Introduction

The world is ruled by a secret cabal. The US government was involved in the 9/11 attacks. HIV is a man-made virus. Such statements, which are promoted on the internet and elsewhere, are examples of conspiracy theories: implausible, unwarranted claims that important social events are caused by malevolent clandestine groups, that usually run in contradiction to the explanations offered by the relevant epistemic authorities, and that are embedded in a more general worldview [1].

Although researchers have tried to understand conspiracy theory beliefs–which to nonbelievers appear unwarranted by evidence, if not entirely irrational–from many perspectives, including personality theory [2, 3], intergroup identity [4–6], and cognitive bias [7–9], arguably the most frequent accounts rely on some notion of “control.” The claim that perceived lack of control is the natural breeding ground for conspiracy theory beliefs is a recurring theme in popular science (e.g. [10–14]), and many academics agree (e.g., [15–18]).

However, on closer inspection, neither the theoretical role of control in conspiratorial thinking, nor the empirical relationship between the two constructs, is entirely convincing. Although specific accounts differ in their emphasis, control accounts of conspiracy theories generally assume that the perceived loss of personal control is threatening to one’s need for order (e.g., [19]) or one’s schemas about the world (e.g., [20]). Conspiracy beliefs help compensate for these threats by ascribing control to external entities or organizations (e.g. [21–23]), or by creating new meaning frameworks via novel connections among unrelated events. For example, perceiving a connection between a tornado and the unusually long contrails observed the previous day, provides a basis for believing that the government (or some other entity) has deliberately sprayed chemicals in the sky to cause the tornado.

However, it seems equally plausible that conspiracy theories would have the opposite effect on the perception of control and order. A common theme among conspiracy theories is that reality is not what it seems; they are counter-normative narratives that suggest institutions and explanations that we take for granted are illusions–a revelation that, if taken seriously, seems bound to challenge the larger meaning-making systems individuals have come to rely on. Furthermore, although people desire order, they seek to optimize, not to maximize it, and it could be argued that conspiracy theories provide “too much” order, structure, and meaning by accounting for errant data, connecting seemingly unconnected events and leaving nothing unexplained [24, 25]. Finally, even if conspiracy theories represent an effective means of restoring order and structure, they are not the only means [19, 26–28], and, it would seem, a means of last resort. While control accounts do not require that sources of control or meaning be benevolent, people theoretically favour control systems that are both “culturally accessible” and “socially acceptable” [27, 29]. In most—if not all—conspiracy theories, however, the alleged conspirators are malevolent [30], thus making them poor candidates for compensatory processes when numerous effective and socially sanctioned alternatives exist.

The experimental evidence for conspiracies as compensatory sources of personal control is as ambivalent as its theoretical rationale. In a typical experiment, researchers ask participants to write about a time they did not have control, and to compare these participants’ conspiracy beliefs with a control group that is asked to describe an event when they felt in complete control (or about a neutral event). Some studies report stronger conspiracy beliefs in the low/neutral control group compared to the high control group (e.g. [31–33]), but others report weaker ones (e.g. [34]), and yet others (e.g. [35, 36]) report no effects of control on conspiracy beliefs at all.

One factor in the variability of results is the corresponding variability–and sometimes dubious validity–of how belief in conspiracy theories is operationalized and measured. For example, Whitson and Galinsky [33, Study 4] found that participants recalling uncontrollable situations were more likely to perceive patterns in ambiguous stimuli (e.g., images in visual noise) and links between events (e.g., an increase in e-mail correspondence and a lack of a promotion). However, while conspiracy theories are related to illusory pattern perception [37], they are not reducible to it. And while novel connections between seemingly unrelated events can, of course, be the building blocks of conspiracy theories, such theories tell a more complex story that taps into a broader worldview. For example, seeing a connection between the “umbrella man” (a man who opened and lifted an umbrella when Kennedy’s limousine approached) and the assassination of Kennedy is a starting point for building a conspiracy theory, but it is not a conspiracy theory per se (e.g., that Kennedy’s assignation was an organized Soviet-backed plot carried out for political purposes).

In another operationalization of conspiracy belief, Van Prooijen and Acker [32] asked participants about the construction of the North-South metro line in Amsterdam, by posing situations that were technically conspiracies, but also plausible abuses of power warranting legitimate skepticism. Accusations such as, “The city council transferred parts of the budget to the bank accounts of others”, and “Members of the city council received money from construction companies to set this plan in motion,” describe political corruption, which differ from “conspiracy theories” in scope, plausibility, official endorsement, and psychological interest. Indeed, recent evidence indicates that belief in conspiracy theories and belief in corruption are two distinct psychological constructs, predicted by different aspects of a larger construct (“conspiracy mentality”; [1]).

Even studies that explicitly ask about beliefs in specific conspiracy theories may not be well suited for assessing the construct of conspiracy theory belief. For one, participants can easily recognize such items as “conspiracy theories,” whose negative connotations are likely to produce socially desirable responses [1]. Moreover, using specific conspiracy beliefs limits the generalizability of the findings to very particular and culturally specific claims, and may say little about a person’s general tendency to employ conspiratorial logic. More relevant, we argue, is an individual’s tendency to engage in conspiracy thinking, rather than their belief in any particular claim.

Given the significance and potential consequences of widespread conspiracy beliefs [38, 39], and the plausible but largely unsubstantiated role of control in their appeal, we here report three studies to test the effects of lack of control on conspiracy theory beliefs using a standard priming paradigm and a validated measure of conspiracy ideation, which reflects the belief that a powerful entity lies behind significant social or political events and that the conventional (official) truth is not the “real” truth. A fourth study confirms the results with a measure of specific claims, and two final studies examine two of the most likely moderators of control effects: the degree of control that a particular conspiracy claim affords; and the effectiveness of alternative means of restoring control. Although we conceptually replicate previous correlational work–lower control beliefs were associated with stronger conspiracy ideation–we find no evidence for a causal effect.

Study 1

Study 2

Study 3

Study 4

Study 5

If people turn to conspiracy theories as a compensatory control mechanism, it may be that some theories are more effective at restoring perceptions of order than others. In particular, threats to control may only prompt belief in conspiracies involving a controlling entity (e.g., the theory that the U.S.S.R. started the “hippie” movement in the United States in order to undermine and weaken America’s traditional values), because the belief that someone is in control is sufficient to satisfy the higher order need for structure and meaning. Other theories in which a controlling entity is absent or only implicit (e.g., the theory that Osama bin Laden died prior to the 9/11 terrorist attacks but was used as a scapegoat) may not restore control, or do so to a lesser extent. Consistent with this proposal, Kay et al. [27] found that participants whose control was threatened increased their belief in God, but only when God was described as controlling, rather than as a creator. Thus, in Study 5, we developed a new stimulus set to test the hypothesis that participants whose control was threatened would be inclined to vest more belief in specific conspiracy theories about controlling entities, but not in other conspiracy theories.

Method

Participants.

The participants were three hundred MTurk workers (108 males, 192 females, M = 37.52, SD = 12.001). The majority (44%) had an undergraduate degree, while 29% had a high school degree, 13% had a master’s degree, 2% a doctoral degree and 1% did not have a degree. The mean age was 37.54 years (range 18 to 73 years). None had taken part in our previous studies.

Materials.

The dependent measure was developed in several stages. First, an independent group of MTurk participants (N = 210) were asked to write five conspiracy theories that they had heard before, from which we identified 112 unique exemplars. Next, two research assistants were asked to code these theories on a 9-point scale anchored at 1 (“the goal of this conspiracy is NOT control”) and 9 (“the goal of this conspiracy is control”). The intraclass correlation coefficient was r = 0.84. The final rating comprised the average of the ratings of the two research assistants. Next, a second group of 85 MTurk participants was asked to rate the conspiracy theories in terms of their prototypicality, using a seven point scale anchored at 1 (“a bad fit” with their idea of a conspiracy theory) and 9 (an “excellent fit” with their idea of a conspiracy theory). A third group of 96 MTurk participants was asked to rate the same conspiracies for likely veracity, using a nine point scale anchored at 1 (“definitely false”) and 9 (”definitely true”). The intraclass correlation coefficient for prototypicality ratings was r = 0.89 and for the veracity r = 0.96. The prototypicality and veracity scores for each conspiracy theory were obtained by averaging ratings, respectively across all participants.

These extensive pretest data were used to create seven pairs of conspiracy theories that were equated on veracity (t (95) = 0.604, p = 0.55, d = 0.06) and prototypicality (t (84) = 0.439, p = 0.66, d = 0.05), but that each differed by at least 5 scale points in terms of control (Ms = 8.21 vs. 1.71). For the full list of the selected conspiracies, along with their control, prototypicality and veracity rating (obtained from the pre-test) see Table 1, for the full list of all pretest conspiracies see the online supplementary materials.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Conspiracy theory pairs.

https://doi.org/10.1371/journal.pone.0237771.t001

Procedure.

The procedure was identical to that of Study 3, except for the use of two types of specific conspiracy theories, varying in the control they imply, and the addition of a high control group as additional comparison group. Participants rated their agreement with the 14 conspiracy statements, presented in a random order, on a 7-point scale (1 = strongly disagree, 7 = strongly agree). Mixed among these statements was an attention check question (i.e., “As a response to this question select strongly agree.”).

Study 6

Another plausible moderating factor determining the value of conspiracy theories for restoring threatened control, is the availability of alternative means of control. As noted above, while compensatory control processes do not require that sources of control be benevolent, people unsurprisingly favour sources of control that are not themselves threatening in other ways. Furthermore, there is a plethora of alternative control and meaning-making systems that are proven substitutes for personal control, such as God [52], government institutions [48], scientific [53], and even more abstract beliefs such as in meritocracy [54] and societal progress [55, 56]. Even if these alternatives are not all equally effective at establishing a sense of control, conspiracy theories would seem to be an option of last resort, given the essentially malevolent worldview they imply, and the social stigma that can accompany their endorsement [57]. Consequently, conspiracy theorizing may be a secondary strategy for restoring control, avoided when one has other means of affirming that the world is ordered [58].

Thus, in Study 6, we tested the hypothesis that threats to control would increase belief in conspiracy theories to a greater degree (or only) when the effectiveness of a second salient source of control–the government [29]–was called into question.

Results and discussion

One hundred and four participants were excluded from the main analysis: 74 due to failures of attention and 30 due to failure to following instructions. This left a total of 501 participants. A sensitivity analysis (α = 0.05, 1 - β = 0.8) indicated that the study is able to detect a minimum effect size of f = 0.12.

A 2 (control priming: low vs. neutral) x 2 (government competence: competent versus incompetent) ANOVA on conspiracy theory ideation revealed no main effect of priming, F (1, 497) = 0.161, p = 0.689, f = 0.02 or of competence, F (1, 497) = 0.243, p = 0.623, f = 0.02. Importantly, there was no interaction F (1, 497) = 1.301, p = 0.255, f = 0.05, (see Table 2 for descriptive statistics). Thus, the study, again, provides no evidence that control priming influences conspiracy beliefs, regardless of whether an alternative source of control has been compromised.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Descriptive statistics by recall task and scenario type.

https://doi.org/10.1371/journal.pone.0237771.t002

Meta-analyses

Although Studies 1–6 varied somewhat in methodology, their consistent inclusion of low control and neutral conditions permitted a meta-analysis, which maximizes both power and accuracy of our effect size estimates. Because in Study 2 the low control manipulation did not lower perceived control compared to the neutral group we excluded Study 2 from the analysis. For Studies 5 (low control compared to neutral) and 6 we looked at the effect at each level of the moderator. Because the effect sizes in Studies 5 and 6 were not independent we used the robumeta [61] package in R, which takes the dependency into account, to conduct the meta-analysis. A test of heterogeneity suggested consistency between the studies, I² = 11. The cumulative effect size did not suggest effect of condition, d = -0.03, 95% CI [-0.20, 0.13], p = 0.60 (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1.

Forest plot of analysed effect sizes from Studies 1and 3–6 (squares) and the cumulative effect size (diamond).

https://doi.org/10.1371/journal.pone.0237771.g001

To investigate support for the null hypothesis relative to the alternative hypothesis that control priming influences conspiracy theory beliefs, we used a Bayesian information criterion (BIC), using the calculator by Masson [62], based on Wagenmakers [63]. The analysis requires transformation of the sum of squares generated by an ANOVA output and provides graded level of evidence for the null and alternative hypothesis. According to Raftery [64], the strength of evidence for the posterior probability of a hypothesis given the data (p_BIC(H₁|D) within the range of 0.50–0.75 is “weak”, between 0.75–0.95 is “positive”, between 0.95–0.99 “strong” and above 0.99 “very strong”.

For those studies containing three groups we analysed the low and neutral (Study 1 and 5) or low and high (Study 2). For studies 5 and 6 we looked at each level of the moderator.

Table 3. shows the posterior probability of the null hypothesis given the data, p_BIC(H₀|D), and the probability of the alternative hypothesis given the data, p_BIC(H₁|D). There is positive evidence for the null hypothesis in five out of the six studies. In Study 1, there is only weak evidence for the null hypothesis, but the support for the alternative hypothesis is even lower (unsurprisingly, given that the effect is in the “wrong” direction).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Overview of study design and posterior probabilities for H₀ and H₁.

https://doi.org/10.1371/journal.pone.0237771.t003

General discussion

Six studies, conducted online and using MTurk participants, tested the hypothesis that lack of control motivates belief in conspiracies. While the hypothesis is plausible, there was in fact little support for it in the (small) literature on the subject. The current findings offer no further evidence, and some positive evidence for the null hypothesis, at least when the most common experimental manipulation of control is used with online samples. Participants who were asked to recall instances in which they felt out of control were no more likely to endorse either general conspiracy beliefs, or specific conspiracy theories, relative to participants who recalled neutral memories, or nothing at all. Plausible moderating variables–the extent to which conspiracies provide evidence of control (Study 5), and the effectiveness of an alternative means of control (Study 6)–had no effect, and a meta-analysis of the studies determined a cumulative effect size not statistically different from zero. On the contrary, a Bayesian analysis indicated that the null hypothesis (i.e., that there is no causal relationship between lack of control and conspiracy theory beliefs) should be retained.

How to explain the inconsistency between previous positive results and our null findings? One explanation, of course, is that previous “positive” findings were simply Type 1 errors, which is not farfetched given social psychology’s historical inattention to power [65, 66] combined with publication bias. Indeed, assuming the “actual” effect size of control threat on conspiracy beliefs is 0.2 (the average effect size reported in psychology research [67]), then previous studies reporting positive results may have included chance findings given their low power. For example, assuming the actual effect is 0.2, then Whitson and Galinsky’s [31] Study 1 has power of 0.17, Sullivan et al.’s [33] has power of 0.18, and van Prooijen and Acker’s [32] power of 0.22. By contrast, studies finding nonsignificant findings, such as those by Hart and Graether’s [35] or Nyhan & Zeitzoff [36], have higher power, 0.52 and 0.88, respectively. Another possibility, however, is that some previous studies measured constructs other than “pure” conspiracy ideation (e.g., pattern perception, paranoid beliefs, corruption beliefs), which are susceptible to control threats. Indeed, studies that examine more typical conspiracy theories and have larger samples tend to report null results. For example, Hart and Graether [35], found no effects of control on the Generic Conspiracist Beliefs Scale (another psychometrically validated measure of conspiracy beliefs, with items such as, “Experiments involving new drugs or technologies are routinely carried out on the public without their knowledge or consent”). Similarly, Nyhan and Zeitzoff [36] found no effect of control priming on Middle Eastern and North African participants’ conspiracy beliefs, using a list of conspiracy theories consisting of items such as “Jewish leaders are secretly conspiring to achieve world domination.”

It is also worth repeating that the current results were obtained entirely with online samples, while most studies in this area (including the majority of those with positive effects of control) have been conducted in the laboratory (typically with student samples). In principle, online manipulations of control could be weaker than those in the laboratory, accounting for the lack of support for the control hypothesis. As an empirical matter, however, there is no evidence that effects in this paradigm depend on sample type [68]. In a recent meta-analysis of experimental manipulations of control on conspiracy beliefs [68] conducted on 45 effect sizes across 23 studies (including those reported here), we found no moderating effect of sample type (MTurk vs. students vs. other). Thus, although it is important to continue to examine the role of control threats in diverse samples and contexts, the current data, despite being collected online, nevertheless challenge the hypothesis that such threats account for conspiracy beliefs in any significant way.

Although there was no evidence in the current studies for a causal effect of control on conspiracy beliefs, Studies 1 and 2 provided evidence for a correlational relationship. Such correlational findings are consistent with previous research [32, 34, 49, 69, 70], suggesting that chronic, rather than acute, feelings of low control are related to higher conspiracy theory beliefs. While longitudinal studies may help shed light on the effects of chronic lack of control on conspiracy beliefs, it is also possible that the link is spurious. There are any number of co-variates of both lack of control and conspiracy theory belief, including stress [71, 72], self-esteem [73, 74], anxiety [75, 76] or uncertainty [77], that could account for their association. Also, the correlation may be due to conspiracy beliefs’ shared variance with paranoia (see for example Imhoff & Lamberty, [78]). Alternatively, the causality may run in the opposite direction: conspiracy theories themselves may reduce feelings of control. Jolley and Douglas [79], for example, found that participants exposed to conspiracy theories about vaccines reported greater feelings of powerlessness compared to participants exposed to anti-conspiracy material, suggesting that conspiracy beliefs threaten rather than satiate one’s sense of control. Furthermore, it is not clear what type of correlational relationship should be taken as evidence for satiation; if conspiracy theories compensate for compromised personal control, should the relationship between control and conspiracy beliefs in cross-sectional studies be positive or negative? Thus, correlational evidence in the absence of experimental evidence is not a strong foundation for claiming that lack of control is a precursor to conspiracy beliefs.

Even if our present null results are confirmed in subsequent research, it is important to acknowledge their limitations. Priming is only one method of threatening (or bolstering) a sense of personal control, and while we found no evidence for changes in conspiracy beliefs, more potent or qualitatively distinct experimental manipulations could conceivably produce effects that we were unable to detect here. Furthermore, we could not assess the effects of chronic lack of control: It may be that prolonged periods of control deprivation do indeed produce reliance on other, even malevolent sources of control, including conspiracy theories.

Another unexplored area is whether lack of control in a specific domain will lead to belief in conspiracy theory in that domain. For example, people diagnosed with HIV or COVID-19 might be particularly inclined to believe in conspiracy theories involving those viruses, rather than conspiracy theories in general. Indeed, in a recent naturalistic study [80], we found a domain-specific link between threat to control and conspiracy beliefs. For example, after a series of tornados, those who experienced such natural disasters were more likely to endorse weather-related conspiracy beliefs than they were other types of conspiracy beliefs. Moreover, the strength of these endorsements subsided over time (i.e., three months after tornado season). Such findings seem to suggest that the link between perceived control and conspiracy beliefs is both nuanced and transitory, and that there is no “one size fits all” explanation for lack of control and conspiracy-theory endorsement in general.

Finally, we relied exclusively on laboratory manipulations of control (or lack thereof). While such empirical techniques are widely used in the literature, the effectiveness of the “recall task” in threatening participants’ sense of control has been questioned [34], and its effects were weak in our studies. “Real life” events, such as natural disasters, terrorist attacks, and pandemics, may produce stronger threats to personal control (and indeed, evidence suggests that traumatic events are followed by an increase in conspiracy discourse [81]). Findings from our own naturalistic study seems to confirm this. However, such naturalistic “manipulations” may confound perceptions of control with other variables, such as threats to identity, uncertainty about the future, and stress. Thus, any naturalistic observations need to be accompanied by rigorous experimental research to triangulate the mechanism(s) responsible for any changes in conspiracy beliefs. On their own, both empirical approaches are problematic, in that experimental manipulations of control may be too “weak” to evoke conspiracy-belief endorsements, and perceived loss of control under naturally occurring conditions are confounded with other variables.

Supporting information