Recurrent Fury: Conspiratorial Discourse in the Blogosphere Triggered by Research on the Role of Conspiracist Ideation in Climate Denial

Conspiracism and Rejection of Climate Science

In the case of climate change, several qualitative analyses have shown that denial is suffused with conspiratorial themes, for example when dissenters are celebrated as “Galileos” who oppose a corrupt scientific “establishment”. Researchers have suggested that the dissenters’ creation of such a shared social reality provides a stage upon which public climate denial can unfold (McKewon, 2012b, a).ⁱ Public accusations of conspiracies within the Intergovernmental Panel on Climate Change (IPCC) were aired in the opinion pages of the Wall Street Journal (WSJ) as early as 1996 (Lahsen, 1999; Oreskes & Conway, 2010), in a piece that alleged a “(...) disturbing corruption of the peer-review process.” The charges focused on a chapter of the 1995 IPCC report that was concerned with the attribution of global warming to human activities. The WSJ piece was authored by an individual who had no part in the IPCC process, and subsequent scholarly work traced the origin of the charge of conspiracy and corruption to a document produced by the Global Climate Coalition, a lobby group representing primarily the energy sector (Lahsen, 1999). In her analysis of this controversy, Lahsen (1999) concluded that conspiracy theories are “(...) rhetorical means by which to cast suspicion on scientific and political opponents” (p. 133). Accordingly, the titles of many recent popular books critical of mainstream climate science hint at a conspiracy. Table 1 shows a non-exhaustive sample of titles. Similarly, Koteyko, Jaspal, and Nerlich (2013) showed in an analysis of UK tabloid readers’ comments on climate change that contrarian constructions such as “tax scam” were a main theme of discourse that arguably could be considered conspiracist.

At a behavioral level, Smith and Leiserowitz (2012) found that among people who reject the findings from climate science, up to 40% of affective imagery invoked conspiracy theories. That is, when asked to provide the first word, thought, or image that came to mind in the climate context, statements such as “the biggest scam in the world to date” would be classified as conspiracist. Similarly, a pair of recent surveys uncovered an association between endorsement of unrelated conspiracy theories and the rejection of scientific propositions, including the findings from climate science (Lewandowsky, Gignac, & Oberauer, 2013; Lewandowsky, Oberauer, & Gignac, 2013). Because those articles form the basis of the present research, we describe them briefly.

The first study sampled visitors to climate blogs (Lewandowsky, Oberauer, & Gignac, 2013), and the “in press” paper and data became available for download in July-August 2012. We thus refer to this paper as LOG12 from here on. The second study involved a nationally-representative sample of Americans and was published online, and the data made available, in October 2013 (Lewandowsky, Gignac, & Oberauer, 2013). We call this paper LGO13 from here on. In replication of much prior research (e.g., Heath & Gifford, 2006; Kahan et al., 2012), LOG12 and LGO13 identified political ideology—in particular, the strength of endorsement of laissez-faire free market economics—as being the principal driver of the rejection of climate science. Conspiracist ideation, measured by endorsement of items such as “A powerful and secretive group known as the New World Order are planning to eventually rule the world” constituted another but lesser contributing factor. Notably, notwithstanding the rather different pools of participants and differences in methodology, the size of the effect of conspiracist ideation on rejection of climate science (r ≈ .20 between latent variables) was virtually identical across both studies. The results mesh well with the findings by Smith and Leiserowitz (2012) and the body of prior research that has linked conspiracist ideation to the rejection of science (Diethelm & McKee, 2009; Goertzel, 2010; Kalichman, 2009; McKee & Diethelm, 2010) and the acceptance of pseudoscience (Lobato, Mendoza, Sims, & Chin, 2014).

Method

Conspiracist Classification Criteria

To process the corpus and to test for the presence of conspiracist discursive elements, we derived six criteria from the existing literature (see Table 3). Our criteria were exclusively psychological and hence did not hinge on the validity of the various hypotheses. This approach follows philosophical precedents that have examined the epistemology of conspiratorial theorizing irrespective of its truth value (e.g., Keeley, 1999; Sunstein & Vermeule, 2009). The approach also avoids the need to discuss or rebut the substance of any of the hypotheses.

The first criterion is that the presumed motivations behind any assumed conspiracy are invariably nefarious or at least questionable (Keeley, 1999): Conspiracist discourse never involves groups of people whose intent is to do good, as for example when planning a surprise birthday party. Instead, conspiracist discourse relies on the presumed deceptive intentions of the people or institutions responsible for the “official” account that is being questioned (Wood, Douglas, & Sutton, 2012). This criterion applies, for instance, when climate science and research on the harmful effects of DDT are interpreted as a globalist and environmentalist agenda designed to impoverish the West and push civilisation back into the stone age (Delingpole, 2011). When presenting the results, we refer to this criterion as Questionable Motives, or QM for short (see Table 3).

A corollary of the first criterion is that the person engaging in conspiracist discourse perceives and presents her- or himself as the victim of organized persecution. At least tacitly, people who hold conspiratorial views also perceive themselves as brave antagonists of the nefarious intentions of the conspiracy; that is, they are victims but also potential heros. The theme of the victimization and potential heroism features prominently in science denial, for example when isolated scientists who oppose the scientific consensus that HIV causes AIDS are presented as persecuted heros and are likened to Galileo (Kalichman, 2009; Wagner-Egger et al., 2011). We refer to this criterion as Persecution-Victimization or PV for short.

Third, conspiracist ideation is characterized by “(...) an almost nihilistic degree of skepticism” (Keeley, 1999, p. 125) towards the “official” account. This extreme degree of suspicion prevents belief in anything that does not fit into the conspiracy theory. Thus, nothing is at it seems, and all evidence points to hidden agendas or some other underlying causal mechanism. We label this criterion Overriding Suspicion or OS.

Fourth, the overriding suspicion is often associated with the belief that nothing happens by accident (e.g., Barkun, 2003). Thus, small random events are woven into a conspiracy narrative and reinterpreted as evidence for the theory. For example, the conspiracy theory that blames the events of 9/11 on the Bush administration relies on evidence (e.g., intact windows at the Pentagon; Swami, Chamorro-Premuzic, & Furnham, 2010) that is equally consistent with randomness. We label this criterion Nothing is an Accident, or NoA for short.

Fifth, the underlying suspicion and lack of trust contribute to a cognitive pattern whereby specific hypotheses may be abandoned when they become unsustainable, but those corrections do not impinge on the overall abstraction that “something must be wrong” and the “official” account must be based on deception (Wood et al., 2012). In the case of LOG12, the “official” account is the paper’s conclusions that conspiracist ideation contributes to the rejection of science; and it is this conclusion that must be wrong according to this criterion. At that higher level of abstraction, it does not matter if any particular hypothesis is right or wrong or incoherent with earlier ones because “ (...) the specifics of a conspiracy theory do not matter as much as the fact that it is a conspiracy theory at all” (Wood et al., 2012, p. 5). We label this criterion Must be Wrong (MbW).

Finally, contrary evidence is often interpreted as evidence for a conspiracy. This idea relies on the notion that, the stronger the evidence against a conspiracy, the more the conspirators must want people to believe their version of events (Bale, 2007; Keeley, 1999; Sunstein & Vermeule, 2009). This self-sealing reasoning may widen the circle of presumed conspirators because any contrary evidence merely identifies a growing number of people or institutions that are part of the conspiracy.

Concerning the rejection of climate science, a case in point is the response to events surrounding the illegal hacking of personal emails of climate scientists, mainly at the University of East Anglia, in 2009. Selected content of those emails was used to support the theory that climate scientists conspired to conceal evidence against climate change or manipulated the data (see, e.g., Montford, 2010; Sussman, 2010). After the scientists in question were exonerated by nine investigations in two countries, including various parliamentary and government committees in the U.S. and U.K., those exonerations were re-branded as a whitewash (see, e.g., U.S. Representative Rohrabacher’s speech in Congress on 8 December 2011), thereby broadening the presumed involvement of people and institutions in the alleged conspiracy. We refer to this criterion as Self-Sealing, or SS for short.

Results

Discussion

The thematic analysis provides an initial sketch of the blogosphere’s collective response to LOG12. The contribution of Study 1 was to classify the blogosphere’s discourse into a number of potentially conspiracist hypotheses on the basis of a number of well-established criteria. This result meshes well with the prior literature, which has repeatedly linked the denial of well-established scientific findings to conspiracist ideation (e.g., Diethelm & McKee, 2009; Goertzel, 2010; Kalichman, 2009; McKee & Diethelm, 2010).

However, Study 1 has two limitations that need to be addressed before we can explore the full implications of the results. First, any thematic analysis necessarily contains a subjective element, which opens the door to alternative interpretations. Second, this problem is compounded by the fact that two of the present authors also contributed to LOG12, thereby creating a potential conflict of interest.

We addressed these problems in a number of ways: First, the data collection for Study 1 (via Internet search) was conducted by two of the present authors who were not involved in analysis or report of LOG12. Second, the fourth and fifth authors (S.B. and E.A.L.), who had no involvement with LOG12 whatsoever, provided an independent audit of Study 1, reviewing the data and the categorization of it presented in Table 3. Specifically, S.B. first examined for semantic clarity the 7 criteria employed in characterizing statements as at least partly conspiratorial in character (questionable motives, overriding suspicion, persecuted victim, must be wrong, no accident, self-sealing, and unreflexive counterfactual thinking). He next examined the application of groups of these criteria to the 10 summary hypotheses in the study. To assess the justificatory question of whether these hypotheses exhibited these characteristics, he examined the original statements on blogs from which the summary hypotheses were drawn. On the basis of these two steps, he concluded that the summary hypotheses appeared entirely accurate as representations of the remarks being classified under those hypotheses, and that they were accurately characterized as exhibiting the combinations of the criteria for conspiratorial thinking found in them by the study. Similarly, E.A.L. independently confirmed both the logic and accuracy of the thematic analysis in Study 1. Her examination focused on sorting out the truth conditions of the cases, as well as analyzing the counterfactual nature of others. Her focus was on the careful and accurate attribution of the citations used in this study. The results of her detailed analysis confirmed the original work done by the fellow authors of this study.

Third, and arguably most important, we now report two behavioral studies that sought independent support for the thematic analysis using naive participants in a blind test. For those studies, an anonymized and enlarged set of comments was harvested off the Internet in a further search conducted by hired research staff with no association with the present authors.

Method

Results

We first asked what proportion of content items participants did not classify into any of the hypotheses in the scoring sheet. For the combined set, the proportion of items that were not classified (by responding with 0) was around 10% (mean N across subjects = 51.4). Those proportions differed slightly between the “depth” set (N = 15.6, 12.4%) and the “breadth” set (N = 35.8, 9.4%). Overall, the vast majority of content items appeared amenable to classification. The 5 “other” categories were used between 8.5% and 19% of the time for primary classifications (mean N across subjects = 63.0).

We next examined the consistency of those classifications (including the “other” categories) across participants. For this analysis, we considered the mandatory primary classifications and the secondary classifications where present (.28 of all cases). The small proportion of optional tertiary (.04) classifications were not considered. Because the instructions to participants did not differentiate multiple classifications by importance (i.e., participants were instructed to use “up to 3” classifications if they thought a “comment refers to several hypotheses”), primary and secondary classifications were considered interchangeably. For example, if for a given comment, 4 participants provided a primary classification into hypothesis X, and the fifth participant’s primary classification was Y but that person’s secondary classification was X, then this was counted as agreement among 5 participants.^vii

Table 4 shows agreement among participants for the overall set of 508 comments, and broken down by search (i.e., the “depth” search for Study 1 and the new “breadth” search for Study 2. The table shows that between 62% and 68% of comments were classified identically by 3 or more judges; given that there were 38 candidate classifications (plus the 0 option) the agreement is at least satisfactory. When classification is measured at the level of the 6 over-arching categories of hypotheses, the agreement increases considerably (to 87%) as shown in Table 5.

We next asked whether the principal hypotheses from Table 3 that arose from the thematic analysis in Study 1 were also detected by the participants in Study 2. Using the “depth” set from Study 1, we tabulated the number of comments that at least 3 participants classified identically. Table 6 shows the results. It is clear that participants broadly recovered the classification and frequency information revealed by the thematic analysis in Study 1, despite the fact that content items here were anonymized and, where necessary, edited for length.

Further statistical confirmation for this independent recovery is provided by correlating the number of comments assigned to each of the principal hypotheses from the thematic analysis by the present participants (bold-faced in Table 6) and the corresponding frequency data from Study 1 that were extracted during the thematic analysis (see Table 3). The Spearman correlation between those two sets of numbers was significant, ρ = .87, p < .003 (Pearson’s r = .93, p < .0003). Note that the absolute numbers of assigned comments may differ between Tables 3 and 6 because participants in this study were provided with a much larger set of candidate hypotheses, comprising more nuanced sub-classifications of the broader categories obtained in Study 1.

A final question concerns potential differences between the two search sets (i.e., the “depth” set used in Study 1 and the “breadth” set generated for this study). Table 7 shows the frequency of classifications for the “breadth” set, presented in the same manner as before. Comparison of Tables 6 and 7 reveals that the frequency pattern of hypotheses is broadly similar between the two sets. In particular, 6 of the 7 principal hypotheses in Table 3 were also identified in the “breadth” set, although several additional hypotheses were added.

The frequency information in Table 7 was again compared to the corresponding information in Table 3, and the Spearman correlation was significant, ρ = .87, p < .002 (Pearson’s r = .88, p < .002). Hypotheses that did not appear in Table 7 were represented by a frequency of zero.

Table 7 also permits a quantitative assessment of the prevalence of conspiracist discourse in response to LOG12: Given that the “breadth” search involved a neutral search string, this part of the corpus (382 content items) was in no way slanted towards inclusion of conspiracist content. It follows that the number of comments that participants assigned to the recursive hypotheses from Study 1 (N = 77; see all bold-faced entries in Table 7) provide an estimate of the prevalence of conspiracist discourse in response to LOG12. On that basis, 20% of the content items can be considered conspiracist. We present further evidence from the extant literature in the General Discussion that further highlights the widespread use of conspiracist language in blogs that reject well-established scientific propositions.

Discussion

Study 2 provided independent support for the classification of hypotheses and their frequency that formed the basis of our thematic analysis in Study 1. Because Study 2 compared content material from two different search sets, the results suggest that the classifications used in the thematic analysis were not contingent on a particular search strategy.

Critics might argue, however, that the present results were biased by the fact that participants were provided with a fixed set of choices (i.e., the 38 hypotheses on the scoring sheet), and that the emergence of the hypotheses in Table 3 is therefore not altogether surprising. Several considerations weaken this potential criticism: First, participants had the option not to classify a content item if they felt that it did not fall into any available category (by placing a 0 into the response column, which occurred around 10% of the time). Second, the scoring sheet offered 5 different generic “other” options, so participants had ample opportunity to sidestep the suggested response options. Finally, participants were given the opportunity to create their own free-form categorizations for the content material, although none chose to do so. On balance, we argue that it is implausible that the results are an artifact of the scoring sheet.

A further criticism might point to the fact that Study 2 yielded several additional hypotheses (see Tables 6 and 7) that were not uncovered by the thematic analysis in Study 1. Could this be taken to imply that the thematic analysis was incomplete? Our response is threefold: First, the thematic analysis was intended to establish the likely presence of conspiracist ideation in the blogosphere, rather than provide a complete inventory of all such discourse in the blogosphere. Hence, Table 3 was intentionally limited to those hypotheses to which at least one criterion for conspiratorial ideation applied. Second, inspection of Tables 6 and 7 suggests that even some of the novel classifications have conspiratorial attributes (e.g., “colluding with journalists”). Third, and most important, we next report an experiment that sought independent confirmation of the conspiratorial attributes of the content giving rise to all hypotheses in Tables 3, 6, and 7. The final study also examined whether the hypotheses advanced by the blogosphere differ from conventional scholarly critique.

Method

Results

Due to equipment failure, data from 3 subjects were missing responses to the final 11, 5, and 37 content items, respectively. All subjects were retained for analysis (the results do not change appreciably if participants with missing observations are omitted). Responses were averaged across trials separately for each content type to yield two data points per subject (one for PhD and one for web) on each of the 5 measures. Those scores were first entered into 5 separate between-within subjects analyses of variance to examine whether the content set (“depth” from Study 1 vs. “breadth” from Study 2) had an effect on responding. Neither any of the main effects nor any of the interactions involving content set were significant in any of the analyses, with the largest F (1, 23) = 2.78, p > .10 for the interaction between type of comment and content set for Suspicion. The remainder of the analysis thus collapsed across content sets and focused on the effect of comment type (i.e., web vs. PhD).

Five independent analyses of variance revealed significant main effects of comment type for all measures. The test statistics are shown in Table 9 and boxplots of the data are provided in Figure 1. It is clear that web comments were rated considerably higher than the PhD comments on all 4 measures that tapped conspiracist attributes, whereas the reverse was true for the single item examining the scholarly nature of a content item.^viii

Click to enlarge

Figure 1

Summary of overall responses for all 5 measures in Study 3. Thick horizontal lines are medians, and medians differ significantly if the notches between boxes in each panel do not overlap. See Krzywinski and Altman (2014) for interpretative guidance of notched box plots.

The last two columns in Table 9 provide further information about the distributions of responses. The columns show the proportion of extreme responses (i.e., the top of the scale; 9) across all trials and participants that were provided for each question and content type. In most circumstances, the incidence of such extreme responses in behavioral research is low as people tend to avoid the extremes, especially with a relatively fine-grained scale, such as the one used here with ten response classes. Accordingly, only about 5% of judgments for the PhD comments used this extreme category. For the web material, by contrast, the proportion of such responses was considerably greater for the questions that targeted conspiracist ideation. For the “something must be wrong” question in particular, more than a third of all responses used the most extreme value, with “questionable motives” and “suspicion” lagging not far behind. It was only for the single question examining the scholarly thrust of a comment that extreme endorsements were rare (3%) for the web comments (and comparatively greater for the PhD comments; 8%).

A visual illustration of the skew in the response distributions for web content is shown in Figure 2. The dark gray histograms represent the distribution (normalized for the area to sum to unity) of responses to web content items across items and subjects. The superimposed dashed line represent a kernel density estimate of the PhD-generated content for comparison. The skew towards extreme values for web content on the conspiracist questions (albeit attenuated for “victim”) is readily apparent, as is the (attenuated) reverse skew for the question targeting legitimate scholarly critique.^ix

Click to enlarge

Figure 2

Histograms (dark gray) for all responses to web content (not averaged across trials within a participant) for all 5 questions in Study 3. The dashed line that encloses the cross-hatched area represents Gaussian kernel density estimates (using 1.5 times the standard bandwidth) of the equivalent responses to PhD-generated content.

Discussion

The results of Study 3 are straightforward: Compared to material generated by junior scholars who were instructed to be as critical as possible of LOG12, the discourse in the blogosphere scored higher on the four criteria for conspiracist ideation tested in this study. One potential criticism of this result is that the junior scholars—notwithstanding the promise of anonymity and notwithstanding the fact that they were not known to any of the present authors—might have been constrained by professional courtesy and therefore might have attenuated the full extent of their critique. This criticism can be deflected on the basis of two aspects of our results: First, participants judged the PhD-generated material to offer a more thought-out criticism than the responses of the blogosphere. This significant difference is difficult to reconcile with attenuation (and could only be even greater if any possible attenuation were absent). Second, the fact that a large proportion of blogosphere material was given an extreme rating on most of the potentially conspiracist attributes is diagnostic by itself, even in the absence of any comparison with the PhD-generated material. The web-content items thus largely spoke for themselves and although the PhD-generated material provided comparative context, it was not essential to our conclusions.

Our results were obtained with anonymized material that was presented with no guiding context, to participants who knew nothing about the issue under consideration. Moreover, the blog comments included content from both Internet searches conducted for this project (i.e., the “depth” search for Study 1 and the “breadth” search for Study 2), and the two different search sets were found to make no difference to people’s ratings. The data thus again support the results of our thematic analysis in Study 1.

Potential Limitations

The present studies were concerned with the blogosphere’s response to a single 4,000-word article. One might therefore question the generality of our results. In response, we note that at least one other scientific report in the climate arena engendered a sustained critique that subsequent scholarly analysis identified as conspiracist (Lahsen, 1999). Likewise, accounts by climate scientists of the strategies of climate denial are replete with accounts of conspiratorial accusations against individual papers (e.g., Mann, 2012).

A second criticism might cite the fact that in our thematic analyis we have considered the “blogosphere” as if it were a single entity, analyzed within the context of psychological processes and constructs that typically characterize the behavior of individuals rather than groups. Our response is twofold: First, at the level of purely descriptive narrative methods, our work fits within established precedents involving the examination of communications from heterogeneous entities such as the U.S. Government (Kuypers, Young, & Launer, 1994) or the Soviet Union (Kuypers, Young, & Launer, 2001). Second, at a psychological level, numerous psychological constructs—such as cognitive dissonance, social dominance orientation, or authoritarianism—have been extended to apply not only to individuals but also to groups or societies (e.g., Moghaddam, 2013). We therefore argue that our extension of individual-level work on conspiracist ideation to the level of amorphous groups fits within precedent in two areas of scholarly enquiry.^x

A further criticism might hold that although we may have presented some evidence for the presence of conspiracist ideation, the evidence falls far short of “real” conspiracy theories involving events such as 9/11 or the moon landing. In response, we note that the hypotheses leveled against LOG12 do not differ qualitatively—that is, in terms of magnitude or scope—from others that have been identified as conspiracist in the context of another paper in the climate arena (Lahsen, 1999) or that have been observed in response to experimental manipulations (Whitson & Galinsky, 2008). We suggest that conspiracist ideation, like most other psychological constructs (e.g., extraversion), represents a continuum that finds expression to varying extents in conspiratorial theories of varying scope.

Critics might further invoke the fact that two of the present authors also authored LOG12, thereby creating a potential adverse conflict of interest. On this view, the response of the blogosphere represented legitimate criticism of LOG12, and the present article is merely an attempt to deflect the impact of that criticism. We argue against this view on multiple grounds. First, the present article fits squarely with precedents in the scholarly literature of researchers analyzing or reporting events arising from their own work (e.g., Landman & Glantz, 2009; Olivieri, 2003). For example, Landman and Glantz (2009) provide an in-depth analysis of the attacks by the tobacco industry on the research of co-author Stanton A. Glantz. Similarly, Olivieri (2003) provides a personal account of events arising from her pharmaceutical research, which was subject to legal threats and interference from a drug company. Both of those articles report events of considerable scholarly and public importance and would not have been possible without the involvement of a concerned party. We argue that this equally applies to the present article. Second, and going beyond those precedents, two of the present authors who had no involvement with LOG12 whatsoever audited, and confirmed, the thematic analysis of Study 1 independently. Finally, and most important, it is difficult to conceive of ways in which the outcome of the behavioral studies in this article—which confirmed the thematic analysis—could have been affected by a potential conflict of interest. All research staff and participants were either completely unknown to the authors or known only in the most casual and superficial manner. None had any prior academic, scholarly, or financial connection to the authors.

It must also be noted that the present article arguably goes against the interests of the LOG12 authors because it places several criticisms of LOG12 into the peer-reviewed literature that previously had been limited to Internet blogs. Given the well-known resistance of information to subsequent correction (e.g., Lewandowsky, Ecker, Seifert, Schwarz, & Cook, 2012; Lewandowsky, Stritzke, Oberauer, & Morales, 2005), the exploration of this criticism without any rebuttal (except for hypotheses 2, 6, 7, and 8 in Table 3 that could be unambiguously classified as false) therefore does not constitute a defense of LOG12—quite on the contrary, it may raise questions about LOG12 in the readers’ mind.

Bearing in mind those potential limitations, we now explore the two goals stated at the outset. We explore the implications of our work for understanding of conspiracist ideation and for the role of the blogosphere, and the public more generally, in the conduct of science.

Implications for Understanding Conspiracist Ideation

Our principal thesis is that some of the responses to LOG12 voiced in the blogosphere satisfy attributes of conspiracist ideation by the criteria defined at the outset. Two attributes deserve to be highlighted: First, most of the hypotheses can be unified under the immutable belief that “there must be something wrong” (Must be Wrong; see Table 3) and that the authors of LOG12 engaged in intentional malfeasance (Questionable Motives, Overriding Suspicion).

Whereas suspicion on its own is insufficient to identify conspiracist ideation, it arguably constitutes one of its core attributes. For example, the suspicion that LOG12 did not contact “skeptic” bloggers tacitly invokes several major presumptions, namely (a) that the authors of LOG12 were willing to engage in research misconduct; (b) that they would invent a claim about a non-event and publish it in a Method section when there was no incentive or reason to do so; and (c) that they should have somehow provided more “evidence” for the method they used beyond writing an accurate Method section.

Indeed, most of the hypotheses advanced about LOG12 included an accusation of intentional wrong-doing by the authors (viz. minimally hypotheses 2, 4, 5, 6, and 8), which goes beyond pointing to problems and errors as would be the norm in conventional scientific discourse (and as was empirically confirmed in Study 3). The ease with which those presumptions about misconduct and malfeasance were made and accepted provides a fertile environment in the blogosphere for the subsequent unfolding of conspiracist ideation (cf. Keeley, 1999; Wood et al., 2012).

Those underlying beliefs infused conspiracist elements even into those hypotheses that would be expected to arise during routine scholarly critique. For example, the scamming hypothesis evolved continuously without being guided by clear a priori assumptions about what would constitute a scammed response profile, thereby ultimately rendering this hypothesis self-sealing and unfalsifiable (criterion Self-Sealing). It is this psychological attribute, combined with the lack of clear a priori standards for a profile for a “fake” response, that points towards a conspiracist component rather than conventional scholarly critique.

Second, self-sealing reasoning also became apparent in the broadening of the scope of presumed malfeasance on several occasions. When ethics approvals became public in response to an FOI request, the presumption of malfeasance was broadened from the authors of LOG12 to include university executives and the university’s ethics committee. Similarly, the response of the blogosphere evolved from an initial tight focus on LOG12 into an increasingly broader scope. Ultimately, the LOG12 authors were associated with global activism, a A$6 million media initiative, and government censorship of dissent, thereby arguably connecting the response to LOG12 to the grand over-arching theory that “climate change is a hoax.”

Notably, even that grand “hoax” theory is occasionally thought to be subordinate to an even grander theory: one of the bloggers involved in the response to LOG12 considers climate change to be only the second biggest scam in history. The top-ranking scam is seen to be modern currency, dismissed as “government money” because it is not linked to the gold standard. The observed broadening of scope meshes well with previous research that has identified stable personality characteristics that predict the propensity for conspiracist ideation (cf. Douglas & Sutton, 2011; Goertzel, 1994; Swami et al., 2010). It is therefore not altogether surprising that suspicions about a single scholarly paper can rapidly mature into more encompassing hypotheses.

Possible Future Research Directions

Our research points to at least two issues that merit further investigation. The first issue concerns the prevalence of the discourse we have documented in this article.

Implications for the Public’s Participation in Science

Could the activity in the blogosphere have constituted legitimate scholarly critique of LOG12, rather than a partially conspiracist discourse as is argued here? Several considerations speak against that possibility. First, the data from Study 3 show that unbiased observers were unable to discern much scholarly value in the blogosphere’s response to LOG12.

Second, we already noted at the outset that the public discourse in the blogosphere was accompanied by non-public events to prevent or delay the publication of LOG12. The attempts to suppress publication of a peer-reviewed paper, in conjunction with the absence of any discernable conventional scholarly activity, speak against the possibility that the blogosphere’s discourse represents legitimate public engagement with science. There are indications that seeking the retraction of inconvenient papers has become a routine practice among individuals who are denying (climate) science: We already noted the circumstances surrounding Recursive Fury at the outset. Moreover, one of the bloggers who was also involved in the response to LOG12 recently called for the retraction of a peer-reviewed paper that had underscored the pervasive scientific consensus on climate change through an analysis of nearly 12,000 peer-reviewed articles (Cook et al., 2013). To date, we have become aware of 7 instances in which editors were subject to what can reasonably be classified as harassment or intimidation in order to achieve the retraction of inconvenient papers. The potentially chilling effects of those activities on academic freedom must be analyzed further.

The present findings add to the body of evidence showing that blog posts and comments sometimes have adverse effects on the rationality and civility of public discourse. For example, Anderson, Brossard, Scheufele, Xenos, and Ladwig (2014) showed that people became more polarized about an emerging technology (nanotechnology) when exposed to uncivil rather than civil online comments. Additional recent evidence has uncovered an association between online “trolling” behavior and three aspects of the “Dark Tetrad” of personality—sadism, psychopathy, and Machiavellianism (Buckels, Trapnell, & Paulhus, 2014)—with the link being strongest with sadism. It therefore appears that cyber-trolling may be the contemporary Internet manifestation of everyday sadism.

In response to those risks, at the time of this writing, several large online news services, such as Popular Science (http://www.popsci.com/; 2.8 million unique monthly visitors) and Reddit (http://www.reddit.com/; 4 million unique monthly visitors) have either banned comments altogether (Popular Science) or have selectively banned comments from climate “trolls” by insisting that arguments be supported by peer-reviewed science (Reddit), citing the harmful effects of comments that are emotive and ad hominem.

Our results therefore contribute to a growing understanding of online behavior and its implications for the conduct of science and public discourse. We offer two tentative pointers towards possible solutions. First, it is increasingly clear that there is a need to provide better architectures for online platforms to help channel public discourse into a more constructive direction. The requirement to back up claims by evidence (as on Reddit) or to consider factual status in search engines (as recently piloted by Google) point in that direction. Other alternatives include strict moderation of comments, as for example practiced by the online newspaper TheConversation.com, which employs a “Community Manager” and has entertained options such as a “community council” to provide moderation (https://theconversation.com/involving-a-community-council-in-moderation-25547).

Second, our results point to the need to educate the public about the difference between scientific and non-scientific forms of discourse. The Internet—as a platform for everyone to voice any opinion and make any claim, however unsupported by evidence—will not go away, and the positives associated with a “free for all” medium should not be under-estimated. However, we need to protect the evidence-bound sphere of scientific arguments from the largely unconstrained buzz outside that sphere. Peer review certainly has a role in defending that boundary: As far as we know, none of the individuals involved in the blogosphere’s response to LOG12 ever submitted a critique for peer review. Conversely, a peer-reviewed critique of LOG12 and LGO13 has recently appeared in print (Dixon & Jones, 2015) (accompanied by a rejoinder; Lewandowsky, Gignac, & Oberauer, 2015), which exhibited none of the features of conspiratorial ideation that we report in this article and which involved authors that were not part of the blogosphere examined here. Crucially, such academic discourse, however critical, does not involve the attempt to silence inconvenient voices, which has become an increasingly clearly stated goal of elements of the climate “skeptic” blogosphere.