Conspiracy theories—the belief that covert, powerful forces are responsible for unexplained phenomena—are a widespread feature of politics, in part because they often focus on the government, especially in the U.S. (e.g., Hofstadter Reference Hofstadter2008; Uscinski et al. N.d.). These theories, which typically (though not always) lack strong evidentiary support, can have a range of pernicious consequences such as undermining trust in political institutions (e.g., Einstein and Glick Reference Einstein and Glick2015) and decreasing political efficacy and participation (e.g., Jolley and Douglas Reference Jolley and Douglas2014b).
The most prominent conspiracy theories often seek to explain unexpected events. These beliefs frequently enjoy widespread acceptance among the public and tend not to dissipate even after official investigations are conducted (Bowman and Rugg Reference Bowman and Rugg2013; Oliver and Wood Reference Oliver and Wood2014). For instance, many Americans reject the conclusions of the Warren Commission about the cause of John F. Kennedy’s assassination and the conclusions of the 9/11 Commission about the causes of the September 11 terrorist attacks (Bowman and Rugg Reference Bowman and Rugg2013; Stempel et al. Reference Stempel, Hargrove and Stempel2007).
Though previous research has identified several factors that may make people more likely to perceive conspiracies (e.g., Oliver and Wood Reference Oliver and Wood2014; Uscinski and Atkinson N.d.; Whitson and Galinsky Reference Whitson and Galinsky2008), little is known about how to reduce belief in conspiracy theories about events that are not supported by convincing evidence. Many of these efforts may be ineffective or even counterproductive (e.g., Nyhan and Reifler Reference Nyhan and Reifler2010, Reference Nyhan and Reifler2012).
In particular, while government disclosure of information may be intended to reduce misperceptions about the events that inspire conspiracy theories, lay epistemic theory (e.g., Kruglanski Reference Kruglanski1990) suggests that common bureaucratic practices may undermine the effects of these efforts among citizens (Harrison and Thomas Reference Harrison and Thomas1997; Sunstein and Vermeule Reference Sunstein and Vermeule2009). Specifically, the presence of redactions—which are often heavily used in documents released by the government, especially in recent years (e.g., Bridis and Gillum Reference Bridis and Gillum2014; Kravets Reference Kravets2014; Bridis Reference Bridis2015)—may make readers more likely to interpret documents as evidence of a conspiracy or coverup and reduce or eliminate any conspiracy-reducing effect. For instance, the 9/11 commission report was intended to reduce misperceptions about the terrorist attacks, but the redaction of 28 pages pertaining to alleged ties between the Saudi government and the hijackers continues to fuel conspiracy theories (Clift Reference Clift2015; Dilanian Reference Dilanian2015).Footnote 1 Redactions have also been prominently featured in recent debates over conspiracy theories about the Sandy Hook massacre (Altimari Reference Altimari2014), the disappearance of flight MH370 (O’Neill Reference O’Neill2014), the crash (apparently due to a surface-to-air missile) of flight MH17 (Associated Press 2015), and the Kennedy assassination (Shenon Reference Shenon2015).
We take a novel approach to the topic of conspiracy beliefs. To our knowledge, this study is the first to test how people react to corrective information about a conspiracy theory depending on the format in which it is providedFootnote 2 and to analyze how those reactions vary depending on people’s predispositions toward conspiracy theories. It is also the first to evaluate the effect of redactions on belief in conspiracy theories, which we test using a new design in which we vary whether or not black boxes are inserted between words and sentences in a series of documents. These boxes appeared to be genuine redactions but did not actually obscure any text, allowing us to hold the information given to respondents constant.
We test two hypotheses. First, we predicted that individuals given seemingly redacted documents would be more likely to believe in a conspiracy theory than those given otherwise identical documents in which the redactions are omitted. As described below, we expect that respondents will infer from the redactions that the government must have something to hide and will therefore be more willing to question the official account and to endorse a conspiracy theory instead. We also predicted that the difference in conspiracy beliefs between the redacted and unredacted conditions would be greater among individuals with high conspiracy predispositions than those with low predispositions. Finally, we estimated how exposure to redacted or unredacted documents changed beliefs relative to controls—a research question of interest.Footnote 3
Consistent with our first hypothesis, participants in an initial study and a replication who read seemingly redacted documents were more likely to believe in a conspiracy than those who read otherwise identical unredacted documents. We did not find support for our second hypothesis—the effects of exposure to redactions did not differ by predispositions toward conspiracy belief in either study. Finally, participants who read unredacted documents had significantly lower conspiracy theory beliefs than controls, but this effect was diminished if redactions were included. The presence of redactions thus appears to undermine the information effect observed in the unredacted condition, preventing evidence in the documents from reducing conspiracy beliefs as effectively. As we show below, this result does not appear to be attributable to a lack of respondent attention or engagement or the absence of a specific rationale for the redactions.
THEORY
We proposed two hypotheses that were preregistered before data from our first study had been fully collected or analyzed.Footnote 4 Our first hypothesis predicted that participants assigned to read redacted documents would believe in conspiracy theories more than those assigned to read otherwise identical unredacted documents (H1).Footnote 5 It has been proposed that selective governmental disclosures such as redacted documents may be construed as “deliberate attempts to suppress information and mislead the public” (Harrison and Thomas Reference Harrison and Thomas1997, 120, 123). A plausible mechanism for this response comes from previous research on lay epistemics, which suggests that people try to explain what they observe by generating and evaluating subjective “if-then” hypotheses about the causes of events (Kruglanski Reference Kruglanski1989, Reference Kruglanski1990; Kruglanski et al. Reference Kruglanski, Dechesne, Orehek and Pierro2009). Members of the public who observe a redacted document may thus make the inference “If a document is redacted, then the government must have something to hide,” which would cause them to attribute redactions to the presence of a coverup or conspiracy. Conspiracy perceptions should thus be higher in the redacted documents condition than in the unredacted condition even if the information in the documents is otherwise identical.
We also expected that participants with high predispositions toward conspiracy belief would be especially likely to interpret redactions as evidence of a potential government coverup or secret plot (Oliver and Wood Reference Oliver and Wood2014; see also, e.g., Goertzel Reference Goertzel1994 and Swami et al. Reference Swami, Coles, Stieger, Pietschnig, Furnham, Rehim and Voracek2011). Redactions are especially consistent with the epistemology of conspiracy theories—which often attribute observed behavior to hidden patterns of wrongdoing—and are thus likely to be perceived as suspicious by these individuals (e.g., Barkun Reference Barkun2013). Our second hypothesis therefore predicted a greater difference in conspiracy beliefs between the redacted and unredacted document conditions among respondents with high conspiracy predispositions than among those with low conspiracy predispositions (H2).
However, we did not propose a hypothesis about the effect of exposure to redacted or unredacted government documents relative to the control group. The evidence in the unredacted documents might either reduce misperceptions or increase them relative to controls—previous research has found differing effects of corrective information exposure (see, e.g., Nyhan and Reifler Reference Nyhan and Reifler2010, Reference Nyhan and Reifler2012; Nyhan et al. Reference Nyhan, Reifler and Ubel2013). The situation is even more uncertain for respondents in the redacted documents condition, who differ in two ways from controls—they see the evidence presented in the documents (which is identical to the unredacted condition) but also see redactions that might seem to suggest the presence of a coverup or conspiracy. The comparison between the redacted condition and controls thus does not identify the causal effect of the redactions alone (the focus of our first hypothesis) but instead the joint effect of exposure to corrective information and redactions. As a result, the effect of exposure to the documents relative to controls was instead designated as a research question of interest.
Finally, it is important to clarify how these hypotheses are tested. The most appropriate test of the redaction effect holds respondent information constant. We isolate this effect by comparing conspiracy beliefs among respondents given the same information in the redacted and unredacted conditions. Likewise, we isolate the effect of information by comparing conspiracy beliefs between the unredacted documents condition and controls. Third, we estimate the net effect of redacted documents relative to controls. However, we emphasize that the comparison between the redacted and control conditions estimates the joint effect of two treatments: the information in the documents and the redactions. As we show, the response generated by redactions can reduce or eliminate the conspiracy-reducing effect of information.
STUDY 1
Subject Matter
We examine beliefs about the crash of TWA Flight 800, which exploded soon after takeoff from Kennedy International Airport on July 17, 1996, killing all 230 people on board. While official accounts concluded that the accident resulted from the ignition of a flammable fuel/air mixture in the fuel tank (National Transportation Safety Board 2000), conspiracy theorists claim it was the result of an accidental U.S. Navy missile strike that is being covered up (e.g., Purdy Reference Purdy1997). This claim, which grew out of testimony by eyewitnesses who claimed to have seen streaks of light before the crash, has fueled a persistent conspiracy narrative that was featured in a recent documentary (Genzlinger Reference Genzlinger2013). As with many such beliefs, these theories seek to explain a shocking or unexpected event as a result of secret actions based on seeming inconsistencies between the official explanation and various details and eyewitness accounts.
We chose to study beliefs about Flight 800 for several reasons. First, conspiracy theories about its explosion are generally non-partisan. As a result, treatment effects are less likely to differ between political groups than other prominent conspiracy theories (see, e.g., Oliver and Wood Reference Oliver and Wood2014). In addition, we wanted a topic that is old enough for a settled conspiracy theory to be established (unlike the disappearance of Malaysia Airlines Flight 370) without being antiquated (such as the J.F.K. assassination). Third, Flight 800 conspiracies are widely known and seemingly plausible (e.g., Bowman and Rugg Reference Bowman and Rugg2013) but not so famous that respondents have relatively fixed beliefs (as they might on, say, Kennedy). Finally, the topic was relevant to Flight 370 conspiracy theories that were circulating when our data was collected (e.g., Frizell Reference Frizell2014; Sanchez Reference Sanchez2014; The Week 2014).
Participants, Design, and Procedure
Participants from Amazon Mechanical Turk, a crowdsourcing website, completed the study on the Qualtrics online survey platform from April 30–May 7, 2014.Footnote 6 Of the 2,524 participants (the maximum allowed by the study budget), 48% were male; 80% were white; the median age group was 30–39; and half had at least a bachelor’s degree. Politically, 41% identified as Democrats, 18% as Republicans, and 42% as independents or something else. (See Table A1 in the appendix for more information on respondent demographics and Table A5 for further details on the procedures used in both studies.)
After providing informed consent, participants completed a series of demographic and attitudinal questions, including evaluating two statements that have been shown to correlate with conspiracy beliefs (Oliver and Wood Reference Oliver and Wood2014): “Much of what happens in the world today is decided by a small and secretive group of individuals” and “Politics is ultimately a struggle between good and evil” (six-point scale: strongly disagree=1, strongly agree=6). Participants with an average answer above three were categorized as having high conspiracy predispositions in a median split; others were categorized as having low predispositions.Footnote 7 After finishing these questions, participants completed a word search task to clear working memory.
All participants were then instructed to read a paragraph describing both the official story of the TWA Flight 800 explosion and the conspiracy theory in a balanced manner, which is a common practice in news coverage of factual disputes (e.g., Cunningham Reference Cunningham2003; Fritz et al. Reference Fritz, Keefer and Nyhan2004). The article was accompanied by a picture of the reconstructed plane to make the experiment more vivid (see appendix for text and image).
Both treatment groups were then asked to read three government documents that provided evidence supporting the official account of the crash:
-
• A transcript of radio correspondence at the time of the crash (Tauss N.d.);
-
• A summary of radar evidence from the official report (National Transportation Safety Board 2000);
-
• The report’s conclusion that the crash was the result of an explosion in the fuel tank (National Transportation Safety Board 2000).
The documents given to the two redacted and unredacted treatment groups included the same visible text, but the redacted documents were manipulated by adding black boxes over blank space, creating the appearance that information was being withheld (see appendix). Controls were instead asked to read three recipes, a realistic real world task (reading the news and then a cookbook) which ensured that respondents in all conditions were exposed to documents of approximately equal length and density before the outcome measures.
After the experimental manipulation, we asked respondents to evaluate the likelihood of a series of statements about the causes of the incident and the validity of the government investigation on a six-point scale:
-
• A mechanical failure caused the explosion of TWA Flight 800.
-
• The U.S. government was involved in the explosion of TWA Flight 800.
-
• TWA Flight 800 was shot down by a missile fired by the U.S. military.
-
• The government thoroughly investigated the crash of Flight 800 and determined its true cause.
-
• The government is covering up the true cause of the explosion of TWA Flight 800 from the public.
The direction of these items, which serve as our dependent variables and were based in part on past polling (Bowman and Rugg Reference Bowman and Rugg2013), varied between the official explanation of the explosion and conspiracy theories about its causes. Answers were coded so that higher values represented greater conspiracy belief. We also created a composite belief measure using the mean response after recoding.Footnote 8
Results
Redaction effects
To assess the results of the experiment, we first compared the average conspiracy belief measure between redacted and unredacted conditions. As noted above, this comparison holds the information provided to respondents constant and is therefore the clearest test of the effect of redactions on conspiracy beliefs. Confirming our hypothesis, respondents exposed to redacted documents (mean=2.52, 95% CI: 2.43–2.61) reported stronger conspiracy beliefs than those who saw unredacted documents (mean=2.32, 95% CI: 2.24–2.41; t = 3.16, p < 0.01)—an increase of 0.15 standard deviations on the average belief measure.Footnote 9 We also estimated the effect of exposure to corrective information by comparing respondents who saw unredacted documents with the control group. Results indicated that providing corrective information in unredacted form decreased average conspiracy beliefs relative to controls (mean=2.60, 95% CI: 2.51–2.68; t = −4.47, p < 0.01). However, the joint effect of the redactions (which led to higher conspiracy beliefs than among people who saw otherwise identical unredacted documents) and the information in the documents (which led to lower levels of belief among people who saw unredacted documents versus controls) was null—redacted documents did not have a significant effect versus controls (t = −1.22, p < 0.23), suggesting that redactions offset or undermined the effects of the corrective information. (We discuss our interpretation of this effect below.)
Table 1 evaluates these findings more systematically by providing regression results for each of the dependent variables and the composite belief measure. The coefficient estimates for the redacted and unredacted document conditions represent effects relative to the control group, which is the omitted category in the regression. The direct effect of the redaction among respondents exposed to the government documents is computed as the difference between the redacted and unredacted coefficient estimates and presented separately in the table. As predicted, participants receiving redacted documents reported significantly higher levels of conspiracy beliefs than those receiving unredacted documents for all dependent variables (p < 0.05 except for beliefs that the government investigation determined the true cause of the crash, which was p < 0.052). In addition, the redacted treatment only reduced conspiracy beliefs relative to the control condition for one dependent variable at the p < 0.05 level (belief that TWA Flight 800 was shot down by missiles). In contrast, almost all dependent variables recorded significant differences between the unredacted and control treatments at the p < 0.05 level except for beliefs that the government investigation determined the true cause of the crash (p < 0.06). Although the conditions displayed identical text, reading the redacted documents had little to no effect versus controls, whereas reading the unredacted documents decreased conspiracy belief significantly. Again, redactions appear to have offset or undermined the conspiracy-reducing effects of the information in the documents that are observed when we compare the unredacted condition with controls directly.Footnote 10
Table 1 Redaction Effects on TWA Flight 800 Conspiracy Beliefs
*p < 0.10, †p < 0.05, ‡p < 0.01. OLS estimates with robust standard errors.
To illustrate the magnitude of these effects using a more intuitive outcome measure, Figure 1 presents differences in a binary indicator of conspiracy belief between conditions. We identify as conspiracy adherents those respondents who were above the midpoint of our six-point scale of average conspiracy beliefs, indicating that on average they thought that the claims that the government was involved in the crash, the flight was shot down by a missile, and the government is covering up the true cause of the crash were more likely than unlikely and were similarly doubtful about the official explanation. We find that the proportion of conspiracy adherents was six percentage points higher in the redaction condition (24%, 95% CI: 21–27%) than the unredacted condition (18%, 95% CI: 16–21%)—a relative increase in prevalence of 31% (p < 0.01). Conspiracy adherence was significantly less common among respondents in the unredacted condition than the control group (26%, 95% CI: 23–29%; p < 0.01), but the redacted condition was again not significantly different from controls. These results suggest that the presence of redactions has meaningful effects on the prevalence of conspiracy beliefs and is not limited to small effects on levels of disbelief among skeptics.
Figure 1 TWA Flight 800 Conspiracy Adherence by Condition
The null effect of the redacted condition relative to controls does not appear to be the result of respondents dismissing or ignoring the stimulus.Footnote 11 As we demonstrate in Table A2 in the appendix, respondents spent almost exactly as long reading the redacted documents (m = 285 seconds) as the unredacted documents (m = 287 seconds; t = −0.23, p < 0.82) and the average response times for the outcome variables were almost identical (m = 7.47 seconds for redacted versus 7.36 seconds for unredacted, t = 0.56, p < 0.58).Footnote 12 In addition, respondents in the redacted condition were more likely to mention the content of the study in a general open text question asking if they had any comments on the survey than those in the unredacted condition—3.2% of those in the redacted condition included the words “TWA,” “800,” “plane,” “crash,” or “flight” compared with only 1.4% in the unredacted condition (t = 2.48, p < 0.05). The evidence we observe is thus inconsistent with the interpretation that redacted documents had no effect on conspiracy beliefs relative to controls due to a lack of respondent attention or engagement. The effects of the redactions seem instead to have offset or undermined the reduction in conspiracy beliefs observed in the unredacted condition. (This issue is discussed further in the conclusion.)
Differences by conspiracy predispositions
Table 2 presents OLS results for the redacted and unredacted conditions and their interaction with the high conspiracy predisposition indicator for each dependent variable.Footnote 13
Table 2 Redaction Effects on Conspiracy Belief by Respondent Predispositions
*p < 0.10, †p < 0.05, ‡p < 0.01. OLS estimates with robust standard errors.
Again, the coefficients for the redacted and unredacted conditions and their associated interaction terms are estimated relative to controls; the key term for testing the second hypothesis is the auxiliary quantity reported below the main coefficient estimates, which represents the difference in the redaction effect (relative to the unredacted condition) between the low and high conspiracy predisposition groups.Footnote 14
Contrary to our hypothesis, the redaction effect did not consistently differ between groups (except for one marginally significant effect in the opposite direction from expectations). Instead, we found that individuals who are predisposed to believe conspiracy theories were more likely to believe in a Flight 800 conspiracy regardless of the available information (the difference between redacted and unredacted was not significant), whereas respondents who lacked these predispositions had higher conspiracy beliefs in the redacted condition than in the unredacted condition (p < 0.01 for average beliefs).
STUDY 2
One possible concern about Study 1 is that no reason was provided for the presence of redactions, which might make respondents more suspicious and inclined to believe conspiracy theories. We therefore conducted a second study to verify that our findings were robust to the inclusion of a substantively plausible rationale for withholding information (protecting aviation safety and national security). As we show in the appendix, which describes the design and results in more detail, conspiracy beliefs were again higher in the redacted than in the unredacted condition despite the inclusion of a realistic rationale, though in this case both treatments reduced conspiracy beliefs relative to controls.Footnote 15
DISCUSSION
Confirming our first hypothesis, people who read redacted documents about the TWA Flight 800 accident were more likely to believe conspiracy theories than those who read otherwise identical unredacted documents in two studies. In addition, while participants who read unredacted documents were significantly less likely to believe in the conspiracy theory than controls, redacted documents (which represent the joint effect of redactions and information) reduced or eliminated the effect of exposure to the information in the documents relative to controls—a result that does not seem to be attributable to a lack of respondent attention or engagement. These findings confirm the expectation from lay epistemic theory that redactions are often seen as evidence that government has something to hide and can thereby contribute to conspiracy beliefs. However, the effect of redactions on conspiracy beliefs did not differ based on people’s conspiracy predispositions, contradicting our second hypothesis.
Our study suggests several directions for future research. First, our design used artificial redactions that did not withhold any information. We believe this approach best isolates the effect of redactions alone and is likely to be a lower bound of real world effects. However, future research should evaluate the external validity of our findings by testing if the results strengthen (as we expect) when redactions obscure text that is present in an unredacted condition, which is the situation observed in practice. Second, scholars might wish to examine the effect of redactions in media accounts;Footnote 16 to test alternate conspiracy belief and/or predisposition measures (e.g., Darwin et al. Reference Darwin, Neave and Holmes2011; Swami et al. Reference Swami, Coles, Stieger, Pietschnig, Furnham, Rehim and Voracek2011; Lewandowsky et al. Reference Lewandowsky, Gignac and Oberauer2013); and to consider other possible moderators such as trust in government. Third, researchers should investigate whether these effects vary depending on the type of document or the frequency or position of redactions within it. Fourth, scholars might consider varying the rationale provided for redactions to test if those that seem less justified or proportional to the volume or importance of the information withheld are especially likely to increase conspiracy beliefs. Finally, though establishing the mechanism for a causal effect is very difficult (Bullock et al. Reference Bullock, Green and Ha2010), it would be worthwhile to further investigate the process by which people react to redactions, which could provide additional insight into why their effect relative to controls was reduced or eliminated.
Despite these limitations, our study makes a valuable contribution to both the study of conspiracy theories and the practice of government. Even the appearance of having something to hide can seemingly cause suspicions about government intentions and doubts in official accounts to grow. These findings suggest that governments should seek greater transparency when releasing documents to dispel conspiracy beliefs.
SUPPLEMENTARY MATERIALS
The appendix is available online as supplementary material at http://dx.doi.org/10.1017/XPS.2015.21
