never has. Check it out.
http://www.econbrowser.com/archives/2008/09/shadowstats_deb.html
This was a great comment.
The first thing that stood out to me about Shadow Stats is that William's figures are usually a constant spread from the official government figures. The derivatives of the graph almost never vary. In order to figure out why this is so, I went hunting through his "about" pages to find his methodology.
It turns out that this is not an accident; the similarity of the graphs is the result of his methodology. It seems that John Williams does not actually recompute CPI etc using the old techniques. What he does is take the current measurements, and then offsets them with a correction factor. This correction factor is derived from some internal government document which was created around the time that the measure is changed (e.g. someone writes a memo claiming that the new CPI will be 2% lower). Hence, his system has the same problems that the new government measurements have, namely that they cannot meaningfully be compared with the original government measurements. They are simply an alternate metric, and one of dubious quality.
This is a really shame. We desperately need a site that actually does what Shadow Stats claims (but fails) to do. If I hear another mathematically illiterate economist/pundit numerically compare 2000s CPI with 1980s CPI (or 2000s unemployment measurements with 1930s unemployment measurements), my head is going to explode. Posted by: Walker at September 5, 2008 06:50 AM