In a normal error analysis for a large data set, scientists would assume that any non-systematic error is random. A miscalibrated thermometer is just as likely to read too high as it is to read too low. Human error has no systematic bias. The only systematic error is UHI, which should cause the corrections to be increasingly negative.

ts.ushcn_anom25_diffs_urb-raw_pg.gif (650×502)

So what are the odds of 840 months in a row having temperatures which were read too low by thousands of observers? Two raised to the 840th power. For all practical purposes, that number is infinity.

In other words, there is a 0% chance that the USHCN adjustments are correct, and 100% probability that they are politically motivated, or at best conformation bias.