Foundations of climate change denial: Anti-environmentalism and anti-science

PLOS One | https://doi.org/10.1371/journal.pone.0334544 November 12, 2025 9 / 23 Deductive learning tests theory, and inductive learning builds theory. Since the denial of AGW science is defined by trend, attribution, and impact denial, each book must contain at least one of the these three denial claims to be included in the dataset [29]. We also identified policy denial as a key claim made by CTTs early on [7], but it was not a requirement for inclusion. However, given the prominence of policy denial, stage one of the analysis uses all four categories to identify denial claims. Our analysis began with one of the authors reading all of the books three times, marking potential passages that reflected one of the four key claims of AGW denial. Then, a team of eight climate literate undergraduate researchers and the same author coded the passages for one or more of these claims in reading and discussion sessions where there had to be agreement on each code. One of the main challenges to deductive content analysis is inter-subjective validity [74]. Rourke and Anderson [75] argue that the process for developing valid protocols involves identifying the purpose and the behaviors/concepts that rep- resent the construct (categories), reviewing these categories, testing them, and developing guidelines for interpretation. To determine if the coding rules were reliable, a second set of eight undergraduate researchers and an author coded 13 passages with basic rules for either a rejection of a warming trend, humans’ contribution to GHG emissions, that warming

will have negative impacts, and the need for policies to reduce GHGs for a total of 117 decisions with 468 potential possi- bilities. Under these rules the inter-coder reliability was near perfect between the nine coders (Cohen’s Kappa,.892; Scot’s pi,.891), indicating our rules were reliable. Then, in order to analyze passages in MaxQDA 11.1.2 software (Stage Two), we digitized relevant coded passages into portable document files (pdfs), paying attention to the semantic ecosystem. It is important to note that climate denial authors are not always consistent in their claims or rationales, and it has already been noted that incoherent claims of the “world is cooling” can coexist with claims that warming is “natural” [76]. For example, Michaels (2004) proclaims, “Global warming is real, and human beings have something to do with it” (p. 9) but then in the same book warns that, “…IPCC’s scenarios of dramatic warming require the continuation of an exponential increase in carbon dioxide that stopped a quarter-century ago….” We solve this problem with data-driven empiricism by coding the plain language rejection of climate science where it occurs in the text, or “manifest” coding. Coding of a book stopped if all four denial categories were found--thus our data do not include every denial claim in every book. B. Stage two: open coding technique After the denial code, we move to unrestricted, inductive open coding of the rationale for denial. Open coding, sometimes

called emergent coding, means the rationale for denial emerges directly from the data allowing for theory building to occur as a result. As Cope [77] indicates, “The purpose at this stage is to ‘open up’ the data,…so that conceptual implications can emerge…” (p. 446). Typically, the narrative under analysis was one or two pages proximate to the denial claim within a chapter section. The length of the text unit was determined by the semantic relations which are “instantiated by context” of chapter sections which usually address a focused aspect of the narrative arc [78]. Each semantic network is in the form of speech acts by a discourse community and over time patterns emerge. As the patterns emerge, so does “coherent discourse” – i.e., the argument is legible [79]. In our data, we identify “cause and effect” as the coherent relation, which Hume [80] described as one of “three principles of connection among ideas” (p. 32)- the other two being “Resemblance, Contiguity in time or place” [67]. Coherent cause and effect discourse infers: “P from the assertion of S1 and Q from the assertion of S2, where normally P → Q” [79]. In other words, we can infer how one assertion must plausibly follow another with or without plain exposition. If P and Q exist within proximate passages, the discourses co-occur and co-occurring discourses inform each other’s meaning. We assume that words used in each book are related, and it is important to understand how the discourses are associated in the spirit of linguist J.R. Firth [81] who

noted that we understand the meaning of words by the “company it keeps.” So, when one claim is in proximity to another claim, together they create larger meta-discourses because the denial claim does not exist alone but within a forest of meaning that comes from the surrounding narrative. This is part of latent semantic analysis: