In early January 2020, we launched a study of belief updating on Amazon Mechanical Turk. It was a follow-up to a study which showed that paranoia, the belief that powerful others are conspiring against us, was relatable to distinctly non-social uncertainty-driven learning mechanisms¹. We took this non-social task (choosing between decks of cards for points rewards) and reframed it socially (choosing between collaborators for a better grade). We sought to compare how performance on each version related to participants’ self-reported paranoia. If the social version was more sensitive, or more strongly related, we would favor theories of paranoia that lean on mechanisms of reputation-management and coalitional cognition², if not, we would conclude that more fundamental, domain-general processes are involved³.

As recruitment progressed through January and February, it became apparent that our world was changing drastically. We began hearing reports of a viral pneumonia in China, and then in Europe. Initially, these were seen as minor and isolated events, however, with time, it became clearer that things were very serious. The World Health Organization declared the situation a pandemic in March. We closed our in-person laboratory and headed into lockdown on Friday March 13^th 2020. At this point (and indeed throughout) we found no differences in the relationship of paranoia to the social or non-social framing of the task⁴.

Since our experiment was already open to recruitment, we kept it open, and continued to gather data, as the pandemic unfolded. This was an unprecedented opportunity – for a lab that focuses on uncertainty and beliefs. We were able to gather data as people’s uncertainty about their world waxed and waned. We could track the impact of that real-world uncertainty on their beliefs about our card deck and collaborator games. Since we were acquiring data across the US, we could compare and contrast the effects of different policy responses to the pandemic on peoples’ beliefs and task performance.

National Public Radio began referring to, “these uncertain times” in their sponsor’s commercials. Historical periods of uncertainty, from the black death⁵ to September 11^th 2001, are fertile grounds for conspiracy theorizing⁶. We tested for that effect empirically, and aimed to unpack the factors that might form and foment such an effect.

Without the usual in-person collegial interactions, we turned to podcasts for stimulation and inspiration. We learned about Michele Gelfand’s work on cultural tightness – the extent to which people in a given place value adherence to social norms⁷ – from Shankar Verdantem’s Hidden Brain podcast. We learned about the Berber people, whose reluctance to expect too much from their conspecifics was not an impediment, but instead was critical to a functioning society⁸ – from the Invisibilia podcast. Perhaps these ideas might help explain our data?

The New York Times published compelling and often troubling infographics – on people’s perceptions that others they encountered were wearing masks. These unconventional sources of ideas and data made invaluable contributions to our analyses of the data we were accruing.

One idea seemed particularly attractive; we could leverage an approach from econometrics to make causal claims about the impacts of different policies on people’s beliefs.

This Difference-in-Differences approach was pioneered by John Snow in the Mid-1800s to discover how residents of London were being infected with cholera⁹. He observed that the Southwark and Vauxhall Company and the Lambeth Company supplied water to high-death-rate districts in South London. Both companies used the Thames—which was highly contaminated—as their water source before 1852. In 1852, the Lambeth Company began to source water from the Thames Ditton river—which was upstream and uncontaminated. Snow was able to show that between 1849 and 1854, districts that remained supplied by the Southwark and Vauxhall Company saw an increase in deaths from cholera while those supplied by the Lambeth Company experienced a reduction in deaths from cholera.

In a fabulous perspective piece¹⁰, Ioana Marinescu, Patrick Lawlor and Konrad Kording describe how these methods might be applied to answer neuroscientific and behavioral questions using laboratory data, rather than large scale studies. Our pandemic approach was really an amalgam of these – adopting a task grounded in preclinical behavioral neuroscience to study belief updating in humans from diverse enough areas so as to be able to track the impact of the policies to which they were subjected, on their task behavior and self-rated paranoia.

The method is not uncontroversial. For example, care must be taken to apportion variance in responses appropriately. There can be confounds at baseline that might explain apparently causal effects – we learned that from a series of papers considering whether the exposure to the MTV show 16 and pregnant led to more teenage pregnancies (it did not)¹¹.

In our own data, and lives, there were events other than policy changes that could have contributed to the increases in paranoia that we observed. It was important for us to recognize the potential impact of the murder of George Floyd in May, which triggered protests in every state and around the world. We attempted to quantify these effects, as best we could with publicly available data.

We found that a vigorous lockdown (closing early, and extensively, and remaining closed) was associated with lower self-reported paranoia. However, at reopening, when paranoia peaked, it was particularly high in states that mandated mask wearing. Indeed, a difference in differences analysis comparing states with a mask mandate to states where masks were merely recommended, found that mask mandates caused paranoia to increase. By incorporating data from the New York Times study of mask-wearing perception (from Dynata), and combining them with Prof. Gelfand’s cultural tightness measure for each state, we were able to show that paranoia was highest in states that had a mandate, wherein people typically follow the rules, but for whom the perception was that people were not adhering to the mandate.

When paranoia increased, so did participants’ erratic behavior on our tasks (switching between card decks or collaborators, even after a win). We fit computational models to these choices and found that these effects were driven by an elevated belief in volatility (of the task). It appeared that paranoia increased as real-world volatility increased, and so did participants’ expectations of how volatile the task would be.

In follow up work from August to November, we assayed other relevant features; more paranoid people were more hesitant to wear masks, and harbored more conspiratorial beliefs about vaccines for COVID. Again, more erratic behavior was associated with greater expected volatility and paranoia. Furthermore, given that the pandemic and its handling became highly politicized, we measured participants beliefs in the QAnon conspiracy (the false belief that President Trump is waging war on the Deep State and pedophile Hollywood elites who are torturing children and harvesting their adrenal glands). QAnon beliefs also correlated with paranoia, as well as erratic task behavior and stronger prior beliefs about volatility.

Next – inspired by a wonderful question after a talk over zoom (thank you Klaas Stephan) - we extended our investigation of the non-social (card deck) task. After the Social task, we asked participants whether they believed any of the collaborators were sabotaging them. They often did, and those beliefs correlated with paranoia more broadly. In a follow up study, we asked those participants who completed the non-social task whether they believed that the card decks were sabotaging them. They did, and they strength of those beliefs also correlated with the severity of their self-rated paranoia. Taken together, our data suggest that paranoia – I deeply social concern, has distinctly non-social roots, but it is nevertheless strongly impacted by social norms and their perceived violation. Our work is consistent with the idea that humans are conditional cooperators – they will follow the rules as long as they don’t perceive too many people defecting, otherwise, they too will break the rules.

To be clear, we do not think that social processes are irrelevant to paranoia. Rather, non-social domain-general mechanisms can explain a lot too. Put simply, there may be value in having an enemy – someone on whim to blame our misfortune, and, in so doing, contingent events might be rendered more explicable and controllable¹². Of course, it may be that our social task was not social enough. We are actively exploring tasks with better face validity and ecological validity.

More broadly, we think our demonstration that real world volatility alters volatility beliefs in the (virtual) lab, may have implications for the replication crisis in psychology. To the extent that a particular phenomenon engages volatility driven belief updating, that phenomenon may not replicate if the original study and the replication attempt are conducted under different real-world volatility conditions.

We know it is often counterproductive to rush to policy prescriptions, but we humbly suggest that contextually appropriate enforcement of new norms, not necessarily with punishments, may keep paranoia at bay, and help people conform. In follow-up work, we are exploring the relationships between advice-taking and paranoia under ambiguity. There too, the initial findings suggest that even deciding whether or not to listen to another person may be anchored in an individual’s beliefs about themselves, rather than their advisor¹³.

In sum, this paper is the product of, and speaks to, our uncertain times. Under uncertainty, we can readily alienate others, precisely at the time when we need them the most.

1          Reed, E. J. et al. Paranoia as a deficit in non-social belief updating. Elife 9, doi:10.7554/eLife.56345 (2020).

2          Raihani, N. J. & Bell, V. An evolutionary perspective on paranoia. Nat Hum Behav 3, 114-121, doi:10.1038/s41562-018-0495-0 (2019).

3          Feeney, E. J., Groman, S. M., Taylor, J. R. & Corlett, P. R. Explaining Delusions: Reducing Uncertainty Through Basic and Computational Neuroscience. Schizophr Bull 43, 263-272, doi:10.1093/schbul/sbw194 (2017).

4          Suthaharan, P. et al. Paranoia and belief updating during the COVID-19 crisis. Nat Hum Behav, doi:10.1038/s41562-021-01176-8 (2021).

5          Cohn, N. The Pursuit of the Millenium.  (Oxford University Press, 1961).

6          van Prooijen, J. W. & Douglas, K. M. Conspiracy theories as part of history: The role of societal crisis situations. Mem Stud 10, 323-333, doi:10.1177/1750698017701615 (2017).

7          Harrington, J. R. & Gelfand, M. J. Tightness-looseness across the 50 united states. Proc Natl Acad Sci U S A 111, 7990-7995, doi:10.1073/pnas.1317937111 (2014).

8          Carey, M. Mistrust: An ethnographic theory.  (University of Chicago Press, 2017).

9          Angrist, J. A., Pischke, J-S. Mostly Harmless Econometrics.  (Princeton University Press, 2008).

10        Marinescu, I. E., Lawlor, P. N. & Kording, K. P. Quasi-experimental causality in neuroscience and behavioural research. Nat Hum Behav 2, 891-898, doi:10.1038/s41562-018-0466-5 (2018).

11        Jaeger, D. A., Joyce, T.J., Kaestner, R. A Cautionary Tale of Evaluating Identifying Assumptions: Did Reality TV Really Cause a Decline in Teenage Childbearing? Journal of Business & Economic Statistics 38 (2020).

12        Sullivan, D., Landau, M. J. & Rothschild, Z. K. An existential function of enemyship: evidence that people attribute influence to personal and political enemies to compensate for threats to control. Journal of personality and social psychology 98, 434-449, doi:10.1037/a0017457 (2010).

13        Rossi-Goldthorpe, R., Leong, Y-C., Leptourgos, P, Corlett, P.R. Paranoia, Self-Deception, and Overconfidence. Psycharxiv Preprint (2021).