- One must carefully distinguish the cause and effect in many relationships.
- Very different mechanisms may give the same predictions, especially if we only verify them by looking at a few numbers such as monthly averages. Compare with the inverse LHC problem.
- That's the reason why higher-frequency data such as the daily data should be used to evaluate ideas. That's what they did in their recent paper.
- One should never assume that there is no correlation between X and Y, especially if the causal relationship between X and Y is what we want to understand. ;-)
- One should never neglect effects that are nominally equally important as those that are used to build an argument. For example, one shouldn't neglect precipitation if he considers evaporation of water - even though this is what most climate feedback people seem do be doing.
- And indeed, clouds and rain are pretty important for the weather and thus the climate. Recall your courses of Kindergarten Science 101.
He offers examples how these basic and other rules have been violated, refers to some colleagues who probably wanted to express similar ideas as himself but who were not sufficiently articulate, and constructs a toy model in which his statements are obvious.
Spencer's talk about cloud feedbacks in New York
Finally, he says another obvious thing: that one should look at very particular signals where different effects are separated because this is where one can learn a lot about the ways how the climate actually works. (I am improving what he says but he essentially gets it.) The 1991 eruption of Mt Pinatubo is an example of an event where such things can be deduced: this particular event seems to imply that the cloud feedback is negative - see Forster and Gregory.
If you think that there is a serious bug in Spencer's ideas, you should tell him because he won't have to write a new paper about it.
Thanks to M. Simon.