The use of artificial corrections, the merging of data from different sources, and the curtailing of graphs to hide trends that conflict with your hypothesis, are frowned-upon in High Energy Physics, Engineering, and Neurology, which are my fields, and according to this eloquent comment, they are frowned upon in Medicine as well.
In my experience, these "tricks" (as Phil Jones calls them), allow us to uphold any hypothesis without supporting evidence. In the case of global warming, the tricks went like this. First, we have a trend from global surface stations, which shows a +0.6°C rise in the 20th Century. We claim, but do not prove, that we have accounted for all sources of error. We now receive other observations of temperature. If these new observations agree with our +0.6°C trend, we accept them. The rumor that ice is melting fast and glaciers are retreating is well-received, as are images of polar bears swimming. These new observations, we claim, give us more confidence in our +0.6°C trend. Observations that disagree with our +0.6°C trend we reject. The Briffa tree ring data is curtailed or "smoothed into instrument data" (Gavin Schmidt) in order to "hide the decline" (Phil Jones's). Satellite measurements showed no rise in temperature when they first came out, but this leads us into a flurry of efforts to find sources of error in the analysis that might have hidden the upward trend that we all know must be there from our many supporting temperature observations. Eventually, one or two groups studying satellite data say they see a slight rise. That's good enough. When we combine temperature histories, we give more weighting to those that agree with our hypothesis than those that disagree (see Mann-Made Warming). Our combined trends look great. Now someone comes and challenges our original global surface trend, and we say, "That actual temperatures have been rising is unequivocal and demonstrated by ocean temperatures, retreating glaciers, melting snow, etc. etc. etc." (Gavin Schmidt).
And so our global warming theory pulls itself up by its own bootstraps.
UPDATE: An interview.