Keywords

Introduction

Religious beliefs and behaviors, although they exist in many forms (Moro & Myers, 2010; Smith, 1991) and may not always be viewed as a formal “religion” by their practitioners, appear to be human universals (Atran, 2002; Norenzayan, 2010). Although there is debate regarding how to define religion, the assertion of religion’s universality is based on a definition by Atran (2002, p. 13) that is broad enough to avoid excluding less “traditional” religions and comprises four components:

  1. 1.

    Widespread counterfactual belief in supernatural agents (gods, ghosts, goblins, etc.)

  2. 2.

    Hard-to-fake public expressions of costly material commitments to supernatural agents—that is, sacrifice (offerings of goods, time, other lives, one’s own life, etc.)

  3. 3.

    A central focus of supernatural agents on dealing with people’s existential anxieties (death, disease, catastrophe, pain, loneliness, injustice, want, loss, etc.)

  4. 4.

    Ritualized and often rhythmic coordination of 1, 2, and 3—that is, communion (congregation, intimate fellowship, etc.)

These components exist in all known societies, converging into what may reasonably be defined as “religion.”

Despite the apparent universality of religion, the degree of religiousness (i.e., religiosity) varies across individuals, societies, and time. For example, Sweden and Denmark are among the least religious nations (excluding nations with state-imposed atheism, which do not accurately represent the religiosity of the populace), with studies reporting a range of 46–85 % of Swedes and 43–80 % of Danes as nonbelievers in God (Zuckerman, 2007). These percentages may be large when compared to other nations, but they also highlight the substantial variability in religiosity at the individual level. Even though these societies are highly secular, there is a substantial portion of highly religious individuals in these populations (Zuckerman, 2008). Furthermore, this predominant secularism has not been constant throughout these nations’ histories, as Zuckerman (2008) notes regarding the late 1700s and 1800s:

…there is no question that heartfelt, faithful Christianity was discernibly pervasive in various parts of Denmark and Sweden,” and “…ever since sociologists began collecting data on religion in Denmark and Sweden—which, admittedly, really wasn’t that long ago—the clear pattern has been that of decline, in both belief and participation. (p. 125)

Besides Denmark and Sweden, developed, postindustrial nations tend to exhibit lower religiosity than less-developed nations (Norris & Inglehart, 2004; Zuckerman, 2009). The United States, however, appears to be an exception. Recent estimates of the percentage of atheists, agnostics, or nonbelievers in the United States range from 3 to 9 %, which places the United States 44th—between Portugal and Albania—on a list of the top 50 countries with citizens self-identified as such (Zuckerman, 2007). Nevertheless, United States religiosity has been declining in recent decades. An analysis of the 1990, 2001, and 2008 waves of the American Religious Identification Survey (ARIS) indicates that the percentage of Americans designated as religious “nones” (i.e., those who do not identify with any particular religion) increased from 14.3 million (about 8 % of the population) in 1990 to 34.2 million (15 %) in 2008 (Kosmin & Keysar, 2009).

How can we explain this individual, societal, and temporal variability in religiosity? Although historical and cultural factors play a role in shaping individual and societal religiosity over time, it can also be useful to analyze religion from an evolutionary perspective.

Evolutionary Psychology and Its Application to Religion

Evolutionary psychology is not a subdiscipline of psychology, such as social psychology or personality psychology, but rather an approach to psychology that applies evolutionary theory. Evolutionary psychology is founded on the premise that the brain, like every other organ, has evolved and is therefore open to analysis from an evolutionary perspective, which means that the products of the brain (i.e., thoughts, feelings, behaviors) are open to evolutionary analysis. For example, an evolutionary psychological approach has proven useful in examining social behavior (Cosmides, 1989; Kenrick, Maner, & Li, 2005), learning (MacDonald, 2007; Weber & Depew, 2003), memory (McBurney, Gaulin, Devineni, & Adams, 1997), development (Bjorklund & Pellegrini, 2000), and perception (Rhodes, 2006), to name a few diverse topics. All aspects of psychology have the potential to be better understood by examining them from an evolutionary perspective, and religious beliefs and behaviors are no exception.

Evolutionary psychologists argue that the mind is composed of a large number of evolved mechanisms shaped throughout evolutionary history to solve specific, recurrent adaptive problems of survival and reproduction. Although the number of evolved mechanisms that exist and the typical scope of such mechanisms (i.e., domain specific vs. domain general) are subjects of debate, a less controversial aspect of this view is the description of evolved mechanisms as information-processing mechanisms sensitive to specific types of information (i.e., environmental stimuli, physiological activity, output from other parts of the brain), and this information is processed, resulting in a specific type of output (i.e., physiological activity, input to other mechanisms, or manifest behavior) (Buss, 2011).

From this conceptual foundation, one can examine religious beliefs and behaviors by considering the possible evolved psychological mechanisms that produce such output. Identifying the types of information—particularly environmental input—that influence religiosity can aid in identifying the mechanisms involved in producing religious beliefs and behaviors and the functions of these mechanisms, which can aid in determining whether religious beliefs and behaviors are produced by specialized mechanisms or as the byproducts of other evolved mechanisms (i.e., output that is merely a consequence of a mechanism’s design rather than output that a mechanism is specialized to produce).

Although this chapter presents a test of a theory of religiosity that was not developed by evolutionary psychologists, an evolutionary perspective can be useful in understanding the theory’s predictions and interpreting findings related to this theory.

Secure Society Theory

Norris and Inglehart (2004) proposed a theory of secularization that may explain the variability in religiosity between nations as well as societal changes in religiosity over time. They argue that a key factor driving secularization is the level of security provided by a society, which influences individuals’ “existential security” or “…the feeling that survival is secure enough that it can be taken for granted” (p. 4). Their theory is referred to throughout this chapter as “Secure Society Theory.”

Secure Society Theory is built on two premises, referred to by Norris and Inglehart (2004) as “the Security Axiom” and “the Cultural Traditions Axiom” (pp. 13–18). The Cultural Traditions Axiom does not play a large role in the theory’s explanation of secularization, but rather emphasizes that the religious worldviews of a society continue to influence that society’s culture even as that society moves toward secularization (e.g., the Protestant work ethic). More relevant to explaining religious variability over time is the Security Axiom, which states that variability exists between societies with regard to the level of security (i.e., people’s vulnerability to risks and dangers, such as environmental disasters, diseases, crime, human rights violations, poverty, etc.) and that societal shifts from agrarian to industrial and from industrial to postindustrial improve societal security. The first stage of modernization (agrarian to industrial) lifts developing nations out of extreme poverty, aiding the most vulnerable portions of the population and improving the standard of living. As societies develop, there are improvements in nutrition, sanitation, access to clean water, healthcare, and education, as well as improved mass communication. These changes have a positive impact on individuals’ perceptions of security (i.e., existential security).

Norris and Inglehart (2004) acknowledge that societal development does not inevitably lead to greater security, at least not for all citizens. This makes sense when considering Secure Society Theory from an evolutionary perspective. If the psychological mechanisms that produce religious beliefs and behaviors are sensitive to societal security, then even a prosperous, developed nation can have a religious populace if cues to insecurity are present. For example, certain events can have significant negative impacts on any nation regardless of the level of development (e.g., natural disasters, war, recession) and will impact individuals’ perceptions of security. Economic inequality is also is an important variable, as a substantial portion of the population may continue to suffer from threats to their security while a small “elite” class of citizens reaps the benefits of development.

Using the Security Axiom and Cultural Traditions Axiom as their foundation, Norris and Inglehart (2004) hypothesize that the variability in security between societies, resulting from varying levels of development and historical events, can partially explain the variability in religiosity between societies, with greater security leading to increased secularization. More specifically, they predict that greater security will result in weaker religious beliefs, values, and participation and that differences in religiosity will be most pronounced between agrarian, industrial, and postindustrial societies.

This relationship between religiosity and security is based on the argument that as individuals’ perceptions of security increase, their need for religion decreases, as religion—particularly supernatural beliefs—provides a coping mechanism for living in less secure and unpredictable conditions. That supernatural beliefs afford coping with uncertainty was hypothesized by Malinowski (1954) and has since been supported by a variety of studies. For example, regular church attendance is linked to a reduced incidence of depression, suggesting greater ability to cope with stress (McCullough & Larson, 1999), and several studies indicate that individuals compensate for uncertainty, ambiguity, and reduced feelings of control through superstitious (Burger & Lynn, 2005; Case, Fitness, Cairns, & Stevenson, 2004; Keinan, 2002) and religious (Frijters & Barón, 2010; Kay, Gaucher, McGregor, & Nash, 2010; Kay, Moscovitch, & Laurin, 2010; Kay, Whitson, Gaucher, & Galinsky, 2009) beliefs and behaviors. Whitson and Galinsky (2008) have even shown that the experience of lacking control can increase the perception of illusory patterns, including developing superstitions and forming illusory correlations about stock market data. In evolutionary terms, these findings suggest that religion functions as an adaptation (i.e., that religious beliefs and behaviors are designed output of psychological mechanisms responding to environmental threats), a view implicit in Secure Society Theory and further considered in the Discussion below.

Norris and Inglehart (2004) expand on the hypothesized relationship between security and religiosity by noting that societal changes in security are not expected to have an immediate impact on individuals’ religiosity, but rather that these effects should take time. If Secure Society Theory is correct, changes in religiosity over time may correspond to societal changes in security. Identifying this relationship requires consistently gathered data over an adequate length of time. Evidence regarding the hypothesized relationship between security and religiosity is discussed next.

Evidence Supporting Secure Society Theory

Norris and Inglehart (2004) conducted a series of analyses to test Secure Society Theory. Their primary source of data for religiosity was the pooled World Values Survey/European Values Survey conducted in four waves from 1981 to 2001. This data set provided information about 76 nation-states. However, not all nations were included in each wave, so time series analyses were limited to 20 societies.

The specific measures of religiosity included religious participation, both collective (attending religious services) and personal (prayer frequency), religious values (the importance of religion in one’s life), and religious beliefs (belief in God, heaven/hell, life after death, and existence of the soul). Regarding societal security, Norris and Inglehart (2004) categorized societies as agrarian (n = 23), industrial (n = 33), and postindustrial (n = 23) based on the Human Development Index, a 100-point scale of societal modernization published annually by the United Nations Development Programme. This measure combines levels of knowledge (adult literacy and education), health (life expectancy at birth), and standard of living (real per capita GDP). Additional measures of security/development drawn from a variety of sources included the proportion of the population living in rural and urban areas, the Gini coefficient of economic inequality, access to mass communications, the number of HIV/AIDS cases, access to an improved water source, immunization rates, the distribution of physicians, and average life expectancy at birth.

Norris and Inglehart (2004) first examined differences in religiosity between agrarian, industrial, and postindustrial nations. As hypothesized by Secure Society Theory, religious participation, values, and beliefs were strongest in agrarian societies and weakest in postindustrial societies. For example, 54 % of respondents in agrarian societies reported praying every day, compared to 34 % and 26 % of those living in industrial and postindustrial societies, respectively.

Norris and Inglehart (2004) then conducted correlational analyses between the various measures of security/development and religious behavior (attending religious services and prayer frequency). These results also supported Secure Society Theory, with each societal indicator correlating with both religious participation and prayer frequency in the predicted direction: As societal conditions improve, religiosity decreases. The correlations ranged in strength from 0.41 to −0.74. For example, the Human Development Index was negatively correlated with both religious participation and prayer frequency, rs = −0.53, ps < 0.001.

Norris and Inglehart (2004) also examined historical trends regarding religiosity, specifically the annual trends in regular (weekly) religious service participation from 1970 to 1998 for 13 European societies. For each society, the year of the survey was regressed on the proportion of respondents reporting weekly religious service attendance. Every model resulted in a negative regression coefficient, and this result was statistically significant for nine of the societies. Although these results show that religiosity is in decline in these European societies, the analyses did not address the causal effect of societal security on these declines.

Finally, Norris and Inglehart (2004) note that the United States appears to represent an exception to their theory. Although the United States religiosity is declining, it remains an outlier compared to most postindustrial nations. The high level of religiosity observed in the United States seemingly contradicts Secure Society Theory, as the United States is a successful postindustrial nation. However, when the United States is analyzed in terms of societal indicators of security, the high rates of religiosity are less anomalous. For example, Norris and Inglehart highlight that the United States exhibits greater economic inequality (as measured by the Gini coefficient) than any other postindustrial nation included in their analyses. They further state that:

Many American families, even in the professional middle classes, face risks of unemployment, the dangers of sudden ill health without adequate private medical insurance, vulnerability to becoming a victim of crime, and the problems of paying for long-term care of the elderly. (p. 108)

Although Norris and Inglehart do not analyze these additional factors and their relationship to religiosity statistically, they make a strong case for the value of Secure Society Theory for understanding religiosity in the United States, as well as throughout the world. Nevertheless, it is important to consider additional evidence regarding the validity of Secure Society Theory.

Additional Supporting Evidence of Secure Society Theory

Since Norris and Inglehart’s (2004) initial presentation of their theory, several researchers have further tested Secure Society Theory either implicitly or explicitly. For example, Paul (2005) tested the hypothesis that popular religiosity is beneficial to society by examining rates of religious belief and practice along with several indicators of societal health and dysfunction (homicide, youth suicide, sexually transmitted disease prevalence, teen pregnancy and birth, and abortion rates) in 18 developed democracies, including the United States. Paul (p. 7) concluded that “higher rates of belief in and worship of a creator correlate with higher rates of homicide, juvenile and early adult mortality, STD infection rates, teen pregnancy, and abortion in the prosperous democracies.” Furthermore, the United States is an outlier regarding most societal indicators of dysfunction, with homicide rates, STD infection rates, early adolescent pregnancies, and abortion rates much higher than in the other countries analyzed.

Given that the United States is an outlier among prosperous democracies on several indicators of societal security, it is important to consider whether the observed relationship between religiosity and security is being driven primarily by the inclusion of the United States in statistical analyses. This possibility was considered in a later study by Paul (2009), in which analyses similar to those conducted earlier (Paul, 2005) were performed with and without the United States included. The newer study also included the creation of a “Successful Societies Scale” (SSS) based on over two dozen indicators—several of which were used in Paul’s (2005) study—and a “Popular Religiosity Versus Secularism Scale” (PRVSS) comprising seven measures of religiosity and secularism (absolute belief in a supernatural creator deity, biblical literalism, religious service attendance, prayer frequency, belief in an afterlife, self-reported agnosticism/atheism, and acceptance of human descent from animals). Higher scores on the SSS indicated less societal dysfunction, and higher scores on the PRVSS indicated higher levels of secularization. Results indicated that scores on the SSS positively correlated with scores on the PRVSS, with the United States included (r = 0.71, p < 0.001) and excluded (r = 0.53, p < 0.01), although the relationship is larger with United States inclusion. These results are consistent with Secure Society Theory, as many of the societal measures are indicators of societal security (e.g., homicides, incarcerations, life expectancy, infant mortality, Human Poverty Index), and the relationship with religiosity is not driven by data from the United States.

A study by Rees (2009) further tested Secure Society Theory with an analysis of 55 countries. This study also tested alternative explanations for changes in religiosity: the traditional modernization theory of secularization and Rational Choice Theory, the latter of which states that secularization occurs “due to competition for attention from secular services and the provision of unattractive products by the monopoly of religious providers” (Rees, 2009; p. 2). After establishing economic inequality—measured by the Gini coefficient—to be a reasonable proxy for personal insecurity given its correlation with several societal indicators of security, Rees developed a model with economic equality and variables related to the alternative explanations of secularization (governmental and social regulation of religion, religious fractionalization, and per capita GDP) as predictors of religiosity. The results supported Secure Society Theory, indicating that while taking into account other variables, economic inequality remains a unique predictor and was indeed the strongest predictor of religiosity.

Barber (2011) provided further support for the link between economic inequality—as well as other variables likely to influence existential security—and religiosity. As a measure of religious disbelief, Barber relied on the proportion of the population reporting that they do not believe in God, as compiled by Zuckerman (2007) for 137 countries. Barber controlled for the effect of living in Communist societies (where religious beliefs are criminalized) and Islamic states that observe Sharia law (where atheism is criminalized). Independent variables included economic development measured in terms of the proportion of the labor force employed in agriculture and third-level education enrollment, economic security measured in terms of the Gini coefficient and the level of personal taxation (a proxy for the extent of the welfare state), and health security measured in terms of the severity of 22 parasites (i.e., “pathogen prevalence”) as reported by Fincher and Thornhill (2008). The results indicated that religious disbelief was correlated with all the independent and control variables, and these variables were all predictors of religious disbelief in regression analyses, explaining 75 % of the variance in disbelief.

Pesta, McDaniel, and Bertsch (2010) provided an indirect test of Secure Society Theory by creating an index of well-being for the United States, using the 50 states as the units of analysis. They identified six “sub-domains” of subjective well-being for which state-level data are available: religiosity, health, crime, education, income level, and g, or general intelligence. Although religiosity was included because of its documented positive effect on well-being (see Pesta et al., 2010), correlational analyses indicated that religiosity was positively correlated with the only sub-domain representing lower well-being—crime—and negatively correlated with every other sub-domain representing greater well-being. In other words, although religiosity has beneficial effects on well-being at the individual level, higher levels of religiosity are associated with lower-state-level well-being. This apparent contradiction is consistent with Secure Society Theory, as individuals living in states with stronger indicators of lower well-being (e.g., higher crime rates, lower health, lower education, etc.) are expected to display greater religiosity in an effort to cope with these conditions.

Finally, a study by Solt, Habel, and Grant (2011) analyzed economic inequality and religiosity over time, providing a test of the temporal component of Secure Society Theory (i.e., changes in societal security result in changes in religiosity over time). Solt et al. analyzed data over a 50-year period, from 1955 to 2005. Grant’s (2008) Aggregate Religiosity Index (ARI), which provides a single value of national religiosity for each year based on available survey data, was used as the measure of religiosity, and economic inequality was measured by the Gini coefficient. GDP per capita was also included in their analysis. To test the effects that these variables have on each other over time, Solt et al. analyzed these data with vector autoregression, a form of time series analysis that comprises a series of regression equations. Each variable under consideration serves as the predictor in one of the equations. More specifically, a time-lagged version of each variable serves as a predictor to determine whether it can predict future values of the other variables. The analyses indicated that GDP per capita negatively predicts future religiosity, whereas economic inequality positively predicts future religiosity (i.e., as inequality increases, future religiosity increases). However, the time lag considered in this analysis was only 1 year, as preliminary analyses indicated this was the most appropriate lag. Therefore, although the results are in the direction hypothesized by Secure Society Theory, such a brief time lag may not provide a proper test of Secure Society Theory’s hypothesis regarding gradual historical change in religiosity.

The Present Study

A major limitation of the previous research on Secure Society Theory is the lack of direct data regarding existential security (i.e., people’s personal perceptions of security). Instead, religiosity has been linked to societal indicators of security. Though this type of investigation is important for testing Secure Society Theory, as existential security should be correlated with societal conditions, it is necessary to demonstrate the relationship hypothesized by Norris and Inglehart (2004) between personal perceptions of security and religiosity.

Another limitation of previous studies is the lack of time series analyses, because Secure Society Theory hypothesizes that changes in security over time will lead to changes in religiosity. Although Norris and Inglehart (2004) analyzed historical changes in religiosity, they did not investigate the hypothesized causal relationship between security and religiosity over time. Solt et al. (2011) attempted to address this, but their time series analysis relied on a time lag of just 1 year. Although their results supported the hypothesized causal relationship between changes in security and religiosity, their brief time lag is not consistent with the gradual (i.e., generational) changes hypothesized by Secure Society Theory.

The present study attempts to address these limitations as well as replicate previous findings regarding Secure Society Theory by analyzing religiosity, societal security, and perceptions of security in the United States from 1972 to 2012 through use of the General Social Survey (GSS) (Smith, Marsden, Hout, & Kim, 2013) and data provided by the Federal Bureau of Investigation and the United States Census. Because Secure Society Theory hypothesizes a negative relationship between societal security and religiosity, another way of framing this is that Secure Society Theory hypothesizes a positive relationship between societal insecurity and religiosity. This is how the hypothesized relationship was framed in the present study, given the variables available for analysis.

The present study tested three hypotheses. By including a measure of personal perception of insecurity, the present study tested the hypothesis that personal perceptions of insecurity are positively related to self-reported religiosity (Hypothesis 1). This personal measure was analyzed along with societal indicators of insecurity used in previous studies (crime rates and poverty rates) to test the hypothesis that both personal perceptions and societal indicators of insecurity uniquely predict self-reported religiosity (Hypothesis 2). Finally, religiosity and insecurity were analyzed at the national level over time to test the hypothesis that insecurity positively predicts future religiosity (Hypothesis 3).

Method

Data Sets and Variable Selection

To test the Secure Society Theory as it applies to religiosity in the United States, data were obtained from several data sets. These data sets, and the relevant variables from these data sets that are analyzed in the present study, are described below.

The General Social Survey. For variables regarding religiosity and one variable regarding perceptions of societal insecurity, the present study relied on data obtained from the GSS. This sociological survey was conducted almost annually from 1972 to 1994 (excluding 1979, 1981, and 1992, due to funding limitations) and biennially from 1994 to the present (although data from 2014 were not yet available at the time of this study), by the University of Chicago’s National Opinion Research Center (NORC), and provides a valuable source of time series data on American demographic characteristics and attitudes on a wide range of topics.

The GSS is a 90-min, in-house interview of a probability-based sample of noninstitutionalized US adults at least 18 years old. From 1972 to 1993, for each year the survey was conducted, the target sample size was 1500 participants, with actual sample sizes ranging from 1372 in 1990 to 1613 in 1972. Since 1994, the GSS has been administered to two samples each year the survey is conducted, each with a target size of 1500 participants. Total sample sizes range from 2765 in 2002 to 2992 in 1994. Aside from an oversampling of black participants in 1982 and 1987 (statistically controlled for in the present study’s analyses), there has been no oversampling in other periods.

The 1972–2012 GSS data set consists of a total of 56,355 participants (25,804 men, 30,551 women) after correcting for an oversampling of black participants. The mean age of participants is 44.37 years (SD = 17.00, range = 18–89). The majority of participants (82.2 %) are identified as white (n = 46,328), with 6906 participants (12.3 %) identified as black and 3120 participants (5.5 %) identified as “other.” The primary religious identifications of participants are as follows: 32,289 Protestant, 14,533 Catholic, 1111 Jewish, and 5994 “none.” The residence of participants is coded into nine regions; this unfortunately reduces the amount of regional variability in religiosity and security that can be analyzed, but provides more detail than data at the national level.

The questions in the GSS are of two broad types: Participants either simply provide an answer and the interviewer is responsible for assigning the appropriate code to the response, or participants are given a hand card with a list of possible responses from which to choose. In 2002, the GSS switched from printed questionnaires to computer-assisted personal interviewing (CAPI), but hand cards are still provided to participants for relevant questions.

Measures of Religiosity. The following variables from the GSS were used as measures of religiosity for the present study. Although other variables related to religion are available in the GSS, the current variables were chosen because they were included in the greatest number of surveys from 1972 to 2012.

Religious Attendance. Participants were asked “How often do you attend religious services?” and responses were coded on a scale of 0 to 9 (0 = “never,” 1 = “less than once a year,” 2 = “about once or twice a year,” 3 = “several times a year,” 4 = “about once a month,” 5 = “2–3 times a month,” 6 = “nearly every week,” 7 = “every week,” 8 = “several times a week,” 9 = “don’t know” or no answer). Participants were not provided a hand card with these options, but interviewers were instructed to use these categories as probes if necessary. This question has been asked every year the GSS has been administered, resulting in 29 years for which data have been collected over a 40-year period (valid n = 55,821; 534 cases of “don’t know” or no answer).

Participants tend to overstate the frequency of church attendance, in both the GSS and other surveys, such as Gallup polls (Hadaway, Marler, & Chaves, 1993). The principal investigator of the GSS has acknowledged this issue and explains it as a result of three factors: social desirability bias, telescoping, and participants relying on a broader interpretation of “attend[ing] religious service” (Smith, 1996). However, these data can still be used to analyze changes over time and relationships with other variables because the difference between reported attendance and actual attendance (which has been more accurately measured with time diary studies) has remained consistent in recent decades (Chaves, 2011).

Prayer Frequency. Participants were asked “About how often do you pray?” and responses were coded on a scale of 1 to 6 (1 = “several times a day,” 2 = “once a day,” 3 = “several times a week,” 4 = “once a week,” 5 = “less than once a week,” 6 = “never”). Participants were not provided a hand card with these options, but interviewers were instructed to use these categories as probes if necessary. The GSS began asking this question in 1983, but it was not included in 1986 and 1991, resulting in 18 years for which data have been collected over a 29-year period (valid n = 27,816; 324 cases of “Don’t know” or no answer).

These data were reverse-coded in the present study so that greater values indicate higher levels of prayer frequency. Additionally, for data from 1983, “Never” responses were collapsed with “Less than once a week.” This was recommended by Smith (1988) due to an unusually high number of “Never” responses coded in that year. Although the wording of the question and instructions for interviewing and coding were not changed, it is possible that the discrepancy was due to inadequate interviewer training or inadequate care by the coding supervisor.

Biblical Fundamentalism. Participants were asked “Which of these statements comes closest to describing your feelings about the Bible?” and responses were coded on a scale of 1 to 4 (1 = “the Bible is the actual word of God and is to be taken literally, word for word”; 2 = “the Bible is the inspired word of God, but not everything in it should be taken literally, word for word”; 3 = “the Bible is an ancient book of fables, legends, history, and moral precepts recorded by men”; 4 = “Other” [volunteered]). The first three options were provided to participants on a hand card. The GSS began asking this question in 1984, but it was not included in 1986, resulting in 18 years for which data have been collected over a 28-year period (valid n = 27,618; 349 cases of “Other,” 644 cases of “don’t know” or no answer). These data were reverse-coded in the present study so that greater values indicate higher levels of fundamentalism. Responses of “Other” were excluded from analyses due to a lack of information about how to interpret these responses.

Strength of Affiliation. Participants who indicated any religious preference were asked “Would you call yourself a strong (preference named) or a not very strong (preference named)?” and responses were coded on a scale of 1 to 4 (1 = “strong,” 2 = “not very strong,” 3 = “somewhat strong” [volunteered], 4 = “no religion”). Participants were not provided a hand card with these options, but interviewers were instructed to use these categories as probes, if necessary. Interviewers were instructed to refer to the religious preference previously identified by the participant when asking this question. If participants indicated that they follow no religion, this question was not asked. The GSS began asking this question in 1974, and it has been asked every year since, resulting in 27 years for which data have been collected over a 38-year period (valid n = 51,436; 1797 cases of “don’t know” or no answer). These data were reverse-coded in the present study so that greater values indicate a greater strength of affiliation. Also, the order of the “not very strong” and “somewhat strong” categories were switched to more accurately reflect a scale of increasing religiosity. Therefore, the final coding as used in the present study is as follows: 1 = “no religion,” 2 = “not very strong,” 3 = “somewhat strong,” and 4 = “strong.”

Religiosity Component. To facilitate analyzing the effects of several variables on religiosity, the present study considered whether religious attendance, prayer frequency, biblical fundamentalism, and strength of affiliation could be combined into a single religiosity composite variable. A principal components analysis was conducted on these variables, with extraction based on eigenvalues greater than 1. The number of factors to extract was not fixed beforehand. Two measures of sampling adequacy, Kaiser–Myer–Olkin (KMO) and Bartlett’s test of sphericity, indicated that it was appropriate to proceed with principal components analysis, as KMO was greater than 0.5 (KMO = 0.760) and Bartlett’s test of sphericity was significant (p < 0.001).

The four religiosity variables are all correlated. A principal components analysis yielded a single component with an eigenvalue of 2.44, explaining 61.04 % of the variance. The second largest eigenvalue was 0.69 and, therefore, was not extracted. The principal component communalities were 0.690 for religious attendance, 0.629 for prayer frequency, 0.436 for biblical fundamentalism, and 0.687 for strength of affiliation. Thus, the majority of the variance in these variables is accounted for by a one-component solution, although variance in biblical fundamentalism is not accounted for as strongly as the other variables. Nevertheless, it is reasonable to consider these four variables as part of a single religiosity component. Therefore, a single religiosity variable was constructed by first calculating z-scores for each of the four GSS variables (since their scales of measurement are not uniform) and then calculating the mean of these z-scores. The four standardized variables exhibited strong internal consistency (Cronbach’s α = 0.79). For each valid case in the GSS data set, a single “religiosity” value was calculated. Unless otherwise stated, all subsequent analyses rely on this religiosity composite variable.

The number of original religiosity variables used to create the values for the composite variable varies by year, since some surveys from 1972 to 2012 only include a subset of the four variables. As a result, the religiosity composite variable is most strongly influenced by religious attendance (n = 55,821), followed by strength of affiliation (n = 51,436), prayer frequency (n = 27,816), and biblical fundamentalism (n = 27,618). This may also partially explain why biblical fundamentalism has the least amount of variance accounted for by the one-component solution.

Measure of Societal Insecurity. One variable from the GSS was used to assess participants’ perceptions of societal security. Specifically, participants were asked “Is there any area right around here—that is, within a mile—where you would be afraid to walk alone at night?” and responses were simply “yes” (coded as 1) or “no” (coded as 2). The GSS began asking this question in 1973, but it was not included in 1975, 1978, 1983, and 1986, resulting in 24 years for which data were collected over a 39-year period (valid n = 33,652; 253 cases of “Don’t know” or no answer). The data for this “fear” variable were reverse-coded in the present study to match the other variables assessing societal insecurity (described below), with higher values indicating a greater degree of insecurity.

Control Variables. The following control variables were included in a subset of the present study’s analyses because of their relationships with religiosity: sex, race, and age. These relationships were tested in the present study through a series of preliminary analyses (performed using SPSS version 21), which are described below.

Several studies indicate that women are, on average, more religious than men; women express a greater interest in religion (Sasaki, 1979), are more strongly committed to their religions (Bensen, Donahue, & Erickson, 1989), and report more frequent religious attendance (Batson, Schoenrade, & Ventis, 1993). These trends are consistent across denomination and type of religious belief system (Stark & Bainbridge, 1985). This relationship is also found in the 1972–2012 GSS data. An analysis of sex and the religiosity composite variable indicated that women are more religious (M = 0.14, SD = 0.82) than men [(M = −0.15, SD = 0.84), t(56,188) = −40.52, p < 0.001, mean difference = −0.29, 95 % CI [−0.30, −0.27], Cohen’s d = 0.34].

Although the relationship between race and religiosity has not been examined as extensively as sex and religiosity, researchers have found evidence of differences between African Americans and white Americans, with African Americans exhibiting greater religiosity (see Levin, Taylor, & Chatters, 1994). This relationship was tested in the 1972–2012 GSS data. A one-way Analysis of Variance indicated a difference in religiosity between races, F(2, 56,187) = 437.75, p < 0.001. However, the effect size is very small, η 2 = 0.02. Nevertheless, a post hoc Tukey test indicated that African Americans reported higher religiosity (M = 0.28, SD = 0.77, 95 % CI [0.27, 0.30]) than both white Americans (M = −0.03, SD = 0.85, 95 % CI [−0.04, −0.02]) and nonwhite others (M = −0.07, SD = 0.81, 95 % CI [−0.10, −0.04]).

Finally, several studies have found a relationship between age and religiosity in the United States (Bahr, 1970; Chaves, 1991; Firebaugh & Harley, 1991; Hout & Greeley, 1990), although there is debate as to what is driving this relationship (see Argue, Johnson, & White, 1999). An analysis of the 1972–2012 GSS data indicated a correlation between age and religiosity, r(56,040) = 0.18, p < 0.001. However, as with race, the effect size is quite small, r 2 = 0.03.

Uniform Crime Reports. In addition to the fear variable from the GSS, societal insecurity was assessed by measuring regional crime rates, as the frequency of crime in one’s surrounding area arguably influences one’s perception of societal insecurity. These crime rates were obtained from Crime in the United States (CIUS), an annual report published as part of the Federal Bureau of Investigation’s Uniform Crime Report program (UCR; FBI, 2014). Like the variables from the GSS, these data were ideal for the purposes of the current study because of the availability of data over a substantial time period, as well as the consistency of measurement over time. These data are available as early as 1930, but the present study only used data corresponding to the time period for which religiosity data from the GSS are available (i.e., 1972–2012).

The UCR divides reported crimes into two broad categories: violent crimes and property crimes. Violent crimes are defined in the UCR as offenses involving force or the threat of force, and this category consists of murder and nonnegligent manslaughter, forcible rape, robbery, and aggravated assault. Property crime consists of burglary, larceny–theft, motor vehicle theft, and arson, specifically when there is no force or threat of force against the victims. However, statistics regarding arson are not included in the UCR summary data of property crime because of limited participation by local law enforcement agencies, as well as variance in data collection procedures by agencies that do participate. Although the CIUS reports provide data on each type of crime listed above (except arson), the present study relied on data from the broader categories of violent crime and property crime.

To control for varying population sizes between states, the current study relied on crime rate data reported in CIUS, rather than the absolute number of crimes reported. For each state and each year, CIUS provides a violent crime rate and property crime rate that is calculated as the total number of crimes reported in each category divided by the total population. These data are presented as the rate per 100,000 inhabitants.

These data sets do not provide a perfect record of how many crimes are committed, as they are limited to those crimes discovered by or reported to law enforcement agencies. Therefore, changes in the values reported over time only partially describe changes in the actual frequency of crimes committed.

Data from the GSS are not provided by state, but rather by region. Therefore, prior to analysis, state-level data on violent crime and property crime were converted to region-level data by calculating means for the states corresponding to each region (e.g., for each year, violent crime data from New York, New Jersey, and Pennsylvania were converted to a single mean value for the “Middle Atlantic” region). These converted data were then added to the GSS data set.

Current Population Survey. Data regarding the percentage of the United States population in poverty, by state and by year, were obtained from The Annual Social and Economic Supplement (ASEC; United States Census Bureau, 2013) of the Current Population Survey (CPS). Poverty is calculated based on a series of dollar value thresholds, and these thresholds are determined by family size and the number of resident children under 18 years old (DeNavas-Walt, Proctor, & Smith, 2013). To be categorized as “in poverty,” the family’s total income (before taxes and tax credits) must be less than the applicable threshold, which is updated annually based on the Consumer Price Index to account for inflation. If the family’s total income is below the threshold, then every individual in the family is considered to be in poverty.

The ASEC provides data from 1980 to 2012 on the number of people in poverty in each state and the poverty rate for each state (i.e., the number of poor divided by the state’s population). The poverty rate is presented as the rate per 100,000 inhabitants. As with the crime data, the poverty rates were converted to region-level data by calculating mean rates for the states corresponding to each GSS region. These converted data were then added to the GSS data set.

Time Series Data Set. To test Hypothesis 3 of the present study, regarding societal insecurity predicting future levels of religiosity, a separate data set was constructed using several of the variables described above. In this data set, each “case” was a year, resulting in 41 cases representing the years 1972–2012. Violent crime rate, property crime rate, and poverty rate data were added to the data set by calculating the mean of the state values for each year. The neighborhood fear variable from the GSS was also added to this data set by calculating the mean responses for each year in the 1972–2012 GSS data set. The religiosity composite variable was added to this data set by calculating the mean values for each year in the 1972–2012 GSS data set.

Given the already limited number of cases available in this data set for time series analysis, the fact that there are several years in which the GSS has not been administered is a considerable problem. Therefore, missing values of neighborhood fear and religiosity in the time series data set were replaced by calculating the median of the nearest value above and below the missing value. For example, a religiosity value of 0.0457 was created for the year 1992 by calculating the median of the existing values from 1991 to 1993. This method of missing value replacement was chosen based on the assumption that values for these variables are unlikely to change substantially in the short term, from year to year, and therefore the missing value for any given year can be reasonably estimated as falling in between the values of neighboring years.

Results

Religiosity and Individual Perceptions of Societal Insecurity

To test Hypothesis 1, that individuals’ perceptions of societal insecurity are positively related to their reported religiosity, an independent sample t-test was conducted (using SPSS 21) to determine the effect of neighborhood fear (i.e., one’s fear of going outside at night in their neighborhood) on the religiosity composite variable. This analysis was conducted to take advantage of the greatest number of cases in the GSS data set, as the subsequent regression analyses described below, which use several variables, are limited by the number of cases with valid data for all variables. The t-test was significant, t(33,554) = −15.57, p < 0.001, mean difference = −0.15, 95 % CI [−0.16, −0.13], Cohen’s d = 0.17. Participants who indicated they are afraid to walk around their neighborhood at night had higher religiosity scores (M = 0.09, SD = 0.83) than those who were not afraid (M = −0.05, SD = 0.83). In other words, as hypothesized, higher perceptions of societal insecurity are related to higher levels of religiosity.

To further test Hypothesis 1, a series of chi-square tests of independence was conducted using SPSS 21 to investigate whether the relationship between the religiosity composite and fear would obtain for each of the four original religiosity variables from the GSS (religious attendance, prayer frequency, biblical fundamentalism, and strength of affiliation).

To facilitate the analysis of religious attendance and interpretation of the chi-square test results, the attendance categories “never,” “less than once a year,” and “about once or twice a year” were collapsed into a “low attendance” category, and the categories “nearly every week,” “every week,” and “several times a week” were collapsed into a “high attendance” category. Other categories of attendance were not included in the chi-square test, as they arguably represent “moderate” levels of religiosity.

The chi-square test of independence for religious attendance was significant, Χ 2(1, N = 23,681) = 101.51, p < 0.001. However, the effect size, as measured by Cramer’s V (φ c), was quite small, φ c = 0.07. Those who indicated that they are afraid to walk around their neighborhood at night were more likely than expected to report high levels of religious attendance (O = 4747, E = 4350; std. residual = 5.7), and they were less likely than expected to report low levels of religious attendance (O = 4496, E = 4873; std. residual = −5.4). Those who indicated that they are not afraid were more likely than expected to report low levels of religious attendance (O = 8017, E = 7640; std. residual = 4.3), and they were less likely than expected to report high levels of religious attendance (O = 6441, E = 6818; std. residual = −4.6). In short, security and religiosity are not independent of each other, and as hypothesized, higher perceptions of societal insecurity are related to higher levels of religious attendance.

To facilitate the analysis of prayer frequency and interpretation of the chi-square test results, the categories of “never,” “less than once a week,” and “once a week” were collapsed into a “low prayer frequency” category, and the categories of “several times a week,” “once a day,” and “several times a day” were collapsed into a “high prayer frequency” category.

The chi-square test of independence for prayer frequency was significant, Χ 2(1, N = 17,835) = 157.58, p < 0.001. Again, the effect size was small, φ c = 0.09. Those who indicated that they are afraid to walk around their neighborhood at night were more likely than expected to report high prayer frequency (O = 4939, E = 4567; std. residual = 5.5), and they were less likely than expected to report low prayer frequency (O = 1626, E = 1998; std. residual = −8.3). Those who indicated that they are not afraid were more likely than expected to report low prayer frequency (O = 3802, E = 3430; std. residual = 6.4), and they were less likely than expected to report high prayer frequency (O = 7468, E = 7840; std. residual = −4.2). In short, as hypothesized, higher perceptions of societal insecurity are related to higher prayer frequency.

Regarding biblical fundamentalism, it was not necessary to collapse any categories because responses were recorded in three categories: belief that the Bible is (1) the literal word of God, (2) the inspired word of God, or (3) a book of fables. The chi-square test of independence for biblical fundamentalism was significant, Χ 2(2, N = 17,550) = 78.89, p < 0.001. The effect size was small, φ c = 0.07. Those who indicated that they are afraid to walk around their neighborhood at night were more likely than expected to view the Bible as the literal word of God (O = 2462, E = 2204; std. residual = 5.5) and less likely than expected to view the Bible as a book of fables (O = 972, E = 1095; std. residual = −3.7). Those who indicated that they are not afraid to walk around their neighborhood at night were more likely than expected to view the Bible as a book of fables (O = 1975, E = 1852; std. residual = 2.9) and less likely than expected to view the Bible as the literal word of God (O = 3469, E = 3727; std, residual = −4.2). The “inspired word of God” category showed a significant difference for the “afraid” group, but no significant difference for the “unafraid” group. It is difficult to interpret the effects for this category because it arguably represents a moderate level of religiosity, but for the less ambiguous categories, as hypothesized, higher perceptions of societal insecurity are related to greater biblical fundamentalism.

To facilitate the analysis of strength of religious affiliation, the categories of “no religion” and “not very strong” were collapsed into a “low strength” category, and the “strong” category was unchanged and used as the “high strength” category. The “somewhat strong” category was not included in this analysis, as it represents a vague, volunteered response, and few participants belong to this category in the first place.

The chi-square test of independence for strength of affiliation was significant, Χ 2(1, N = 27,787) = 155.66, p < 0.001. The effect size was small, φ c = 0.07. Those who indicated that they are afraid to walk around their neighborhood at night were more likely than expected to report strong religious affiliations (O = 5110, E = 4608; std. residual = 7.4), and they were less likely than expected to report weak religious affiliations (O = 5766, E = 6268; std. residual = −6.3). Those who indicated that they were not afraid were more likely than expected to report weak religious affiliations (O = 10,247, E = 9745; std. residual = 5.1) and less likely than expected to report strong religious affiliations (O = 6664, E = 7166; std. residual = −5.9). In short, as hypothesized, higher perceptions of societal insecurity are related to stronger religious affiliations.

Does Societal Insecurity, at the Individual and Regional Level, Predict Religiosity?

To test Hypothesis 2, that both individual perceptions of societal insecurity and regional factors indicative of societal insecurity positively predict religiosity, multiple linear regression analyses were conducted using SPSS 21. Two regression models were computed to determine whether regional societal insecurity as measured by violent crime rate, property crime rate, and poverty rate, along with individual perceptions of societal insecurity as measured by the GSS fear variable, predict religiosity when controlling for age, sex, and race (dummy coded with “African American” and “other” entered into the model, and “white” omitted to serve as the reference category). Two models were computed to avoid issues of multicollinearity because of the large correlation between violent crime rate and property crime rate, r(27,526) = 0.58, p < 0.001.

Model 1 excluded property crime rate. The variables were entered into the model in two blocks: The first block (the partial model) included the control variables, and the second block (the full model) introduced the societal insecurity variables. This allowed for determining whether the inclusion of the societal insecurity variables contributed to the model fit. The partial model predicted and explained roughly 8 % of the variance in religiosity, adjusted R 2 = 0.081, SE = 0.79, F(4, 27,523) = 607.66, p < 0.001. All of the control variables contributed to the model, and their coefficients were all positive, which is consistent with the previous literature and preliminary analyses described in the Method section. The standardized coefficients (β) indicated that age was the strongest predictor (β = 0.185, p < 0.001). The next strongest predictor was sex (β = 0.168, p < 0.001), followed by the African American racial category (β = 0.133, p < 0.001), and, finally, the “other” racial category (β = 0.012, p = 0.046).

The full model was also significant and explained about 10 % of the variance in religiosity, adjusted R 2 = 0.097, SE = 0.78, F(7, 27,520) = 425.57, p < 0.001. Using a criterion of VIF values less than 10, there were no apparent issues of multicollinearity in the full model, as all VIFs were less than 1.2. The change in explained variance was significant, ΔR 2 = 0.017, ΔF(3, 27,520) = 168.04, p < 0.001, indicating that the collection of societal insecurity variables contributed to the model beyond the control variables. The control variables all remained significant predictors in the full model, and although their coefficients changed slightly, their relative strengths as predictors remained the same. Aside from the “other” racial category (β = 0.016, p = 0.005), all of the control variables were stronger predictors of religiosity than the societal insecurity variables. Among the societal insecurity variables, poverty rate was the strongest predictor (β = 0.119, p < 0.001), followed by violent crime rate (β = 0.027, p < 0.001) and neighborhood fear (β = 0.013, p = 0.027). As hypothesized, all of the coefficients for the societal insecurity variables were positive, indicating that as both individual perceptions of societal insecurity and regional factors indicative of societal insecurity increase, religiosity increases.

Model 2 excluded violent crime rate. The variables were again entered into the model in two blocks, following the same procedure as Model 1. As block 1 was identical to that of Model 1, the results of the partial model were identical and are not repeated here. The full model was significant and, like Model 1, explained about 10 % of the variance in religiosity, adjusted R 2 = 0.098, SE = 0.78, F(7, 27,520) = 425.96, p < 0.001. Using a criterion of VIF values less than 10, there were no apparent issues of multicollinearity in the full model, as all VIFs were less than 1.2. The change in explained variance was significant, ΔR 2 = 0.017, ΔF(3, 27,520) = 168.67, p < 0.001. The control variables all remained significant predictors in the full model, and although their coefficients changed slightly, their relative strengths as predictors remained the same. Aside from the “other” racial category (β = 0.017, p = 0.003), all of the control variables were stronger predictors of religiosity than the societal insecurity variables. Among the societal insecurity variables, poverty rate was again the strongest predictor (β = 0.122, p < 0.001), followed by property crime rate (β = 0.028, p < 0.001) and neighborhood fear (β = 0.013, p = 0.035). As hypothesized, all of the coefficients for the societal insecurity variables were positive, again indicating that as both individual perceptions of societal insecurity and regional factors indicative of societal insecurity increase, religiosity increases.

Does Societal Insecurity Predict Future Religiosity?

To test Hypothesis 3, that both perceptions of societal insecurity and factors indicative of societal insecurity can predict future levels of religiosity at the national level, a time series analysis was conducted. More specifically, this hypothesis was tested through the use of vector autoregression (VAR; Sims, 1980)—following the methodology of Solt et al. (2011)—using the statistical software program STATA 12.1.

The VAR equation is similar to that used in multiple linear regression. However, the VAR model consists of n equations, where n is the number of variables being investigated (Stock & Watson, 2001). In each equation, one of the variables takes a turn as the criterion and is predicted by lagged values of itself, all other variables, and a serially uncorrelated error term. Each equation is then estimated by ordinary least squares (OLS) regression. This is referred to as a standard, “reduced-form” VAR, which is the type of VAR used in the present study.

For these analyses, violent crime rate and property crime rate were combined into a “total crime rate” variable to avoid issues of multicollinearity, as running separate models with each type of crime is less feasible in this case because of the number of equations involved in a single model. The total crime rate variable was constructed by first standardizing the violent crime and property crime data in the time series data set into z-scores. The total crime variable was then computed as the mean of the two standardized variables.

Model selection criteria are used to determine the best lag length for the VAR model. The most common criteria used are the Akaike (AIC), Schwarz–Bayesian (BIC), and Hannan–Quinn (HQ) (Zivot & Wang, 2003). These criteria are produced in STATA after specifying the variables to be included in the VAR and inputting a set number of potential lags. The recommended procedure is to choose the lag for which the selection criteria values are minimized (Lütkepohl, 2005). For the present study, although a lag of 8 years was found to best minimize the selection criteria (AIC = 3.55, HQIC = 3.82, SBIC = 4.52), this model failed to pass the Lagrange multiplier (LM) test for autocorrelation, Χ 2(16, n = 25) = 36.28, p = 0.003. Despite collapsing violent and property crime into a single variable, there was still an issue of multicollinearity. The LM test output indicated that a lag of 10 years should be chosen. Although the selection criteria values were slightly larger (AIC = 3.99, HQIC = 4.24, SBIC = 4.98), this model passed the LM test, Χ 2(16, n = 23) = 17.82, p = 0.33.

These analyses relied on a separate time series data set, with 41 cases representing the years 1972–2012. However, with a lag of 10 years and poverty data only going back to 1980, the VAR model was limited to analyzing 23 cases, from 1990 to 2012. The variables entered into the VAR equations were total crime rate, poverty rate, neighborhood fear, and the religiosity composite. In this data set, these variables reflect values at the national level, as the values are means of state- or region-level data for each year. The VAR model can be illustrated as follows:

$$ \begin{array}{l}{\mathrm{Religiosity}}_t={a}_{10}+{a}_{11}{{\mathrm{Religiosity}}_t}_{-10}+{a}_{12}{{\mathrm{TotalCrime}}_t}_{-10}+{a}_{13}{{\mathrm{Poverty}}_t}_{-10}\;\\ {} +{a}_{14}{{\mathrm{Fear}}_t}_{-10}+{e}_{1t}\end{array} $$
$$ \begin{array}{l}{\mathrm{TotalCrime}}_t={a}_{20}+{a}_{21}{{\mathrm{Religiosity}}_t}_{-10}+{a}_{22}{{\mathrm{TotalCrime}}_t}_{-10}+{a}_{23}{{\mathrm{Poverty}}_t}_{-10}\;\\ {} +{a}_{24}{{\mathrm{Fear}}_t}_{-10}+{e}_{2t}\end{array} $$
$$ \begin{array}{l}{\mathrm{Poverty}}_t={a}_{30}+{a}_{31}{{\mathrm{Religiosity}}_t}_{-10}+{a}_{32}{{\mathrm{TotalCrime}}_t}_{-10}+{a}_{33}{{\mathrm{Poverty}}_t}_{-10}\;\\ {} +{a}_{34}{{\mathrm{Fear}}_t}_{-10}+{e}_{3t}\end{array} $$
$$ \begin{array}{l}{\mathrm{Fear}}_t={a}_{40}+{a}_{41}{{\mathrm{Religiosity}}_t}_{-10}+{a}_{42}{{\mathrm{TotalCrime}}_t}_{-10}+{a}_{43}{{\mathrm{Poverty}}_t}_{-10}\;\\ {} +{a}_{44}{{\mathrm{Fear}}_t}_{-10}+{e}_{4t}\end{array} $$

Because STATA does not provide standardized coefficients as part of the output for VAR analyses, prior to running the VAR model, the poverty rate, fear, and religiosity variables were transformed into z-scores. Thus, the unstandardized coefficients provided by STATA can be interpreted in the same way as standardized coefficients (i.e., how many SDs the criterion changes for every 1 SD change in the predictor). The total crime rate variable was not transformed as it already represents the mean values of two standardized variables and thus approximates a standardized variable itself (M = 0.00, SD = 0.97).

The results of the VAR model indicated that all four equations were significant predictors of their respective criteria, explaining 76 % of the variance in religiosity, 73 % of the variance in total crime, 72 % of the variance in poverty rate, and 55 % of the variance in neighborhood fear (all ps < 0.001). The primary portion of the model to consider for the present study is the equation for predicting religiosity. For this equation, the strongest predictor was neighborhood fear ( β = 0.46, p = 0.003), followed by poverty rate ( β = 0.30, p = 0.03). Both coefficients were positive, indicating that, as hypothesized, increases in fear and poverty predict an increase in future religiosity (specifically, religiosity 10 years later). However, counter to expectations, total crime rate did not predict future religiosity. Thus, the results provide only partial support for Hypothesis 3.

It is also worth noting from the other equations that religiosity is a significant predictor of future increases in crime (  β = 0.58, p = 0.001) and future increases in neighborhood fear ( β = 0.89, p = 0.003). These results were not hypothesized, yet they suggest the possibility of a feedback loop between religiosity and societal security, which is considered in more detail in the Discussion section.

Discussion

The present study tested three hypotheses derived from Norris and Inglehart’s (2004) Secure Society Theory of religiosity, with the goal of explaining variations in religiosity within the United States at both the individual and societal level. Hypothesis 1 was supported; individual perceptions of societal insecurity, as measured by reporting whether one was afraid to walk around their neighborhood at night, were positively related to religiosity. Individuals who reported that they were afraid were more religious than those reporting they were not afraid, as indicated by a religiosity composite score comprising religious attendance, prayer frequency, biblical fundamentalism, and strength of religious affiliation. This relationship remained when examining each religiosity measure separately. Hypothesis 2 was also supported, providing further evidence for the positive relationship between societal insecurity and religiosity. Neighborhood fear, violent crime rate, property crime rate, and poverty rate all predicted religiosity, even after controlling for the effects of each other and the variables of sex, age, and race.

Hypothesis 3 was partially supported, in that neighborhood fear and poverty rate (but not crime rate) predicted levels of religiosity 10 years later. As with the previous analyses, there was a positive relationship, indicating that increases in societal insecurity predict increases in future religiosity. Interestingly, religiosity was also positively related to future crime rate and neighborhood fear. These results were not predicted by Secure Society Theory, and it is unclear why increased religiosity would lead to increases in actual societal insecurity (i.e., higher crime rates). However, the relationship between religiosity and perceived societal insecurity (i.e., neighborhood fear) is more readily interpretable as indicative of a feedback loop, in which religiosity is both influenced by perceptions of insecurity and influences those perceptions. In other words, it may be that not only are people who perceive greater societal insecurity likely to be more religious, but people who are more religious are likely to perceive greater insecurity.

Given the overall support found for the hypotheses in the present study, what conclusions can be reached regarding the relationship between societal insecurity and religiosity? Overall, the apparent relationship is consistent with the Secure Society Theory. Perceptions of insecurity, whether assessed directly (through neighborhood fear) or indirectly (through regional crime rates and poverty rates, which are likely to influence the perceptions of people living in those regions), are positively related to religiosity: The less secure one perceives society to be, the more religious they are, now and 10 years later.

However, the present results do not allow one to make strong inferences regarding causality. The chi-square tests of independence indicated that religiosity and perceptions of societal insecurity are not independent of each other, but this does not mean that a causal relationship necessarily exists. Likewise, the multiple regression analyses indicated that societal insecurity variables uniquely predict religiosity, but their predictive power is merely an indication of the relationship between the variables, not an indication that the predictors cause changes in religiosity. The time series analysis provides the strongest evidence for causality in the present study by showing that some aspects of insecurity predict future levels of religiosity, but technically the results only indicate that the variables are related, as with the other multiple regression analyses performed earlier.

From an evolutionary perspective, the present study’s results indicate that the mechanisms (or a subset of the mechanisms) associated with religious beliefs and behaviors are sensitive to environmental input regarding societal insecurity, and these mechanisms respond to increased insecurity by strengthening religiosity. This is consistent with Norris and Inglehart’s (2004) proposed function of religion as a coping mechanism in the face of societal insecurity. However, this does not necessarily support the idea that religion is produced by a specialized psychological adaptation. A possible alternative explanation is that there are mechanisms designed to cope with unpredictable and unsafe environments by increasing one’s perception of control, as this may serve as a buffer against helplessness or negative affect (Case et al., 2004). This may occur regardless of the existence of religious beliefs; indeed, experimental manipulations of perceived control have been found to elicit increased superstitious behavior (Keinan, 2002) and illusory pattern perception (Whitson & Galinsky, 2008). However, given the existence of religious beliefs, particularly beliefs that provide meaning for events or circumstances that may otherwise seem unpredictable, religion in this context may be best viewed as an “exaptation,” serving as a form of compensatory control (Kay et al., 2010) even though religious beliefs and behaviors were not necessarily originally selected for this purpose.

This interpretation leads to the question: Why would the illusory perception of increased control be adaptive? Norris and Inglehart (2004) suggest that perceptions of control are useful in unpredictable and unsafe environments because “Individuals experiencing stress have a need for rigid, predictable rules. They need to be sure of what is going to happen because they are in danger—their margin for error is slender and they need maximum predictability” (p. 19). This does not explain what is going on, however, because the superstitions, perceptions of illusory patterns, and religious beliefs do not provide “maximum predictability.” They provide the illusion of increased control without actually increasing one’s control over events. If one’s margin for error is slim, it seems maladaptive to invent meaning behind events just to reduce negative affect. After all, natural selection does not care how happy or sad you are, and these feelings are only useful to the extent to which they motivate solutions to adaptive problems.

When dealing with threats in the natural environment, the most accurate interpretation of one’s control would be favored, as this would allow one to allocate limited resources appropriately. However, Kurzban (2010) offers an interesting evolutionary explanation for the phenomena described above, which are examples of what he calls “strategic ignorance.” The key to explaining why inaccurate information may sometimes be favored is the fact that humans deal with more than just the natural environment: We are an extremely social species, and we have spent much of our evolutionary history living in small groups in which we depended on each other for survival. It is important that others view you as being a valuable member of the group; therefore, it is important to persuade others that you are more valuable than you actually are. Rather than consciously lying, an effective method of persuasion is to believe the inaccurate information yourself. Although this explanation is speculative, there are a variety of phenomena that it can potentially explain (see Kurzban, 2010), including the illusory perception of control in unpredictable and unsafe environments.

Limitations of the Present Study

Although it is not technically a limitation of the present study, this section begins with addressing the small effect sizes obtained from the tests of Hypotheses 1 and 2, because these small effects may be partially explained by some of the limitations of this study. For example, one of the weaknesses of using archival data is that the data were collected without the present study’s hypotheses in mind. As a result, the variables used in the present study regarding societal insecurity were not ideal. The question assessing insecurity in the GSS (neighborhood fear) was very specific and, arguably, only assesses a small portion of overall perception of societal insecurity. Therefore, the effect of neighborhood fear on religiosity should not be misconstrued as the effect of overall perceptions of societal insecurity on religiosity. More comprehensive measures of perceived societal insecurity would provide a clearer picture of how this perception is related to religiosity, and would possibly result in a larger effect size.

Another limitation that may have reduced the effect sizes is the lack of state-level data from the GSS. Because the GSS participants were only identified as belonging to one of nine regions, the state-level data on crime and poverty rates obtained elsewhere had to be converted to the same nine regions for analysis. To the extent that crime and poverty affect one’s perceptions of insecurity, it is likely that the effect weakens the further away the crime and poverty are from the individual. Because some of the GSS regions contain as many as eight states, one should expect the crime and poverty data to only partially represent one’s perception of insecurity in their local environment. It is therefore encouraging that, despite this lack of precision, the insecurity variables in the present study were all significantly related to religiosity in the hypothesized direction, and it is likely that data more specific to one’s local environment would produce a larger effect size.

Although the effect sizes identified in the time series analysis were larger, there was a limitation to this portion of the study as well, namely, the restricted number of observations. Although the data used in the present study span several decades, the time lag of 10 years used in the vector autoregression model resulted in 23 observations. Therefore, the results of these analyses should be interpreted with caution, as they may not accurately describe the relationship between religiosity and insecurity over time. The results were promising in that they were, for the most part, in the hypothesized direction, but studies analyzing this relationship over a longer period of time are necessary before making any strong conclusions.

Future Directions

The results of the present study support the hypothesis that perceptions of societal insecurity influence religiosity, which serves as a useful foundation for future studies to investigate the possible causal nature of this relationship. Several possibilities for building on the present results are described below.

The present study relied on measures that arguably assess one’s perceptions of societal insecurity. An important next step for future studies would be to develop a more direct measure of this perception. By constructing a scale that is specifically focused on measuring perceptions of societal insecurity, researchers could assess the relationship between these perceptions and religiosity more accurately. Additionally, assuming such a scale is constructed and validated, researchers could design experiments to test whether these perceptions cause changes in religiosity. For example, researchers could prime participants to have an increased perception of insecurity, perhaps by exposing them to either real or fabricated news stories regarding crime in their local environment. The effectiveness of the prime could be assessed by measuring differences between groups on the societal insecurity scale, and researchers could then investigate whether such primes lead to an increase in reported religiosity relative to participants who are primed to have a decreased perception of insecurity.

Similarly, given the possible feedback loop between religiosity and perceptions of insecurity suggested by the results of the time series analysis, researchers could prime participants’ religiosity and subsequently have them complete the societal insecurity scale. This would allow one to test for causality in the opposite direction—i.e., to test for an increase in perceived insecurity relative to participants who are not primed with religious ideas and concepts.

Finally, future studies can investigate the possible benefits of increased religiosity as a result of increased societal insecurity. Assuming an effect on religiosity is found when priming societal insecurity, one could investigate whether this effect is strengthened in a group setting. If Kurzban’s (2010) argument regarding strategic ignorance applies to this phenomenon (i.e., increased religiosity leads to increased perceptions of control, which makes one appear more valuable to the group), one could hypothesize that the presence of others may strengthen this effect. This line of reasoning could also be extended to investigating effects in the context of cooperative/competitive games; perhaps people are more willing to cooperate with individuals who react more strongly with compensatory control in the face of insecurity.

Conclusion

This chapter presented evidence consistent with the Secure Society Theory of religiosity. There appears to be a relationship between societal insecurity and religiosity in the United States over the last 40 years. Although the nature of this relationship is still unclear, the present study builds on previous research by illustrating that this relationship is not driven solely by societal conditions, but by individuals’ perceptions of societal insecurity. Given the number of studies that now reliably demonstrate a relationship between societal indicators of insecurity and religiosity, it is important to move forward by more directly assessing perceptions of societal insecurity at the individual level. The present study serves as a useful foundation for moving research forward in this way. Although religiosity is influenced by a variety of factors, further investigation of perceptions of societal insecurity may provide insights regarding variations in religiosity at the individual and societal level.