‘Big Gods’ book club #3 : Testing more specific hypotheses and going beyond correlations in the orig

(This week, cognitionandculture.net is hosting a “book club” webinar discussing Ara Norenzayan’s latest book with its author. This précis introduced the discussion.)

In his insightful précis, Ara Norenzayan proposes a model of how religions with “Big Gods” have become almost universal. According to this model, the human cognitive architecture coupled with fitness enhancing cultural practices and beliefs allowed human societies to increase in size and become more cooperative than they would have been without them. Religions with big gods, according to this view, are the products of this interplay between our cognitive architecture and cultural innovations. One problem faced by big societies is that the bigger they are the more anonymous interactions between individuals are. And the more anonymous these societies are, the more prone they are to carrying unpunished selfish individuals which in turn increase the probability of these societies to collapse. One solution to this problem is the existence of policing mechanisms.

The evolutionary mechanism proposed by Norenzayan is designed to work as a policing mechanism that prevents large societies from being exploited by selfish individuals by turning a byproduct of social cognition into a cultural adaptation. In this model, the belief in big gods interested in human affairs, such as gods imposing moral imperatives on humans, are thought to have an active causal role in deterring humans from their selfish behaviors. The cognitive mechanisms proposed to underlie this evolutionary functional role are rather simple. If individuals unconsciously or even consciously perceive that they are being watched, they might, as a result, act as if they were surrounded by people observing them — to act more prosocially. This is the supernatural-monitoring hypothesis. But that alone would not explain why religions with big gods are so prominent in human societies. The fear of supernatural punishment has been proposed to explain specifically why Big Gods are so concerned with morality. If one believes that they will be punished if they behave in a different way than the one prescribed by the Big God(s) of their society, this could have important effects on cooperation at the society level. This is because these beliefs might allow individuals who believe in them to monitor their reputation more efficiently than other individuals. Second because these behaviors in turn might allow societies to get bigger and more functional (as a cross-level by-product, or according to some, as a group-level adaptation). As Norenzayan rightly points out, there is now a fairly strong body of evidence that in anonymous situations people tend to be more selfish. There is also evidence that religion is associated with cooperation. This body of evidence is consistent with the elegant multilevel model proposed by Norenzayan. In fact the model is consistent with both the view that the origins of beliefs in supernatural agents are a by-product of our cognition and the explanation as to why big gods, in large societies have moral imperatives. As coherent and insightful this model is, I would like to challenge both its theoretical underpinnings and the empirical evidence used in support of it.

As I suggested earlier, I am sympathetic to the ‘fear of the supernatural’/’supernatural monitoring’ models for the origins of Big Gods religion. Because of their explanatory power, from cognition to inter group(society) interactions, but also because of their bottom-up approach – as it explains social facts from an individual perspective rather than the other the way around. Yet, I think some of the best evidence for them has been misinterpreted, is sometimes insufficient, and their theoretical underpinnings are sometimes not specific enough. In the remainder of this commentary, I will focus on two empirical studies used by Norenzayan to support the ‘Big Gods’ model. I hope I will convince my fellow readers that the “classical” interpretations given of them are problematic.

In the introduction of his précis, Norenzayan mentions the work of Richard Sosis, who in a series of publications defends the ‘costly signalling’ theory of religion. The idea behind the costly signalling hypothesis is that some religions elicit cooperation by causing people to signal their commitment to the group, by paying a cost that they would not be willing to pay otherwise. By signalling the commitment to the group, individuals are effectively cooperating, which is supposed to have a benefit at the societal level and lead to a more cooperative society, or group. Displaying costly signals is thus another form of the policing mechanisms mentioned earlier. One limitation with this hypothesis is that there are always situations in anonymous societies in which cheating is possible and thus costly at the society level: hard-to-fake behaviors are not always an efficient solution against cheaters. Another complementary way to enforce cooperation in anonymous societies is thus, following the fear of supernatural punishment hypothesis, for people to believe in punishing gods. (An evolutionary explanation of why people would acquire such beliefs in the first instance is not easy to demonstrate but let us assume it for now.) Both the ‘costly signalling’ and the ‘fear of supernatural punishment’ hypotheses predict a higher level of cooperation for religious people.

In a 2003 study that was designed to test the costly signaling hypothesis, Sosis measured the average life span of 200 utopian communes during the 19th century. He found that on average religious utopian communes “survive” longer than the non-religious ones. He also found that religious communes with more “costly requirements” (e.g., bans on communicating with outsiders, drinking coffee or wearing particular outfits) lasted significantly longer than the non-religious ones. For one thing, I think that Sosis’ study cannot separate the costly signaling hypothesis from the fear of supernatural punishment hypothesis. The grain of analysis used in the study is too coarse to distinguish the two hypotheses. As a result, I think that no conclusion as to whether the best explanation for this pattern is fear of supernatural punishment or costly signalling should be drawn from this study. In fact, a religious commune might ask its members, for instance, to wear particular clothes or to have specific behaviors, but not all of those mentioned in Sosis’ study are hard to fake. To supplement the costly signalling theory, it is easy to draw a link between moral requirements and the beliefs in big gods. As outlined earlier, big gods can be a way of enforcing compliance to the rules of the commune.

Although the two hypotheses cannot be separated, it seems that a combination of the two mechanisms could explain why religious communities persist longer than secular ones. Although this hybrid model can explain the pattern found by Sosis, there is another possible explanation consistent with the data, which I believe, is more parsimonious and deserving of consideration. In Sosis’ study, only the persistence of communes is measured, but not their migration flow (how many people enter and leave the communes over time). It could well be the case that utopian religious communes have lasted longer because they held a higher turnover than non-religious ones. This itself would be the result of a higher popularity of religious communes over non-religious ones. This, I claim, is more consistent with Sosis’ main findings than the hypothesis of religious beliefs and practices as enforcing cooperation (by costly signals and/or fear of supernatural punishment). Why? In his study, Sosis found that only religious communes with a high number of moral imperatives lasted longer. The presence of a high number of imperatives is, however, not significantly associated with the life span of secular communes. This represents a problem for the costly signaling hypothesis, which predicts that the higher the number of moral requirements, the higher the level of cooperation and thus the higher the life span of a community. If costly signaling is correct, hard-to-fake demonstrations of commitment to a group should be irrespective of whether the commune is religious or not. The fear of supernatural punishment hypothesis can of course explain why in the religious communities, and not the secular ones, this pattern is observed: simply because there is no supernatural agent ensuring that people comply with these imperatives in secular communes, the explanation goes. But there is another, simpler hypothesis. This is the hypothesis that the number of moral imperatives is not an important factor in explaining the life span of utopian communes. The real determinant, in my view, is the popularity of a commune i.e. how many people they attract over a given period of time. The idea that religious communes would attract more people over time than (perhaps more obscure and less open) secular communities (even if they let go the same number of people overall every year) is, I believe, quite a reasonable and simple one that would need to be controlled for. Without this control, the evidence from Sosis’ study only supports weakly the view that religion (and even more specifically big gods) promotes cooperation at all.

I will now turn to another study. In a 2007 paper by Shariff and Norenzayan (“God is watching you”, mentioned in the précis), one group of people were primed with what Shariff and Norenzayan at times call “God concepts” and at other times “religious concepts” (without distinguishing them). Another group was primed with neutral concepts. (Yet another one was primed with some secular moral concepts but that is not important for my purpose here). In his précis, Norenzayan seems to keep this ambiguity between “religious” and “God” primes. Yet, I want to stress here that they are far from being equivalent, and that distinguishing them is crucial to test the fear of supernatural punishment and supernatural monitoring hypotheses, but also to tease them apart. Take the religious primes used by Shariff and Norenzayan: “God”, “Divine”, “Sacred”, “Prophet” and “spirit”. Only “God”, “Divine” and “Spirit” are directly compatible with the idea that when people are primed they feel monitored by a supernatural agent which is the hypothesis they are testing. For “Sacred” and “Prophet” the link between those concepts and being watched is not as obvious and at any rate it is much more indirect. More generally, the claim that religious concepts make people “feel” as is they are watched by a supernatural agent is only one possible hypothesis among many other with the study performed by Shariff and Norenzayan. For example, it could well be the case that these religious primes elicit prosocial behaviors because they make individuals feel they are interacting with family members i.e. not because they are watched, but because one is on average nicer with their kin than with anyone else. Because kinship concepts (such as brother, father, sister etc…) are used in some religions with Big Gods, it is plausible, among many other hypotheses, that what underlies the cooperation of individuals that have been primed with religious concept is that they elicit a sense of family with the person they are interacting with in the dictator game. Without specificity both in the evolutionary mechanisms hypothesized to explain the patterns, but also in the observable variables used to test our hypothesis, inferring what causal mechanism led to the patterns observed today becomes almost impossible. This is another way to formulate the problem with adaptationism and just so stories.

The point I am trying to make here, is basically, that “religious” primes and “God” primes are different and that Shariff and Norenzayan’s study (and its follow-ups) although it shows that a link between cooperation and religion exists, does not however show that the supernatural monitoring hypothesis is the correct one. In a 2010 unpublished study Ryan McKay and I tried to disentangle this problem. We started from the idea that if supernatural monitoring is the basis for why people cooperate, then priming them with supernatural religious agents or supernatural non-religious agents should have no effect on cooperation. Our findings are consistent with this hypothesis, but the level of cooperation was not significantly higher in both groups when compared to the level of cooperation in the control group. This is not incompatible with the hypothesis that religion makes people more prosocial, but this suggests that supernatural monitoring is not like a pair of factitious eyes watching you. The fear of supernatural punishment hypothesis, in my view, which relies on a more conscious notion of cooperation in which individuals are aware that the choice they make might reward or punish them, might be better at explaining prosocial attitudes of believers, but I am not sure Shariff and Norenzayan’s study can distinguish them. However, both the fear of supernatural punishment hypothesis and the supernatural monitoring hypothesis are extremely hard to disentangle, first between themselves and more importantly from other hypotheses (adaptive or not) compatible with the patterns observed. This is, in my view, primarily due to the fact that beliefs in supernatural big gods are naturally correlated with many other traits that provide fitness benefits (whether at the individual or cultural-group level). Thus another problem with Shariff and Norenzayan’s study is that it does not distinguish between the fear of supernatural punishment and supernatural monitoring hypotheses. Of course, one cannot demand that a single study tackle every problem, but there seems to be a body of work that does not make this distinction very clearly. Yet the two hypotheses are different and rely on different mechanisms (actually the supernatural monitoring mechanism is a component of the fear of supernatural punishment hypothesis). To make progress on the supernatural natural/fear of supernatural punishment hypotheses we must find a way to control for other traits and a way to distinguish the two hypotheses empirically and have a clear rationale (perhaps more formal models) for how and supernatural monitoring and fear of supernatural punishments would lead to more cooperative societies.

At the moment, although the hypotheses involving Big Gods seem promising and stimulating, I think the empirical evidence supporting them has too many possible confounding variables and the theoretical underpinnings are too blurred. I have outlined several problems with two widely cited studies in this field, and there are certainly many others. My hope is that future empirical studies will explicitly recognize the different hypotheses involving supernatural monitoring and thrive to control for alternative hypotheses that must be carefully delimitated.


  • Ara Norenzayan 11 December 2013 (02:27)

    I couldn’t agree more with Pierrick’s call for refining and testing the several specific hypotheses about underlying mechanisms. In this précis I did not delve into these important details, but I do so in the book. This book is really meant to outline the lay of the land, and start the conversation about the origins and spread of world religions with their Big Gods. There are many open questions. My hope is that there will be more empirical and theoretical work along the lines that Pierrick is discussing.

    I would like to point out two broad issues. First, what we call “religion” is an over-determined set of phenomena. This means that there are multiple mechanisms at play, which are not necessarily incompatible. There likely are several arrows that go from religious representations and practices to cooperative tendencies. We see a good example of this in Pierrick’s own work with Quentin Atkinson (see abstract here). In their cross cultural survey, they found that belief in a personal God, afterlife beliefs, and religious attendance independently predicted harsher judgments towards a wide range of moral transgressions. As researchers we want to tease out these mechanisms and their boundary conditions, but we want to be careful not to assume that they are mutually exclusive – they often aren’t. Second, the argument that I outline in Big Gods is based on a synthesis. Whatever its flaws, it’s not based on one study, research paradigm, or field. The specific details and mechanisms are extremely important to work out, but at the same time we want to connect the dots and find the best explanation that accounts for the most facts.

    When Azim Shariff and I first published these religious priming results, it was at the time a lone finding, and we were careful to say that there is more than one way to interpret it. What a difference 6 years makes! There is now a growing literature that has replicated and extended this finding, and we have some preliminary answers — more than Pierrick might be thinking. For example, we now have evidence that the same religious primes that increase generosity in the dictator game also increase the sense of being under surveillance (see here), a finding consistent with the supernatural monitoring account. Meta-analysis of all the priming effects indicate that the effect is robust for believers, but goes away for non-believers. This is important because it is more consistent with the idea that the effect of religious awareness on behavior is culturally-contingent (one has to be socialized into religion to show the effect reliably). Also there is growing evidence that fear of supernatural punishment is a key factor in encouraging prosociality, see for example, here, here, and here. Interestingly, supernatural benevolence, if anything, has the opposite effect (encourages cheating and anti-social behavior). These findings also rule out a number of alternative accounts, but there is much more work to be done.

    I also agree with Pierrick that Rich Sosis’ groundbreaking finding on commune longevity is open to multiple interpretations (I don’t think Rich Sosis himself would disagree). The effect he found is very large, so I wouldn’t be surprised if there are, once again, multiple independent mechanisms that are converging. Rich interprets his finding in a signaling framework, but in the book I discuss some challenges with that account that could be addressed if that framework is embedded in a cultural evolutionary logic. The argument I favor looks at these extravagant religious displays and their effects not as costly signals, but as credibility enhancing displays (or CREDs, and related concepts such as epistemic vigilance, discussed in detail in chapter 6). These, once again are distinct mechanisms that are grounded in different evolutionary scenarios. Signaling models are genetic adpatationist accounts that argue that people who display these signals are more likely to be accepted as cooperators. CREDs are a social influence account, whereby cultural learners who are exposed to these credible displays are more likely to adopt the beliefs that these displays communicate. CREDs do not assume that these displays are costly; signaling accounts do. Signaling models don’t predict that exposure to these extravagant displays cause beliefs anchored to these displays to spread in a population. CRED models do. Once again, these are not incompatible and can be mutually energizing explanations. Experimental work that teases apart these mechanisms is just beginning.

  • Pierrick Bourrat 11 December 2013 (09:20)

    Thanks for these clarifications Ara.

    I agree that it is much more productive to try to understand the origins and evolution of religion with a set of non-mutually exclusive hypotheses. I agree also on Rich Sosis’s work. There are many ways to interpret this work, and Rich seems happy with that plurality. That said, my hypothesis, if correct, would have nothing to do with cooperation per se, which is why I think it would deserve to be tested and is what I contend. The fact that religious communes last longer could merely be a by-product of migration flow. Let us not forget, that longevity of communes, in this study, is a proxy for cooperation. But as any proxy, there is not a one-to-one mapping with what the proxy is intended to measure.

    I am also still a bit confused about the difference(s) in predictions between the fear of supernatural punishment (FSP) and the supernatural monitoring hypothesis and the relation with big gods. In my view, although the FSP is the most interesting in relation to big gods, because it gives big gods a particular function with evolutionary consequences: threatening individuals and preventing them from cheating. The supernatural monitoring hypothesis is also interesting but any religion (or like-religion phenomenon) with supernatural agents (invisible or able to know what humans do when no-one is watching) should have a similar effect on cooperation. Am I missing something?

    If for example, some cross cultural study was to found that people worshipping their ancestors (and thinking about them as being able to live after death and watch them) and people believing in God (everything else being equal) did not have the same level of cooperation after they have been primed with their respective worshipped supernatural agent(s), that would be, in my view, bad news for the supernatural monitoring hypothesis as having a distinctive role for in religions with big gods. In fact that would suggest that monitoring is not what makes individuals distinctively cooperating in those religions. For that reason, it seems to me that the kind of theoretical article in the fashion of Johnson and Bering would need to be written for the different flavors of supernatural monitoring hypotheses, their link with cooperation and their exact predictions so that we can clearly see which hypothesis is the most plausible. Maybe I am a bit out of date with the literature, but I have not seen other more recent theoretical papers on this issues.

  • Ara Norenzayan 11 December 2013 (22:30)

    Pierrick’s comment about the distinction between supernatural monitoring (SM) vs supernatural punishment (SP) is an ongoing discussion in the evolutionary studies of religion (see for example this paper by Schloss & Murray in RBB here). The way I see it, this is an important nuance and there are theoretical issues to sort out, but it doesn’t fundamentally change the overall argument in Big Gods. This is because the cultural evolutionary argument says that monitoring or punishing supernatural agents co-opt our ordinary psychology involving mind reading, agency detection, and social instincts that predispose human to norm-following. These effects emerge over historical time and are “packaged together” through cultural evolution — they are not assumed to be unique to religions. In the book I discuss, how, for example secular societies use similar mechanisms of monitoring and third-party punishment to achieve high levels of cooperation and as a result, in some places, Big Governments replace Big Gods. Other accounts that argue that supernatural punishment is a genetic adaptation, such as Johnson and Bering’s account, make different predictions (the similarities and divergences of these two approaches require going over some important technical issues, discussed in some detail in the book, see pp. 135-138). But briefly, I’d say that SP subsumes SM, since there has to be some monitoring or accountability before there could be any punishment. On the other hand, monitoring can have effects without any overt threat of punishment. Currently, I see evidence supporting both accounts. In the earlier post, I cited several empirical papers that show that supernatural punishment (including belief in Hell) increases prosocial tendencies, whereas supernatural benevolence (including belief in Heaven) actually creates some form of moral permissiveness and reduces prosocial behavior (if God is all loving and will not punish, why not cheat and benefit?) There is also a growing literature showing that the mere presence of monitoring or accountability, such as mirrors, cameras, audiences, even illusory monitoring such as incidental exposure to schematic drawings of human eyes, or wearing dark glasses, is sufficient to nudge people to cheat less and be nicer with strangers. My guess is that monitoring works to some extent, but monitoring, combined with the threat of punishment is even more potent. Prosocial religions have stumbled on this powerful package of features: gods who know everything, care about human affairs, particularly bad behavior, and intervene to punish norm-violators. What’s interesting is that to the extent that supernatural benevolence is present, is not there because it encourages cooperation, it is probably gets into the mix because it has other effects (possibly because it provides comfort in times of trouble, and might help attract converts).