Environ. Res. Lett. 12 (2017) 084019 Table 1. Inventory of documents analyzed. Shown for each document category are the total number of documents, their date range, source(s), and assigned types. Among peer-reviewed and non-peer reviewed documents, eight publications were found to be redundant, with similar or identical wording to seven other (strictly unique) publications. All 15 are included in our analysis. Among non-peer-reviewed documents, there are two citations provided by ExxonMobil that are identical to two others. The identical two are not included in our analysis. Sources: ‘Peer-Reviewed’ and ‘Additional’ publications are cited in the ‘Exxon Mobil Contributed Publications’ list [15]; ‘Supporting Materials’ are internal documents offered by ExxonMobil [59]; ‘Other’ sources refers to documents discovered independently during our research; ICN = InsideClimate News; NYT = The New York Times. NYT advertorials were collated by Polluter Watch [61]. For details on document types, see section S2, supplementary information, available at stacks.iop.org/ ERL/12/084019/mmedia. Miscellaneous Opinions include, for example, commentaries, opinion editorials, and speeches. Sources Document Types Provided by ExxonMobil Dates ‘Peer- ‘Additional’ ‘Supporting ICN NYT Other Academic reviewed’ materials’ journal Internal Ad Misc. doc. opinion No. Internal Documents PeerReviewed Non-PeerReviewed Advertorials 32 1977!1995 0 0 22 28 0 1 0 0 0 0 0 32 0 0 72 1982!2014 50 19 0 0 0 3 53 2 13 4 0 0 0 0 47 1980!2014 3 29 0 3 0 12 0 24 5 2 2 0 0 13 36 1989!2004 0 0 0 0 36 0 0 0 0 0 0 0 36 0 To characterize each document, we read its abstract, introduction, and conclusion, and either skim or read thoroughly the rest as necessary. In the case of long documents (over ∼30 pages) in which executive summaries are provided, we rely on those summaries. The documents are binned into four categories as shown in table 1: Internal, Peer-Reviewed, Non-Peer-Reviewed, and Advertorial. This allows us to distinguish communications according to degree of accessibility–a key variable in assessing the consistency of ExxonMobil’s representations of AGW. Each document’s manifest content is then further characterized in four ways: type, topic, position with respect to AGW, and position with respect to risks of stranded assets. Details of document types and topics are discussed in sections S2!3, supplementary information. 2.1. Document position Research has shown that four key points of understanding about AGW—that it is real, humancaused, serious, and solvable—are important predictors of the public’s perceived issue seriousness, affective issue involvement, support for climate policies, and political activism [62–66]. These four elements have also been found to underpin most narratives of AGW skepticism and denial (namely ‘it’s not happening’, ‘it’s not us’, ‘it’s not serious’, and ‘it’s too hard’) [28, 43, 67, 68]. We therefore use, a priori, these recognized elements as axes along which to characterize ExxonMobil’s positions on AGW in its communications; positions on each of these elements form the primary codes in our content analysis (table 2). Our coding scheme is summarized below (see section S1, supplementary information for further details). One of the authors coded all of the documents, and ambiguities were resolved through discussion between authors. To verify intercoder reliability and intercoder agreement, both authors independently 3 Conference/ workshop proceeding Gov. Book Industry report white paper Category coded a random subset of 36 documents (approximately 19% of the total number of documents in each category). Intracoder reliability was also calculated (see section S1.7, supplementary information). 2.1.1. ‘Real & human-caused’ Tailoring the approaches of Cook et al, Feldman et al, and Elsasser and Dunlap, each document is coded by assigning ‘Endorsement Points’ (EP1 to EP4b, defined in table 2) to pertinent text and figures based on whether each acknowledges or doubts the scientific evidence that AGW is real and human-caused (intercoder reliability of Endorsement Points: percentage agreement = 93%; Krippendorff ’s (Kripp.) a ¼ 0:84) [22, 27, 28]. We recognize that all science involves uncertainties, and therefore that doubt is not, ipso facto, an inappropriate response to complex scientific information. Uncertainties are an innate and important part of reasonable scientific discourse. However, it has also been shown that uncertainty may be amplified or exaggerated in ways that are misleading and unreasonable, sustaining doubt about claims that are scientifically established [42, 52, 57, 69]. Therefore, to distinguish reasonable and unreasonable doubt, we apply two first-order filters to our Endorsement Point codings. First, in documents published on or before 1990, we exempt expressions of doubt that AGW is real (i.e. we deem such expressions to be reasonable at that time). Second, in documents published on or before 1995, we exempt expressions of doubt that AGW is human-caused. 1990 and 1995 are when the Intergovernmental Panel on Climate Change (IPCC) first concluded that AGW is real and human-caused, respectively (these are conservative thresholds insofar as many scientists had arrived at these conclusions prior to the IPCC reports; indeed, IPCC reports are based only on already-completed work) [70, 71]. Finally, based on its individual Endorsement Points, each document is assigned one overall Endorsement