Environ. Res. Lett. 12 (2017) 084019 (a) 100 Real & Human-Caused Percentage of documents (% ) 90 80 70 60 50 40 30 20 10 0 Internal PeerReviewed Non-PeerReviewed (b) 100 Serious 90 Percentage of documents (%) Advertorials 80 70 60 50 40 30 20 10 0 (c) 100 Percentage of documents (% ) 90 80 70 Internal PeerReviewed Non-PeerReviewed Advertorials Acknowledge Solvable Acknowledge (incl. reasonable doubt) Acknowledge and Doubt Reasonable Doubt Doubt 60 50 40 30 20 10 0 Internal PeerReviewed Non-PeerReviewed Advertorials Increasingly accessible Figure 2. Percentage of documents taking each overall position on AGW as (a) real and human-caused, (b) serious, and (c) solvable. For each document category and for all documents that express a position in figure 1, the cumulative fractions of documents taking that position are shown. been filtered for reasonable doubt. Table 3 presents exemplifying quotations (coding units) of varying ‘strength’ that illustrate the assigned positions for a selection of the documents. For each category and for all documents that express a position, figure 2(a) shows the cumulative fraction of documents that take that position. Positions on AGW as real and human-caused vary significantly across document categories (p < 3:7 # 10!13 , Fisher’s exact test, FET; see table S3, supplementary information, for details and all probability values). Figure 2 is based on all documents in figure 1; the same trend is observed when only documents with an overlapping date range are considered (section S4, supplementary information). 7 3.1.1. Peer-reviewed publications Figures 1(a) and 2(a) show that ExxonMobil’s peerreviewed publications overwhelmingly acknowledge AGW as real and human-caused (‘Acknowledge’). Of the 65% (47/72) of peer-reviewed documents that express a position, more than three-quarters hold an ‘Acknowledge’ position (39/47 = 83%). Table 3 provides sample quotations (see section S7, supplementary information, for substantiating quotations for all documents). ExxonMobil’s listed publications include chapter 8 of the 1995 IPCC report (ExxonMobil’s principal climate scientist, Haroon Kheshgi, was a contributing author), which observed a ‘discernible human influence on global climate’ [15, 76]. Kheshgi also co-authored the Summary for Policymakers and several chapters of the next IPCC report in 2001, which found ‘there is new and stronger evidence that most of the warming observed over the last 50 years is attributable to human activities’ [77– 80]. Of the minority of peer-reviewed documents holding a position of ‘Acknowledge and Doubt’ (5/47 = 11%), ‘Reasonable Doubt’ (2/47 = 4%), or ‘Doubt’ (1/47 = 2%), we judge that most of the expressed doubt constitutes normal scientific discussion about uncertainties; for example, ‘the estimation of the uncertainty of this causal chain [linking human activities to changes in climate]’ [81]. 3.1.2. Non-peer-reviewed documents The predominant stance taken in non-peer-reviewed communications is also ‘Acknowledge’, although compared to peer-reviewed work, it loses ground to the ‘Acknowledge and Doubt’ and ‘Doubt’ stances in roughly equal measure (p ¼ 0:044, FET). Figures 1(a) and 2(a) show that, of the 74% (35/47) that take a position, 66% (23/35) ‘Acknowledge’, 17% (6/35) ‘Acknowledge and Doubt’, and 17% (6/35) ‘Doubt’ that AGW is real and human-caused. The more frequent expressions of doubt in non-peer-reviewed documents, compared with peer-reviewed ones, reflect the mixed nature of these documents. Some are technical, academic analyses, while others are industry-targeted speeches, reports, conference proceedings, company pamphlets, etc (see sections S2, S3, and S6, supplementary information). 3.1.3. Internal documents The bulk of ExxonMobil’s internal documents also take the ‘Acknowledge’ stance. Figures 1(a) and 2(a) show that, of the 63% (20/32) that take a position, 80% (16/20) adopt ‘Acknowledge’, with most of the rest expressing ‘Reasonable Doubt’ (3/20 = 15%). Unlike other document categories, however, our characterization of internal documents shifts dramatically if we remove filters for reasonable doubt from our analysis (see section 2). Then, 61% (11/18) take the mixed position (‘Acknowledge and Doubt’), with the remainder split between ‘Acknowledge’ and ‘Doubt’ (3/18 = 17% and 4/18 = 22%, respectively).