What did Cook et al. measure?
What do we mean by anthropogenic global warming? That human activities are the main cause of observed global warming. Cook et al. found only 64 of 11,944 articles that met that definition, 0.5% (their Level 1). They found a slightly higher number who reject AGW either implicitly or explicitly (Levels 5-7). What are we to make of such a tiny percentage of acceptance? There are two possibilities.
The first is that 99.5% of publishing scientists truly do not accept the humans are the main cause of global warming, in which case there is no consensus. Moreover, with so little support from publishing scientists, AGW would almost certainly be false.
The second is that (a) a high percentage of authors do accept AGW but (b) only a tiny percentage explicitly say so. This would falsify the Cook et al. method.
Thus either AGW is not the consensus position and likely false, or the Cook et al. method is false. But we know from other studies that there is a strong consensus on AGW. We know from the massive amount of evidence in its favor that AGW is almost certainly not false. Thus the Cook et al. method must be false. This means that they did not measure the consensus. But did they measure something useful?
To better understand what Cook et al. measured, like them I reviewed the abstracts of those 64 level 1 articles, assigning them to categories by subject matter. This is a highly subjective task. Moreover, the articles do not divide neatly into any particular set of categories—some fit into two or more. Still, the exercise provides some useful information.
I judge 35 of the 64 articles to be mostly about attribution (causes) and modeling, 25 about mitigation, and 4 about impacts. The category is not critical, because what I am after is why the authors might appear to endorse AGW, as Cook et al. required. Moving one of the 64 from one category to another would not make any difference to my concclusion. Let us consider the categories one-by-one.
As with the Hansen et al. (1993) article that I discuss here, articles about attribution (identifying the causes of global warming) will of course include a statement about causes. Many modeling papers will do the same if determining the effect of CO2 on temperature was a point of the research. Papers about mitigation often start by stating what it is that needs to be mitigated: man-made global warming. Those about the impacts of global warming may also set the stage with an opening statement like this one from one of the 64 articles: “Over the past 150 years the amount of CO2 in the atmosphere has been increasing, largely as a result of land-use change and anthropogenic emissions from the burning of fossil fuels.” This is not an "endorsement" but a statement of scientific fact.
In my opinion, none of the 64 articles that Cook et al. placed in level 1 in any sense “endorses” AGW. Rather, in each case the authors made a statement about the cause of global warming either because their subject matter and findings required it or because they found it a useful device to set the stage. There is no reason to believe that authors in Cook et al. level 1 more strongly accept AGW than those in other non-rejecting levels, including level 4: No position.
This leads me to conclude that Cook et al. measured not the consensus but the fraction of published articles that lent themselves to a direct statement about AGW.
Cook et al. did uncover some useful information: that only 0.2% of 11,944 articles explicitly rejected AGW (Levels 6 and 7). They wrote that "...the number of papers rejecting the consensus on AGW is a vanishingly small proportion of the published research." They would then have been on firm ground in concluding that the consensus of authors averaged over the 20 years of their survey is well above 99.8%. (Rejecting articles tend to have a single author, others an average of about three.) This would have denied the deniers the wiggle-room that the 97%-3% provides. But regrettably, Cook et al. passed up that opportunity.