A while ago the GWPF (yes, just as dumb as trying to pronounce Joe Btfspik) tried one on, claiming that they were going to conduct a "temperature data review". Nick Stokes has made a habit, indeed a retirement of going there, in his own way, of challenging the ungodly in hell, or Climate Audit and the like, but Eli repeats himself. Bunnies should read Nick's submission, but there is one paragraph that Eli would like to emphasize, because while it is obvious, it is so obvious that entire castles of fancy have been built on ignoring it
Global averages are averages of millions of individual readings. Unbiased noise is very heavily damped in the result. The big source of error is bias, which is not damped in the same way. The essential purpose of homogenisation is to identify and minimise this bias. The tradeoff vs added unbiased noise is advantageous, because of that damping.In short averaging handles noise, homogenization is necessary to take care of bias.