The study tests two competing explanations of the secularization process related to rationalizing worldviews and decreasing existential insecurity. While the former explanation argues that people are unwilling to join religious groups because of increasing mechanistic understanding of the world that clashes with religious views (and is rather irreversible), the latter argues that it is the decreasing insecurity that causes secularization and that this trend can be reversed with increasing insecurity. In the present study, 811 secular participants from the USA and Poland played a modified version of the Nash demand game, which simulates dilemmas indexing cooperative insecurity. Participants were randomly assigned to either a secure or insecure environment, manipulated by the parameters of the Nash demand game, and we assessed whether they would be willing to join costly normative groups that regulate cooperation in the game. Crucially, participants were randomly assigned either to a secular condition (choosing between a secular normative group and a group with no norms)—our manipulation check—or a religious condition (choosing between a normative group with religious framing and a group without norms)—main test between the two theories. The results showed that participants in the secular condition were more likely to choose the normative group in the insecure compared to the secure environment, but this difference was inconclusive in the religious condition. However, when re-assigning participants from insecure to secure environments and vice versa, we found strong support for the existential insecurity theory. We discuss potential explanations for the discrepancy between stated and actual behavior as well as potential motivations for joining religious normative groups. This submission has been positively recommended by PCI RR (links to Stage 1 and Stage 2 recommendations).
1. Introduction
Over the past 60 years, scientists observed an intensified decline in the commitment to religious beliefs, values, and practices in various countries worldwide—a phenomenon labeled as ‘secularization.’ One of the early theoreticians of secularization, Peter Berger (1967), defined secularization as “…the process by which sectors of society and culture are removed from the domination of religious institutions and symbols” (p. 113). Berger argued that secularization is both a macro-process by which religious aspects gradually disappear from public institutions (arts, politics, science) as well as a micro-process whereby an increasing number of people cede interpreting the world and their lives through the lenses of religious doctrine. Illustrating these trends, Inglehart (2020, p. p. 15) showed that while 83% of people in the U.S. described themselves as ‘religious’ during the 1982 data collection, this number dropped to 55% in 2017.
While the secularization process intensified in the past decades (Inglehart, 2020), the decline in the importance of religion in both the public and private spheres can be traced to the 19th century in some countries (Brown, 2019; Stark, 1999). As a matter of fact, secularization was a vital topic already for Max Weber—one of the early sociologists—who observed that through economic development, the world becomes more rationalized and ‘disenchanted’ (Weber, 1946, p. 139). Curiously, rather than looking for an outside force, Weber saw the inception of the secularization process in the specific aspects of religious teaching, namely Calvinism. According to Weber, Calvinist ideology first promoted material success as an indication of God’s grace; yet, this promotion had unintended consequences in that to advance material success, people created capitalist structures and bureaucracies that mechanized and rationalized the work, eventually leading to rational and disenchanted worldviews (Hirschle, 2013).
Webers’ ideas were further developed by Berger (1967) and others (e.g., Wilson, 1966), who argued that religious worldviews are incompatible with mechanistic worldviews advanced by science and transmitted through increasingly more accessible education. The proponents of secularization were also quick to predict a near demise of religions, at least in the West. However, the end of the 20th century witnessed an increase in religious affiliation, for example, in the post-Soviet countries (Northmore-Ball & Evans, 2016; Schnabel & Bock, 2017; Stark, 1999; Stolz et al., 2023), and it was clear that the predictions of the secularization theory would not be supported (Stark, 1999). Despite the failed predictions of the early proponents of the secularization theory (as admitted by Berger himself, 1999) and the subsequent departure from the secularization theory by many researchers (Stark, 1999), the theory enjoyed a strong come-back, especially in the light of the recent increase in secularization observed globally (Inglehart, 2020).
The impetus for the renewal of the secularization theory was the analysis of data from large-scale World Value Surveys that suggested a novel explanation for the varying levels of secularization worldwide, namely, existential insecurity (Norris & Inglehart, 2004). Using the Human Development Index published by the United Nations Development Programme and the index of socio-economic inequality, Norris and Inglehart found that both indices predict country-level religious commitment. While more development predicted less religiosity, higher inequality predicted higher religiosity, suggesting that a country’s level of religiosity may be driven by economic, material, and existential insecurities. Norris and Inglehart (2004) further argued that when state-level welfare and individual financial welfare provide individuals with increased security and prosperity, people experience a heightened sense of control and may not feel as compelled to commit to religious values, beliefs, and practices that are more ‘expensive’ than their secular counterparts (Inglehart, 2020; Norenzayan et al., 2016). That is, increasing existential security in countries across the world drives secularization.
Following these initial insights of Norris and Inglehart, other researchers using large-scale sociological surveys amassed supportive evidence for the positive relationship between existential security and secularization. For instance, the country-level importance of religion negatively correlates with economic development and income security across 114 countries (Barber, 2013; see also Höllinger & Muckenhuber, 2019; Kusano & Jami, 2022). Similar results were reported for a subset of European countries where lower income, GDP, and social welfare were associated with higher levels of religiosity (Immerzeel & Van Tubergen, 2013; Storm, 2017; c.f., Te Grotenhuis et al., 2015). Moreover, the individual-level negative relationship between material security and religiosity was found in a sample of 15 small-scale societies, including hunter-gatherers, pastoralists, and wage laborers (Baimel et al., 2022; c.f. Purzycki & Bendixen, 2024), although the same relationship was not observed using the World Values Survey (Höllinger & Muckenhuber, 2019). Summarizing this evidence, some researchers concluded that religiosity declines wherever state spending on social welfare increases (Dhima & Golder, 2021; Gill & Lundsgaarde, 2004; Stolz, 2020).
Importantly, the existential insecurity theory posits that the commitment to religious beliefs, values, and practices is dynamically sensitive to insecurity, implying that wherever the trend in increasing security is reversed, the reversal should be followed by an increase in religiosity. Several studies support this proposition. For example, people with professions negatively affected by the farming crisis in the 1980s in the U.S. were more likely to increase their church attendance compared to people with other professions, and a similar relationship was found for the 2007/8 housing crisis (Orman, 2019). In Indonesia, an economic crisis increased the probability that parents would send children to Islamic schools and attend prayer groups more often (Chen, 2010). Increase in religiosity was also observed in countries with a recent history of violent conflict (Cesur et al., 2020; Henrich et al., 2019; Keinan, 1994; Shai, 2022; Uecker, 2008; Van Tubergen et al., 2023), following natural disasters (Sibley & Bulbulia, 2012; Sinding Bentzen, 2019) or the Covid-19 pandemic (Bentzen, 2021; Chvaja et al., 2024).
Notwithstanding the evidence supporting the existential insecurity hypothesis, the proponents of the rationalization theory, which roughly follows the insights of early secularization theoreticians, argue that the main driver of secularization is the mechanization of worldviews associated with increasing technological mastery over the world, not decreasing insecurity (Bruce, 2011). For example, a recent study on automation and religious beliefs found a negative relationship between these variables both between and within countries, and a higher work-related exposure to AI was associated with a later decrease in the intensity of religious beliefs (Jackson et al., 2023). Using education as a proxy for the rationalization and mechanization of worldviews, other studies found that increasing the compulsory years of schooling was associated with decreased religiosity in Canada (Hungerman, 2014) and that earning a college degree was associated with a sudden decline in religiosity in a U.S. longitudinal study (Schwadel, 2016). The negative relationship between a college degree and religiosity holds across 39 nations (Schwadel, 2015). Finally, when explaining the religiosity gender gap using data from 14 small-scale societies, Vardy et al. (2022) found that differential access to education between men and women was one of the main mediators of the positive difference between men and women in religiosity.
The reviewed support for both theories presents a conundrum for secularization researchers since both explanations seem valid at the same time. Part of the issue stems from the fact that both theories often rely on similar data or data that cannot disentangle their predictions. Illustrative of this issue is the use of the variable tracking educational attainment, which is part of the Human Development Index, used by the proponents of the existential security theory, as well as the main explanatory variable utilized by the proponents of the rationalization theory. Of course, disentangling complex social processes is challenging, but it is even more challenging using correlational, cross-sectional data, as is often the case with the research reviewed above (with some notable exceptions). While secularization is a long-term, population-level process that is most often of interest to sociologists working with large-scale surveys, these surveys cannot clearly distinguish between the entangled causal forces of secularization. To help overcome this issue and contribute to understanding the secularization process, the current study employs an experimental paradigm, which allows manipulation and comparison of the key mechanisms hypothesized to drive secularization.
In the present study, we take advantage of the different predictions regarding the reversal of the secularization process made by the rationalization and existential insecurity theories; namely, that individual rationalization is mostly an irreversible process decreasing religiosity while individual insecurity dynamically affects religiosity in both directions (Inglehart, 2020, p. p. 5; see also Stark, 1999, p. p. 253). We manipulated cooperative insecurity (Peysakhovich & Rand, 2016) and investigated whether secular participants would be willing to join religious institutions safeguarding cooperation, mimicking the real-world processes described above (Chen, 2010; Henrich et al., 2019; Orman, 2019). However, in contrast to the real-world data tracking the intensification of religious beliefs and behaviors, we sampled secular participants and examined whether they would use religious institutions as a fallback option in the absence of secular institutions.
Specifically, we asked participants to play a modified Nash demand game (Nash, 1951) where two participants simultaneously draw money from a common pool, and the remaining amount is multiplied by a factor χ and equally redistributed between participants. Participants were randomly assigned to a secure or insecure environment. While in the secure environment, cooperation (i.e., withdrawing nothing) was the most profitable choice and freeriding was limited and, therefore, a suboptimal choice, freeriding was highly beneficial and tempting in the insecure environment and could result in cooperative players earning nothing.
Furthermore, participants were randomly assigned to a religious or secular condition that differed in the type of group participants could choose to play the game with. In both conditions, participants were able to choose between a group without norms regulating withdrawal or a group with norms (religious or secular, depending on the random assignment). While in the secular condition, norms requiring abstaining from withdrawal were presented as the law of the group, in the religious conditions, the same norms were legitimized and mandated by a supernatural deity. The choice of the group was our main outcome variable.
The rationale for this experimental paradigm follows the emphasis on the economic development variables used in the previous research to explain secularization (e.g., Immerzeel & Van Tubergen, 2013). While the existential insecurity theory assumes that religious beliefs and behaviors respond to various types of insecurities and anxieties (Lang et al., 2020; Pargament et al., 2000), most of these insecurities can also be ameliorated by a social safety net; in other words, by the cooperative affordances of an ego’s social network that are responsive to external hardships (Bauer et al., 2016; Lang, Xygalatas, et al., 2022; Majolo & Maréchal, 2017). Therefore, we focused specifically on cooperative insecurity as this insecurity broadly underlies other insecurities and likely drives some of the previous findings on the positive effects of increased insecurity on religiosity (Henrich et al., 2019). Furthermore, rather than aiming to make people less religious (i.e., more secularized), we simulated decision-making of whether to join a religious group without the assumption of underlying religious beliefs, analogically to real-world decision-making such as whether to attend a collective ritual (often used as the main outcome variable in research on secularization; Orman, 2019).
Using this experimental setup, we tested the predictions of the rationalization and existential insecurity theories, investigating competitive hypotheses derived from the respective theories. First, as a manipulation check, we tested the effectiveness of our insecurity manipulation on the choice of normative groups in the secular condition (HMC). We predicted that there will be a difference of at least 15 percentage points (%pt.) between the secure and insecure conditions in the probability of choosing the normative group. In other words, cooperative insecurity should affect participants’ group choice.
We further preregistered that following supported HMC, we would test whether insecurity increases the probability of choosing the normative group in the religious condition (H1). That is, whether participants would be more likely to join religiously framed groups to protect their cooperative interests in the insecure compared to the secure environment. We set the smallest effect size of interests (SESOI) for this difference at 10 %pt. An estimated difference including SESOI or larger effects was preregistered as supporting the existential insecurity theory while a differences below SESOI would be interpreted as supporting the rationalization theory (see Figure 1 for illustration).
Manip check = Manipulation check; Existential insecurity = Existential insecurity theory; Rationalization = Rationalization theory. Values in the secure environment (model intercepts) mapped from pilot data. The thick lines represent the SESOI supporting manipulation check and existential insecurity theory. Manipulation check assumes at least 15 %pt. difference between the secure and insecure environments in the secular condition. Support for the existential insecurity theory assumes at least 10 %pt. difference between the secure and insecure environments in the religious condition. Support for the rationalization theory assumes this difference to be < 10 %pt. The thin lines illustrate other possible outcomes consistent with the theories and checks.
Manip check = Manipulation check; Existential insecurity = Existential insecurity theory; Rationalization = Rationalization theory. Values in the secure environment (model intercepts) mapped from pilot data. The thick lines represent the SESOI supporting manipulation check and existential insecurity theory. Manipulation check assumes at least 15 %pt. difference between the secure and insecure environments in the secular condition. Support for the existential insecurity theory assumes at least 10 %pt. difference between the secure and insecure environments in the religious condition. Support for the rationalization theory assumes this difference to be < 10 %pt. The thin lines illustrate other possible outcomes consistent with the theories and checks.
To test these hypotheses, we planned to recruit a sample of 440 participants from Poland and the USA (together 880 participants). The rationale for conducting the experiment across two countries was to tease apart individual and national-level relationships (Schwadel, 2015). While we were primarily interested in the individual-level effects, these effects may be sensitive to nationwide characteristics (Ruiter & Van Tubergen, 2009). We chose to compare the U.S. and Poland because both societies are currently undergoing an intensive process of secularization (Inglehart, 2020) and thus allow us to sample secular participants who are actively refusing religious institutions that are still present in the public domain (as opposed to countries with a majority of religiously non-affiliated population such as the Czech Republic). For example, out of 46 countries surveyed worldwide by the Pew Research Centre, Poland has the largest difference between young and old cohorts on the importance of religion in their lives (Pew Research Center, 2018). Importantly, secularization theorists often contrast the USA with European countries to demonstrate either the lack of secularization in the former region or different processes driving secularization (Marger, 2013; Pérez-Agote, 2014; Schnabel & Bock, 2017; see also Voas & Chaves, 2023 for a review). Thus, our sample should be able to test whether secularization theories might apply across both contexts or if one may be more fitting to the U.S./European context.
Finally, since we focus on cooperative uncertainty in this study, we also need to consider that cooperative uncertainty does not affect all participants to the same extent. Specifically, the need for institutions safeguarding cooperation may be diminished or even opposite for uncooperative participants (Lang, Chvaja, et al., 2022; Peysakhovich et al., 2014). For such participants, these institutions may hinder their plans and the selection of institutions might be disfavored by uncooperative participants, especially in the insecure condition where freeriding is the most profitable option. Thus, we also preregistered to test H1 on a subsample of participants playing a cooperative strategy.
2. Methods
2.1. Ethics Information
This study was approved by the Ethics Committee for Research at Masaryk University. Participants provided informed consent and received a show-up fee of 1.3 USD (and its equivalent in PLN in Poland; we use USD as an example henceforth) plus any amount they earned in the Nash demand game.
2.2. Design
The whole study was conducted in a survey format using the Qualtrics software (see the OSF folder associated with this project). The study utilized a 2 (environment: secure vs. insecure) x 2 (institution: religious vs. secular) between-subjects and double-blinded design. Participants were blind to our hypotheses and did not interact with any researchers nor directly with any other participants (the Qualtrics software handled random assignment to various conditions). The survey was distributed in English to U.S. participants and in Polish to participants from Poland. After logging into the survey, participants were first asked to sign an informed consent and then fill out questions on age, gender, generalized trust (Rosenberg, 1956), risk aversion (Dohmen et al., 2011), religious affiliation, belief in punitive god, attitudes about religion (Pew Research Center, 2024), and the Social Value Orientation (SVO) measure, which assessed general cooperative preferences using a battery of nine scenarios where participants decided how to distribute money between themselves and others (Van Lange et al., 1997). After reporting the registered analyses, we used these variables to refine our statistical models and get more nuanced insights that may inform future studies/theory-building.
Next, participants were introduced to a variation of the one-shot Nash Demand game (ND; Nash, 1951). We used a variation of ND reported by Ruffle and Sosis (2007), who showed that men in religious kibbutzim in Israel were more cooperative in this version of the ND game when playing together compared to men from secular kibbutzim. The rules of the ND game are as follows: participants are paired in dyads and can withdraw from a common pool of size η, either 0 or Σ. Whatever is left in the common pool is multiplied by a factor χ and redistributed equally between the two participants. However, if the sum of the withdrawal by the two players exceeds the size of the pool—η, participants get nothing from the common pool.
Participants were provided with three illustrations of how the game could evolve to affect players’ earnings, and we asked participants a check question on one of those scenarios to ensure they understood the game’s rules. In case of an incorrect answer, participants were given one more chance to answer, and if they failed again, they were re-directed to the end of the survey without the possibility of partaking in the ND game (but receiving the show-up fee). Those who demonstrated sufficient understanding of the rules were randomly assigned to a secure or insecure environment.
To simulate secure and insecure cooperative environments (the environment factor), we manipulated the η, Σ, and χ factors to benefit freeriding more in the insecure condition compared to the secure condition. In the insecure condition, participants could withdraw from a larger pool of money (ηinsecure = 6 USD; ηsecure = 2 USD), but there was no multiplication factor compared to the factor of three in the secure condition (χ insecure = 1; χ secure = 3). Moreover, in the insecure condition, Σ = η, allowing for full freeriding, while in the secure condition, Σ = 0.25, making freeriding a suboptimal choice and securing earnings even for players who were taken advantage of.
In contrast to the secure condition where cooperation was the option where one can earn the most (3 USD for full cooperation vs. 2.88 USD for full freeriding) and cooperation should be perceived as the best choice even without institutional safeguarding, participants in the insecure condition could be tempted to withdraw more from the common pool for personal gain (or worry about this temptation in others), potentially destroying cooperation (3 USD for full cooperation vs. 6 USD for full freeriding). Thus, participants in the insecure condition should experience a larger demand for institutions safeguarding cooperation, even if these institutions are somewhat costly and repressive. We tested a range of parameter values for χ, η, and Σ and selected the current parameters as giving an optimal contrast between the secure and insecure conditions to test H1 (see the Pilot section in SM).
Following the random assignment into the secure or insecure condition, participants were asked about the amount they expect other players to withdraw from the common pool. Next, participants were randomly assigned to the religious or secular condition (the institution factor). This assignment affected the type of groups participants could choose from when deciding with whom they would like to play the ND game. In both conditions, participants were introduced to a no-norms group and a group with norms regulating withdrawal. However, in the religious condition, regulations in the normative group were said to be guaranteed by a supernatural agent, while in the secular condition, these regulations were guaranteed by the laws of the group.
In both conditions, the groups without norms were defined as follows:
Group X does not regulate withdrawal and has no entry requirements. There is no need to punish for withdrawing. Members pay no tax on earnings.
In the secular condition, the instructions for the normative group was as follows:
Group Y asks members to forgo the possibility to withdraw. To enter, it is required to spend 0.5 minute reading the groups’ rules. Those who withdraw in this group may face punishment for withdrawing which will confiscate part of their earnings. Members of this group will be taxed at 5% of any earnings by the game.
In the religious condition, the instructions was as follows:
Group Y asks members to obey God’s will and forgo the possibility to withdraw. To enter, it is required to spend 0.5 minute reading a religious text. Those who withdraw in this group may face punishment by God. Members of this group will be taxed at 5% of any earnings by the game.
The text participants who chose the normative group read Biblical verses, namely Proverbs 10:9, Leviticus 19:11, and Luke 16:10. The text is identical for the religious and secular conditions: we selected passages that do not directly refer to God and are general enough as not to be recognized as Biblical verses in the secular condition. The inclusion of the tax and the reading task was meant to stimulate the cost of upkeeping policing institutions as well as deter free-riders from joining these normative groups (Lang, Chvaja, et al., 2022, 2024). The possibility of punishing non-zero withdrawals was added for the same effect. We reasoned that if secular participants regard supernatural punishment as a delusion, as the rationalization theory would have it, they would not be willing to pay the entry costs and rely on the supernatural punishment mechanism. Participants should be willing to rely on the secular mechanism, though. Likewise, if religious institutions are flexibly recruited in times of insecurity, even secular participants may be willing to pay the tax and the reading task cost and rely on the supernatural punishment mechanism. Note that to make the religious and secular normative groups comparable, the mentioned possibility of punishing withdrawals was not realized after data collection.
After choosing the group in which they would like to play—our main outcome variable—participants were asked to state the reason for their choice with their own words. Participants who chose the normative group then engaged in the reading task. Next, all participants continued to make their withdrawal choice and we asked them about the expected behaviors of other participants in both normative and non-normative groups as well as expected religiosity of those participants. Finally, we introduced participants to the opposite condition of the environment factor they were randomly assigned to and asked them how much they think others would withdraw from the common pool and which group they would choose under this game setup (in the same condition of the institution factor they were assigned into). These additional data assessed how sensitive participants were to our manipulation and how flexible their group choice was.
Upon finishing this procedure, participants received their show-up fee. After the end of data collection, we randomly paired participants in the same conditions according to the groups they chose and paid them the bonus from the ND games according to their game decisions and the decisions of a randomly selected partner. Participants who chose the normative group were further taxed at 5%. For an overview of the design, see Figure 2.
Manipulated variables are in italics, and the main outcome variable is in bold.
Manipulated variables are in italics, and the main outcome variable is in bold.
2.3. Sampling
To estimate the expected statistical power to find an effect hypothesized in H1, we first selected the smallest effect size of interest (SESOI; Lakens et al., 2018) in terms of the percentage-point (%pt.) difference. Three streams of evidence informed the selection of SESOI. First, prior studies using Likert scales to measure the effect of adversity exposure (e.g., war, earthquake, pandemics) on belief in God and importance of religion indicate that when such measures are scaled between 0-1, the average effect of adversity amounts to 0.075 increase in outcome variables. This effect ranges between 0.04 and 0.1 (Cesur et al., 2020; Chvaja et al., 2024; Shai, 2022; Sinding Bentzen, 2019). However, note that such re-scaling of ordinal variables does not directly translate to %pt. increase, and these studies used a variety of measures/controls in their models. More relevant to the current study, two studies measured religious affiliation after earthquake and war exposure (Henrich et al., 2019; Sibley & Bulbulia, 2012). The former shows that an earthquake in New Zealand increased religious conversion by 3.4%pt. in the affected regions; the latter shows a difference in religious affiliation in people differentially affected by war exposure in Siera Leone, Uganda, and Tajikistan, ranging between 6 and 19%pt. Finally, in a prior study where religious and secular participants chose between groups with whom they would like to play a public goods game (Lang et al., 2023; OSF preregistration), we observed the lowest probability of choosing a religious group by secular participants at 11%. Nevertheless, this study lacked manipulation of insecurity and studied a different mechanism stabilizing cooperation compared to the current study.
Taking these previous studies as rough guidelines, we observe differences between 3 and 19%pt., with the lowest effect of an experimental online study placed around the middle of this continuum. Since our current study is an online experiment, we set the SESOI of insecurity on choosing the religious institution (testing H1) at 10%pt. increase, which is roughly in the middle of the observed effects and congruent with the previous experimental study. We consider effects with 90% CIs below the 10%pt. increase as practically negligible (Lakens et al., 2018).
To select the needed sample size for the detection of this SESOI at 90% statistical power, we tested the power of sample sizes ranging from 200 to 500 (in the steps of 50) in the religious condition and estimated the power to detect non-equivalence between the predictions of the rationalization and existential insecurity theories (see R code at OSF). To this end, we used the command powerSim from the simr package (Green & Macleod, 2016) in R. Specifically, we performed 500 Monte Carlo simulations to re-fit the planned logistic model, assessing the binomial ratio of models with statistically significant/non-significant interaction results (at the significance level α = 0.05). Note that since we planned to use logistic regression as our main statistical model (see below), the power to detect a 10%pt. difference crucially depends on the size of the intercept of this comparison (in our case, the secure environment in the religious condition). We assumed an 8% probability of the normative-group choice based on our pilot data (see Figure S1 in SM). To detect a 10 %pt. increase in the insecure condition with 90% statistical power, 450 participants in the religious condition were needed. However, if the probability of normative group choice in the secure religious condition would rise to 10%, our statistical power to detect the 10%pt. difference between the secure and insecure environments in the religious condition decreased to 81%, which we still deemed acceptable. This sample size also gave us a probability of 95% that the 90% CIs would include SESOI if the real effect would be 10%pt.
Second, we performed a statistical power analysis for our manipulation check (HMC). While our pilot data already showed a well-estimated difference of 21 %pt. between the secure and insecure environments in the secular condition (β = 1.56; 95% CIs [0.56 – 2.55]; see Pilot section in SM), we performed a power analysis to detect the smallest effect considered to validate our manipulation (15 %pt. difference). This effect was selected to be larger than SESOI for H1 to assure that if the effect of insecurity in the religious condition would be weaker than in the secular condition, our manipulation would be impactful enough to allow for the weaker effect to be detected. The power analysis suggested that the 15 %pt. difference in the secular condition would be detected with 89.6% power with a sample of 350 participants.
Combining the required samples for testing HCM and H1, the power analysis suggested a sample of 800 participants. To ensure 800 participants for the analyses, we planned to recruit 880 participants, 440 per country. This oversampling assumed that 10% of participants would be excluded according to the criteria specified below (in a pilot study, these criteria disqualified 10% of participants). Participants were recruited through the platform Prolific.co, which already pre-screened their participant pool on religious affiliation so that we could select participants reporting no religious affiliation. We planned to recruit a sample with a balanced gender across our conditions. Participants had to be at least 18 years old. We further preregistered that participants who would not pass the understanding check or complete the whole study in under three minutes would be excluded from the analyses (three minutes is the minimum length of filling out a similarly long survey in our past research; Lang et al., 2023). We also planned to exclude participants who would report a religious affiliation (which would conflict with information provided by Prolific.co) or those who would report belief in punitive God (as per our manipulation where punitive god is the securing mechanism in the religious institution). Regarding handling missing data, we made answering questions informing our key hypotheses compulsory. For the missing control variables, we planned to exclude participants pairwise. The primary outcome variable was bounded between 0-1, so we did not expect to detect any outliers.
2.4. Analysis
Analyses were conducted in R (R Core Team, 2020) using the glmmTMB package (Brooks et al., 2017). As planned, we to built one generalized linear mixed model (model1.1) to assess HMC, and the same model structure (model1.2) with a different target group was used to test H1. Specifically, we used a Generalized Linear Mixed Model (GLMM) with a binomial distribution that modeled the difference in the probability of choosing a normative group in the secure/insecure conditions, accounting for the assumed data-generation process as well as the hierarchical structure of our data (participants nested within countries). As per SESOI described above, our preregistered inference criteria state that the positive effect of the environment variable on choosing the religious institution supports the existential insecurity theory if 90% CIs of this effect would include SESOI and 95% CIs exclude 0. If the 95% CIs of the effect would exclude 0 but 90% CIs would not contain SESOI, we interpret this effect as negligible (Alter & Counsell, 2023) and in support of the rationalization theory. Similar interpretation applies if 95% CIs would include 0, but 90% CIs would not include SESOI. Finally, if 95% CIs would include 0 and 90% CIs SESOI, this result can be interpreted as inconclusive regarding H1. See Figure 3.
The x-axis represents the %pt. difference estimated by the logistic regression between secure and insecure environments in the religious condition in the probability of choosing the normative group. The vertical dashed line denotes SESOI. The thicker horizontal error lines are 90% CIs, and thinner horizontal error lines 95% CIs.
The x-axis represents the %pt. difference estimated by the logistic regression between secure and insecure environments in the religious condition in the probability of choosing the normative group. The vertical dashed line denotes SESOI. The thicker horizontal error lines are 90% CIs, and thinner horizontal error lines 95% CIs.
Finally, as preregistered, we test H1 on a subsample of prosocial participants as categorized by the SVO measure (model 1.3). We reasoned that the effect of insecurity on the choice of normative group in the religious condition may be estimated more precisely in prosocial participants because people with selfish strategies would likely select the non-normative group in the insecure condition, biasing the estimates. Based on the pilot data, we expected about 2/3 of participants to be scored as prosocial. Looking at the simulations of statistical power for H1, 2/3 of the sample size (300 participants) gives us 75.8% power to detect SESOI, that is, 10%pt. increase in the probability to choose the religious normative group the insecure compared to the secure condition. See Table S1 in SM for an overview of the design and minor deviations from the registered plan.
3. Results
3.1. Participants
We collected data from 880 participants on Prolific.co (440 from US and 440 from Poland). In addition, 60 participants returned the study during filling out and did not submit their data. From 880 participants, two participants did not finish filling out the survey, 12 were excluded because they filled out the survey under 3 minutes, and 119 were excluded for reporting either affiliation to a religious tradition or belief in God. After exclusion, the dataset contained 757 participants, which did not reach the planned statistical power. Hence, we recruited additional 60 participants, from which 1 did not fill out the whole survey and 5 were excluded due to their religious affiliation. The final data set comprises 811 participants: 355 in the secular condition (Mage = 32, SD = 10, range = 18-71; 173 women and 14 other gender) and 456 in the religious condition (Mage = 33, SD = 11, range = 19-76; 227 women and 7 other gender). Regarding the geographical distribution, 395 participants came from Poland and 416 from USA. Additionally, 1% of our sample view religion positively, 57% had mixed views and 42% negatively. Using SVO, we further classified participants as prosocial (66%) or individualist (27%; the remaining 6% had missing data).
3.2. Auxiliary Manipulation Checks
In the secular condition, participants in the insecure environment reported that other participants will be more likely to withdraw money from the common pool compared to participants in the secure environment (β = 1.40, 95% CIs = [0.98 – 1.82]). The cumulative link model estimated that participants had 72% probability of saying that others are likely to withdraw money in the insecure condition while this probability dropped to 39% in the secure condition. A similar, yet smaller, effect was observed in the reversed scenario where participants from the insecure condition were assigned to the secure condition and vice versa (β = 0.55, 95% CIs = [0.16 – 0.94]).
Focusing on the religious condition, the cumulative link model estimated that participants in the insecure environment expected that it is more likely others will withdraw from the common pool (67% probability) than participants in the secure environment (29%; β = 1.60, 95% CIs = [1.23 – 1.97]). This effect held in the reversed scenario, albeit to a lower extent (β = 0.40, 95% CIs = [0.06 – 0.74]). These results suggest that participants in the religious condition were also sensitive to the environmental manipulation. Furthermore, participants in the religious condition expected that religious people would choose the normative group more likely (80% probability) than the no-norms group (34%; β = 2.04, 95% CIs = [1.83 – 2.25]).
We also checked whether prosocial participants would be less likely to withdraw money from the common pool, an assumption we hold when reasoning for using the subset of prosocial participants. Indeed, the GLMM estimated that prosocial participants were less likely to withdraw (β = -1.36, 95% CIs = [-1.77 – -0.95]), and this difference was larger in the insecure compared to the secure environment (βinteraction = -0.24, 95% CIs = [-0.35 – -0.13]). Specifically, in the secure environment, individualist participants had a probability of withdrawing 14% while prosocial participants 6%, and in the insecure environment, these probabilities rose to 44% and 12%, respectively.
3.3. Main Manipulation Check
We predicted at least 15%pt. positive difference between the secure and insecure environments in the probability of choosing the normative group as our manipulation check. The GLMM estimated that participants chose the normative group with probability of 26% in the secure and 44% in the insecure condition, a difference that was well estimated (β = 0.81, 95% CIs = [0.36 – 1.25]; see Figure 4A, secular institution). Furthermore, we observed a similarly well-estimated effect in the reversed scenario (β = 0.74, 95% CIs = [0.31 – 1.17]), indicating that our manipulation check was not caused by sampling bias in one of our environmental manipulations.
Hollow shapes are raw data, full shapes estimates from GLMM and error bars are 95% CIs. (A.) Participants in the insecure environment in the secular condition were more likely to choose the normative group compared to participants in the secure condition. This difference was inconclusive in the religious condition. (B.) The effect of environment held in prosocial participants in the secular condition but was attenuated in the religious condition. (C.) Pooling the regular and reversed scenarios (where participants were asked to imagine their group choice in the environment opposite to which they have been originally assigned), we observed the hypothesized effect in both the secular and religious conditions, and (D.) this effect held in the subset of prosocial participants.
Hollow shapes are raw data, full shapes estimates from GLMM and error bars are 95% CIs. (A.) Participants in the insecure environment in the secular condition were more likely to choose the normative group compared to participants in the secure condition. This difference was inconclusive in the religious condition. (B.) The effect of environment held in prosocial participants in the secular condition but was attenuated in the religious condition. (C.) Pooling the regular and reversed scenarios (where participants were asked to imagine their group choice in the environment opposite to which they have been originally assigned), we observed the hypothesized effect in both the secular and religious conditions, and (D.) this effect held in the subset of prosocial participants.
3.4. Main Analysis of H1
We observed 10% probability of choosing the normative group in the secure environment and this probability increased to 15% in the insecure environment. The upper 90% CI of this effect was at 10%, our SESOI. However, the 95% CIs of this effect included zero (β = 0.48, 95% CIs = [-0.10 – 1.05], Figure 4A, religious institution), yielding this result to be exactly on the verge of the tested theories and inconclusive (see Figure 3). The observation that 95% CIs for the main effect included zero might have been caused by the fact that the probability of choosing the normative group in the secure condition rose to 10% (we expected 8%, see Sampling plan), making the focal difference more difficult to detect (due to the constraints of the binomial distribution) and decreasing our planned statistical power.
3.5. Additional Planned Analyses
We therefore analyzed only a subset of data focused on prosocial participants (as planned), yielding support for the rationalization theory: the difference in the probability of choosing the normative group was 10% in both environments after rounding (β = 0.10, 95% CIs = [-0.66 – 0.85]; see Figure 4B, religious institution) with the upper 90% CI of this effect reaching only 7%. We further planned that in this situation, we would pool our data sets from regular and reversed scenarios, increasing our statistical power by including the within-subject effect. This pooled model revealed substantial support for H1 where the probability of choosing the normative group was 7% in the secure and 18% in the insecure environment. This effect was well estimated (β = 1.07, 95% CIs = [0.64 – 1.49]) and the upper 90% CI was at 15% (see Figure 4C, religious institution). Moreover, this effect held when the pooled data set was limited to prosocial participants (β = 0.94, 95% CIs = [0.34 – 1.53]; Figure 4D, religious institution).
We also explored whether the effect of environment in the pooled data set holds when adjusting the model for potentially confounding variables we collected. Indeed, we found the effect of environment on group choice held (β = 1.14, 95% CIs = [0.69 – 1.59]) when accounting for the effects of negative attitudes about religion, expected number of religious people in the normative group, gender, age, cooperative type (prosocial vs individualist), generalized trust and risk proneness. Finally, directly comparing the effects of the environmental manipulation between the secular and religious conditions on group choice, we observed a 6%pt. larger increase in the secular compared to the religious condition (βinteraction = 0.06, 95% CIs = [0.004 – 0.12]). See Table S2 in the SM for the overview of results.
4. Discussion
This registered report aimed to test between two theories of secularization: the existential insecurity theory and the rationalization theory. Specifically, we investigated whether secular participants in the insecure environment would be more likely to join a religious normative group compared to secular participants in the secure environment. Whereas our manipulation check confirmed that participants are sensitive to our insecurity manipulation and are more willing to join normative groups in insecure environment, the focal test of the two theories was inconclusive: although participants in the insecure environment were more likely to choose a religious normative group than participants in the secure environment and the 90% CIs of this effect included the preselected SESOI, the 95% CIs of this estimated difference contained zero. Following the preregistered plan, we further examined the difference in group choices between the insecure and secure environments in the subset of prosocial participants, finding only a negligible difference that suggested support for the rationalization theory. In the final step, we investigated group choices by combining regular and reversed scenarios (adding a within-subject factor) and found strong support for the existential insecurity theory.
We believe that the main factor affecting the inconclusive H1 test was the somewhat higher baseline rate of joining the religious group in the secure environment. Given the specific shape of the binomial distribution underlying logistic regression, reliably detecting concrete %pt. change requires increasingly larger samples with increasing baseline rate to which this %pt. difference is being added to (see Methods). Indeed, while the 95% CIs for the difference between the secure and insecure environments included zero in the religious condition, most of the probability mass was in the positive direction, and with a larger sample size, it appears reasonable to expect that the focal difference would be detected.
4.1. The Stated vs Actual Behavior Discrepancy
To increase sample size, we pooled data from regular and reversed scenarios. While we included the reversed scenario to assess whether the potentially detected effect is sensitive to our insecurity manipulation (and is not just a sampling error), we did not expect the effects of our insecurity manipulation on religious normative group choice to be stronger in the reversed scenario. Indeed, the results of the reversed scenario provide substantially stronger support for the existential insecurity theory, requiring a further explanation of the discrepancy between the actual and stated behavior in a different environment. One possible explanation relates to the sequence of our study: participants were first assigned to the behavioral test and only later stated their group preference in the reversed scenario. This contrast might have made the insecurity threat more salient for participants who were originally secure or vice versa, effectively increasing the effect of insecure environment on the normative group choice. However, looking at the auxiliary manipulation checks, the estimated difference in expected withdrawals between the secure and insecure environments was smaller in the reversed scenario compared to regular scenario, a finding that empirically contradicts the proposed explanation.
Another explanation of the discrepancy between stated behavior in the reversed scenario and actual behavior in the regular scenario may be that in the reversed scenario, paying the cost of joining a normative group may seem reasonable when the payment is in the future, but this cost becomes too large when it needs to be paid in the regular scenario—a phenomenon labeled temporal discounting (Critchfield & Kollins, 2001). In other words, participants may be willing to join normative groups in the future because they may discount the future costs. In support of this explanation, we observed such a dynamic in the secular condition (our manipulation check) across the secure and insecure environments, where participants in the reversed scenario were more likely to select the normative group than in the regular scenario (βregular vs reversed = 0.47, 95% CIs = [0.07 – 0.87]). On the other hand, in the religious condition, this was true only for the insecure environment where we observed a 9 %pt. increase in choosing the normative group (between the regular and reversed scenarios), while in the secure environment, this difference decreased by 3 %pt. (βinteraction = 1.09, 95% CIs = [0.22 – 1.96]). This latter result suggests that the proposed temporal discounting was moderated by our environmental manipulation in the religious condition, a result that contradicts the general assumption that any future costs should be discounted.
Finally, it could also be that the SVO test we used to categorize participants into prosocial and individualist may lack contextual sensitivity, decreasing our ability to detect the effect of insecurity on participants with cooperative intentions in the regular scenario (where we expected a stronger effect in cooperative participants). While our auxiliary manipulation check showed that SVO was highly predictive of withdrawal decisions, an alternative analysis may test H1 using a subset of participants who did not withdraw (rather than a hypothetical SVO measure of prosociality). One caveat to this approach is that withdrawal decision procedurally followed group choice, so such an analysis needs to assume that group choice was made with a withdrawal strategy already in mind—an assumption we cannot test with the current data. Bearing this caveat in mind, we again did not find support for the existential security hypothesis in the regular scenario, although the results do not provide such an obvious support for the rationalization theory as when using the SVO measure: (βsecure vs insecure = 0.32, 95% CIs = [-0.29 – 0.93]; upper 90% interval of the difference between secure and insecure environment was at 9%pt., just below our 10%pt. SESOI).
4.2. Opportunistic Motivations for Joining Religious Groups
Although our data do not afford a satisfactory explanation of the discrepancy between the regular and reversed scenarios, overall, even hypothetical willingness to join a religious group should signal support for the existential insecurity theory. Recall that the rationalization theory assumes mechanistic worldviews that do not rely on supernatural forces punishing norm trespassing. In our design, the safeguarding function of the religious normative group depended on belief in supernatural punishment, so even entertaining the possibility that such normative groups may be functional should conflict with the mechanistic worldview. Of course, cooperative participants might have stated that they would join religious groups in Voltaire’s fashion (“If God did not exist, we would have to invent him”; cited from Williams, 2009, p. 202), relying on others’ belief in supernatural punishment in the religious normative group to deter them from freeriding. The same may be true for participants who planned to withdraw from the common pool, relying on religious groups to contain participants who are less likely to withdraw (e.g., when choosing the religious normative group, a participant reasoned: “They might believe that their god would really punish them and thus not withdraw”). Nevertheless, the same caveat applies to real-world support for the existential insecurity hypothesis where religious affiliation rises after natural disasters, wars, and economic crises (Henrich et al., 2019; Orman, 2019; Sibley & Bulbulia, 2012): while such disaster-related conversions contravene the secularization trend, the actual motivations of converts are unknown, and many may affiliate with opportunistic motivations.
Importantly, we should not expect that such conversions start with a newly found belief in supernatural agents and acceptance of doctrinal knowledge provided by the respective religious traditions. Influential sociological models of religious conversion (Lofland & Stark, 1965; Rambo, 1993; for a review, see Gooren, 2007) posit that while personal tension/crisis is often a trigger for seeking religious/spiritual change (cf., Snow & Phillips, 1980), conversion is a complex gradual process facilitated by social networks and personal attachments to church affiliates rather than the specific teaching of the church. As a matter of fact, interviews with converts reveal that they initially stay in the church because of the social bonds and despite the church’s teaching, which often conflicts with their views (Lofland & Stark, 1965, p. 871). It is the social network and joint ritual activities that make the doctrinal teachings and beliefs gradually conceivable (Petranek, 1988; Sosis, 2003).
In light of these conversion theories, participants who stated they would join the religious normative group in the insecure condition may be considered as potential converts (Lofland & Stark, 1965), even though they would join this group opportunistically. Of course, this result does not mean that participants would entertain the same idea in real life during cooperative tension/crisis—it is substantially easier to join a religious group opportunistically in an experimental scenario where other options are artificially closed than in real life. Moreover, religious groups often have costly mandates that may deter potential opportunistic converts, especially those that would join the group with exploitative motivations (Bulbulia, 2012; Iannaccone, 1992; Ruffle & Sosis, 2007; Sosis, 2003). While we aimed to simulate this cost in our experimental design, such one-shot costs are less effective compared to repeated costs because free-riders may still reap strategic benefits without punishment (Lang, Chvaja, et al., 2022, 2024). Thus, to compensate for the potential overestimation of ‘joining rates’ in our experiment compared to real life, we set higher SESOI than would be expected in real life, providing a less sensitive but more robust test of the two secularization theories.
4.3. Limitations and Future Directions
While our experimental setup had somewhat limited ecological validity, as much as participants expected interactions with religious individuals (see Results), their decisions should resemble a potential real-life decision-making process. Importantly, we trade ecological validity for the ability to test an essentially causal theoretical prediction that was previously tested using correlational data. The somewhat lower ecological validity of such an experimental test is also necessitated by ethical concerns of assigning participants to insecure conditions in real life or manipulating the conversion process. Although such concerns may be to some extent ameliorated using longitudinal data (see Lang, Palíšek, et al., 2024 for support of the existential insecurity hypothesis using longitudinal data), even longitudinal data do not guarantee Granger causality without careful theoretical modeling of potential confounds (Rohrer et al., 2022). However, in combination with experimental manipulation, they may provide a more robust assessment of inspected causal processes.
Using such manipulation, our design aimed to distinguish between two opposing theories, but we manipulated factors affecting only one of the purported secularization mechanisms (insecurity) while relying on null results to test the other mechanism (rationalization). Ideally, we would also manipulate rationalization and mechanistic worldviews to test whether those indeed cause lower willingness to join religious normative groups and rely on a supernatural-punishment mechanism. However, while perception of insecurity can fluctuate with relatively high frequencies (e.g., within months or years) and can be experimentally manipulated, it is difficult to imagine that people with rational, mechanized worldviews would switch between rational and supernatural explanation with similar frequencies. Whereas the absence of rationality manipulation decreases the sensitivity of our test between the two theories, we believe that such manipulation would provide less valid conclusion due to its artificial nature.
Furthermore, while we interpreted the stated willingness to join religious normative groups as a support for the existential insecurity theory, it is still true that the majority of participants refused to join the religious normative group in the insecure condition. This refusal may have been motivated by their mechanistic worldviews. It is possible that insecurity affects only a certain portion of population, and while this portion of population may resist or partially reverse secularization trends, in the long-term, the majority with mechanistic worldviews would prevail. That is, while secularization may be reversible with increasing insecurity contra the expectations of the rationalization theory (Bruce, 2011, p. 68), it may still be that rationalization by and large causes the secularization process. A potentially fruitful avenue to reconcile these opposing theories would be to investigate rationalization as a continuum with specific thresholds where some secular people remain open to different worldviews while others strictly adhere to a mechanistic worldview absent of the possibility of supernatural powers. Formal models of population dynamic of cultural transmission (Henrich, 2004; McElreath & Henrich, 2012) might further reveal the proportion of open-minded secular individuals and the amount of existential insecurity that would be needed to upkeep religious institutions in secularizing countries as well as identify breaking points that would lead to irreversible loss of religious institutions.
The above-suggested simulation, as our study, critically relies on cooperative insecurity as the pressure that should trigger religious affiliation. Nevertheless, there are other potential insecurities, as reviewed in the original formulation of the existential insecurity theory and its later extensions (Inglehart, 2020; Norris & Inglehart, 2004) that may also be relevant. For example, insecurities related to health (Krause, 2010; Sibley & Bulbulia, 2012; Xygalatas et al., 2019), physical threats through crime and war (Henrich et al., 2019; Laurin et al., 2008), or fear of death (Jong et al., 2012, 2018) may have similar or even stronger effects than cooperative insecurity. Similarly, people may search for meaning and solace in the face of disasters and personal tragedies, and while secular institution may fulfill material-security needs, deeper, meaning-laden insecurities may be best assuaged by religious institutions. Future research may use the current design to test the strength of these pressures, although we caution about the validity of such manipulation—the advantage of cooperative insecurity is that participants can be ethically faced with such a situation rather than only reminding themselves of this situation or reading hypothetical scenarios.
Finally, the selected samples limit the inferences from the current study: we purposefully selected countries undergoing the secularization process where we expected to find enough secular participants yet where, at the same time, conversion to religious tradition would be a feasible possibility and option people could realistically entertain through their social network (Lofland & Stark, 1965). The contextual factors preceding conversion are as important as the active search process for religious solutions (Rambo, 1993). It may be that we would not detect the same effect in highly secularized countries where religions seldom enter the public arena. On the other hand, the lack of country effect in the current results indicates that our design may tap into a more general decision-making process, which is partially independent of the context of the given country. Nevertheless, our sample of countries is obviously very limited for such a conclusion (N = 2).
5. Conclusion
In summary, this registered report found mixed support for the existential insecurity theory when secular participants stated that they would be willing to join religious normative groups under hypothetical cooperative insecurity, but this effect was inconclusive during actual insecurity. While the discrepancy between actual and stated behavior suggests that the detected effects would be much smaller in real life where participants face actual costs of joining religious groups and may have other feasible alternatives, this study adds support to the proposition that the secularization process may be reversed with increasing socio-ecological pressures. This result does not invalidate the rationalization theory—the majority of our sample did not choose the religious normative group, and this pattern may have been motivated by rationalized worldviews in some participants. Yet, the fact that secularization appears to be a reversible process suggests that rationalization is either not as uni-directional as its proponents suggested or that it is a rather than the factor contributing to secularization.
Data and Code Availability
Data, materials and R code are available at: https://osf.io/9qtnj/?view_only=298f21f78168442cbe1b7fd79ac122c2
Stage 1 Manuscript
Acknowledgements
We would like to thank Radek Kundt and Eva Kundtová Klocová for providing feedback on the Stage-2 manuscript and two reviewers and editor at PCI-RR for their helpful feedback. This publication was made possible through the support of Grant 61928 from the John Templeton Foundation managed by The Queen’s University of Belfast awarded to ML. RC was supported by the Templeton Religion Trust (Grant ID: TRT-2022-30378). The funders have/had no role in the study design, data collection and analysis, decision to publish or preparation of the manuscript. The opinions expressed in this publication are those of the author(s) and do not necessarily reflect the views of the John Templeton Foundation or The Queen’s University of Belfast.
Author Contributions
M.L. and R.C. developed the idea for the current study and designed the experiment. M.L. collected and analyzed data. M.L. drafted the registered report and R.C. provided comments.
Competing Interests
The authors declare no competing interests.