Measuring Nonreligion as Absence: Testing Various Approaches

Introduction

The nonreligious are on the rise in numerous countries around the world (Kasselstrand, Zuckerman and Cragun 2023). A growing body of research has explored the characteristics of the nonreligious (Hayes 2000; Hayes and Mcallister 1995; Thiessen and Wilkins-Laflamme 2020; Zuckerman 2010) and scholars are now trying to understand the substantive or positive content of nonreligion (Coleman and Jong, 2021; Lee 2015; van Mulukom et al. 2023). In this article, we are doing the opposite – we examine some of the most common measures of religiosity to determine how different measures affect who is categorized as nonreligious.

By “absence” measures we mean variables that are generally used to give an indication of whether or not someone is religious. For instance, someone can either have a religious affiliation or not have one. In effect, such a variable captures the “absence” of a religious affiliation. Likewise, an individual can regularly or semi-regularly attend religious services or they can never attend. This would be the “absence” of regular religious service attendance. Someone can hold a belief in a god or a higher power but people can also not hold such beliefs, which would be the “absence” of these religious beliefs. All of these measures privilege religiosity and situate nonreligion as effectively the “absence” of religion. We recognize that this is a problematic way to think about nonreligion as the focus is not on who nonreligious people are or what they believe or do but rather on what they are not and what they do NOT believe or do (Cragun and McCaffree 2021).

As a growing number of scholars are beginning to point out (Lee 2015; Taves, Asprem, and Ihm 2018; Strhan and Shillitoe 2025; Wildman and Shults 2024), focusing on what the nonreligious lack tells scholars very little about the nonreligious. It’s akin to describing the vast majority of Americans as people who rarely if ever eat snails or frog legs or scorpions. This is only marginally useful information in that it tells us what Americans do not eat but does not tell us much about what Americans do eat. Even so, we are unaware of any prior research that examines “absence” measures to determine: (a) what percentage of the population would be classified as nonreligious based on the measure used, (b) how these absence measures overlap, and (c) whether there is an absence measure that is a stronger predictor of other variables of interest than others.

Literature Review

Perhaps the most common way to delineate who is nonreligious is a single-item measure of religious affiliation (Cragun and Smith 2024; Smith et al. 2024). Numerous studies in several national contexts have used this approach either through surveys or from census data (Bruce and Voas 2023; Coutinho and Golgher 2014; Esquivel 2021; Rosa González and Cabrera 2023; Thiessen and Wilkins-Laflamme 2020; Voas and Chaves 2016). The primary drawback to using religious affiliation as the sole indicator of whether or not someone is nonreligious is that it does not provide gradation (Lee 2014). It is a dichotomous measure; either someone is religiously affiliated or they are not. As a result, there is no additional information provided on whether the individual has predilections or preferences for religious affiliation or is strongly opposed to organized religion. While this is a limitation of this measure, it is also one of the advantages – it makes classification as nonreligious quite straightforward – people fall into just one of two categories. It is easy to implement, but it is not entirely clear how well this overlaps with other measures that could be used to capture nonreligiosity (or secularity), and there is some evidence regarding its variability or liminality in longitudinal studies (Hout, 2017; Lim, McGregor and Putnam, 2010).

Other studies have used the absence of religious attendance to delineate who is or is not nonreligious (Putnam and Campbell 2012). This measure is also not without its problems. There are plenty of people who report a religious affiliation and retain many religious beliefs but who never attend religious services (Brenner 2011; Rosa González and Cabrera 2023). As Stolz et al. recently argued, for many people the first dimension of religiosity to decline as they secularize is behavior because behavior is costly (Stolz et al. 2025). It is much less costly in terms of time or resources to hold religious beliefs or to report a religious identity. As a result, it’s not uncommon for people to infrequently or never attend religious services but still retain a religious identity and religious beliefs. Of course, the inverse is also true – people can report no religious affiliation but still attend religious services. At the same time, frequency of religious service attendance has been questioned as a Christian-centric and American-contextual measure (Brenner 2016; Romero, 2016). In contrast to the dichotomous religious affiliation variable, however, frequency of religious attendance has the benefit of being a continuous variable, which would suggest that it may account for more variation in potential dependent variables. However, for the purposes of classifying individuals as nonreligious, frequency of religious service attendance can be a bit challenging. For instance, are only those individuals who never attend religious services the nonreligious or should those who attend once or twice a year also be included as nonreligious?

Some studies have focused on belief, denoting those who are nonbelievers as arguably the nonreligious (Blanes and Oustinova-Stjepanovic 2015; Taves 2020). Belief measures can be useful for delineating who is or is not religious, but they are also problematic for a number of reasons. Belief has been described as a core component of Protestant Christian religions and figures less prominently in Judaism and some Eastern religions (e.g., Hinduism; Asad 1993). Given that beliefs are specific to religions, using beliefs as a measure of (non)religiosity is likely to be problematic. Some scholars have suggested that the absence of beliefs may be a more universal measure that works in all cultural and religious contexts (i.e., a nonbeliever in India is similar to a nonbeliever in Thailand or Costa Rica; see Cragun et al. 2015). This argument suggests that using beliefs to measure religiosity may be problematic but using the absence of beliefs to measure nonreligiosity may be more defensible. Even so, there is still the problem of which beliefs are absent. The absence of belief in a monotheistic god may work well in predominantly or formerly Christian countries but will be highly problematic in Hindu or Buddhist cultures, among others (Xu 2024). Likewise, belief in an afterlife can be tricky to measure as afterlife beliefs vary by religion (Burris and Bailey 2009). Thus, while the absence of supernatural beliefs may indicate nonreligion, measuring the absence can be challenging. Additionally, how beliefs are measured can provide more or less statistical information if scholars try to capture the strength of belief rather than the existence or absence of a belief. Delineating who is or is not nonreligious, however, will ultimately require specifying whether it is exclusively those who do not hold a belief or whether it also includes those who are questioning a belief or are skeptical. If we consider belief in a monotheistic god, are only those who do not believe in the existence of such a god (i.e., atheists) nonreligious, or would agnostics – those who do not believe the existence of such an entity can be determined – also not religious? There are also other options, as many attempts to measure god belief include additional gradations, including believing in a higher power, believing sometimes, and believing but doubting. Where does one draw the line to specify who is or is not religious?

There are also studies that utilize the importance of religion and/or self-reported religiosity (Luo and Chen 2021) as indicators of religiosity, though very few studies use this as the only measure to indicate whether or not someone should be considered nonreligious (cf. Beyer et al. 2019). The limited use of self-reported religiosity by scholars to delineate who is or is not nonreligious is intriguing but also understandable. It is intriguing because self-reported religiosity varies among both the religiously affiliated (The Pew Forum on Religion 2009) and the religiously unaffiliated (Langston et al. 2017), allowing for variation in both groups. But that is also part of the challenge. Such measures usually fall along a gradient, giving them the same advantages and disadvantages as measures of attendance or strength of belief measures. On the strength side, they provide a continuum on which to measure someone and may be better at predicting other variables because of their greater mathematical richness. On the weakness side, it’s not clear where to divide self-reported religiosity to demarcate the religious and nonreligious. If people are asked how important religion is in their life, are they only considered nonreligious if they select the lowest possible value? This adds some fuzziness to classifying individuals as religious or nonreligious.

There are, of course, other approaches to measuring religiosity (Hill and Hood 1999) and nonreligion (Coleman and Jong 2021). Some scholars have created measures that combine variables to create a classification system. Putnam and Campbell (2012), for instance, combined religious engagement with self-identified religiosity to restrict the “religious” in their study to only those who regularly attended and reported a religious affiliation, grouping those who reported a religious affiliation but who didn’t frequently attend religious services with the nonreligious. Davie (2015) argued that a single measure of religiosity like frequency of attendance or religious affiliation is not adequate to understand the full picture of what is happening in a country and that other measures like vicarious religiosity are also necessary. We have intentionally opted not to attempt to recreate some of the more complex measures and scales and instead focus on just four single-item measures as these or similar items are available on large-scale surveys and, as a result, are widely used by scholars.

Our examination of the literature illustrates that there are numerous approaches to delineating who is nonreligious, including identity, belief, behavior, and self-reported religiosity. What prior research has not done is compare these measures to determine: (a) to what extent the measures overlap, (b) how the selection of one measure over another changes the percentage of people who are classified as nonreligious, and (c) whether one of the measures is a stronger predictor of outcomes. In this paper, we attempt to address all three of these issues using a novel dataset with data from eight countries. Our aim is not to provide a definitive answer as to the best measure to use as that will likely vary based on context and question availability but rather to suggest that there are strengths and weaknesses to the various approaches.

Methods

Data for this study come from a broader survey that asked questions about a variety of topics related to morals, values, and religious/nonreligious beliefs and practices. It was fielded in eight countries in the fall of 2023: Argentina, Australia, Brazil, Canada, Finland, Norway, the UK, and the USA (Beaman et al. 2024). These countries were selected for three reasons. First, we wanted variation by hemisphere, with countries in the northern and southern hemispheres. Second, we wanted variation in development, with some highly developed countries and some countries that were not as highly developed. Third, while we wanted variation, we intentionally chose countries that had a history of some form of Christianity to allow for comparisons across countries. This last selection criterion is also a limitation we discuss in our limitation section. We paid Dynata to collect the data. Dynata used its online panel to field the survey and used quota sampling to match the sample with the age, gender, and regional makeups of the countries. While our samples are not probability samples, they do align closely with the demographics of the countries. The final sample size in each country was roughly 1,010 individuals (N = 8,080 combined), though some individuals did not answer every question or selected “choose not to respond.” Missing values (<3% across variables) were addressed with multiple imputation by chained equations using the mice package (v3.17; van Buuren and Groothuis-Oudshoorn 2011) in R (v4.3.3; R Core Team 2024). Five variables were imputed under predictive mean matching (pmm) or multinomial logistic (polyreg) models, producing ten imputed data sets and 30 iterations per chain. Convergence diagnostics showed stable means and variances across iterations.

We examine four variables as possible indicators of nonreligiosity. Participants were asked about their religious affiliation with the following question, “Do you identify with any particular religion, religious tradition, and/or spiritual tradition? In other words, are you…”. Response options varied by country but were recoded to include the following categories: (1) Buddhist, (2) Christian, (3) Jewish, (4) Hindu, (5) Muslim, (6) Sikh, (7) Another religious tradition, and (8) no religion or none. For the analyses below, we dummy coded the responses into a binary variable with all those reporting a religious affiliation grouped together in contrast to those not reporting a religious affiliation (none).

Participants were asked about the frequency of their religious service attendance with the following question, “Apart from such special occasions as weddings, funerals and baptisms, how often nowadays do you attend religious services or meetings?” Response options included: (1) “Not at all in the past year,” (2) “Once or twice in the past year,” (3) “Less often than once a month, but several times in the past year,” (4) “Less often than once a week, but at least once a month,” and (5) “At least once a week.” Given the response options, the level of measurement of this variable is technically ordinal. Recognizing this, we examine this variable two ways. First, we use the option “Not at all in the past year” as an indicator of nonreligion. Second, we use the full set of responses when examining how well this variable predicts other variables.

We asked participants about their belief in a god as part of a series of questions that asked about beliefs. The primary question stem was “Do you believe in any of the following?” The specific item was then worded as follows, “A monotheistic (One) God.” Participants could select between three options, (1) “Yes,” (2) “No,” and (3) “Maybe, Not Sure.” For our analyses, we combined “Yes” and “Maybe, Not Sure” into one response and used just the “No” response to indicate nonreligion, turning this into a dummy coded variable (1 = “No”).

Participants were asked about their self-reported religiosity with the following question, “On a scale of 1 to 5, where 1 means not religious at all and 5 means very religious, how religious would you say you are?” Response options included: (1) “Not religious at all,” (2) “A little religious,” (3) “Somewhat religious,” (4) “Moderately religious,” and (5) “Very religious.” Like religious affiliation, we analyzed this variable two ways. First, we categorized those who reported “not religious at all” as nonreligious for some analyses. Second, we used the entire range of responses when examining the ability of this variable to predict other variables. Figure 1 depicts the four variables and illustrates the various measures we use with each in our analyses. Descriptive statistics for these four variables are included in Table A1 of the supplementary materials.

Figure 1

We also analyzed the ability of these four measures to predict scores on two variables. We wanted variables that nonreligiosity should clearly be related to. As a result, we selected two questions from the survey. The first asked participants to indicate their level of agreement with the following statement, “Religion is good for those who believe in it, but it’s not for me.” The second also asked participants to indicate their level of agreement but to a different statement, “Religion is bad and only leads to violence, hatred, prejudice, and delusion.” The response options included, (1) “Strongly Agree,” (2) “Somewhat Agree,” (3) “Neither agree nor disagree,” (4) “Somewhat Disagree,” (5) “Strongly Disagree,” and (6) “Don’t know/Can’t choose.” For our analyses below, the “Don’t know/Can’t choose” option was coded as missing but then imputed as indicated above.

Our initial analyses were largely descriptive and not inferential. We were interested in (1) how many people would be considered nonreligious based on each of the questions we examined and (2) to what extent do these measures overlap? We then used inferential statistics to determine how strongly related these measures of nonreligiosity are with other variables. For the inferential statistics we used regression and correlation. Ordinary Least Squares (OLS) regression was employed to examine the relationship between the independent and the dependent variables. Although the dependent variables are ordinal, previous studies in this area have also utilized OLS regression, allowing for comparison of results (Kim 2013). Additionally, the dependent variable has five categories, which is considered sufficient to approximate a continuous distribution. OLS was therefore deemed appropriate given the study’s focus on identifying linear relationships, ensuring the practical applicability of the findings, and illustrating the explanatory power of the different models. The correlation analyses were conducted to assess the strength and direction of the relationships between the variables, though we are not arguing that this is a causal relationship. The sample size is sufficiently large to mitigate potential biases that may arise from slight deviations from normality.

Results

The first question we attempted to address involved a simple descriptive analysis to determine what percentage of the population in each of the eight countries would be considered nonreligious based on each of the four measures. Figure 2 shows the results. There are some suggestive patterns in the data. For all of the countries, never attending religious services resulted in the highest percentages being considered nonreligious. Likewise, in all of the countries, the measure that resulted in the lowest percentage classified as nonreligious was those who reported a “1” on the self-reported religiosity measure (a.k.a. “Not Religious At All”). Separate analyses (Table A2 in the supplementary materials) that included both “1s” and “2s” on the self-reported religiosity measure increased the percentage nonreligious substantially such that it would result in one of the largest classifications of people as nonreligious – comparable to religious attendance.

Figure 2

The second question we wanted to address is what happens when we combine the measurements. In other words, if we assume that someone is only nonreligious if they meet all four criteria, how many people would be classified as nonreligious? We performed a descriptive step-by-step analysis overlapping the four single-item measures of (non)religiosity. Our starting point was whether or not the person was (non)religious based on their affiliation since this is the measure most often used in other studies. We then added the criteria of not believing in a monotheistic god, followed by never attending religious services, and finally those who self-reported not being religious at all. To provide a point of comparison, the same step-by-step aggregation of characteristics but on the other end of the spectrum was also calculated to determine how many people would be considered religious if they had to meet all four criteria. The results are shown in Figure 3.

Figure 3

The first set of bars on the left of Figure 3 show the results as we aggregate nonreligious absence measures. Using just affiliation, 44.1% of the individuals in the eight countries would be considered nonreligious. When we add the requirement that they do not believe in a monotheistic god, the percentage drops to 29.5%. This drop is in line with prior research illustrating that many unaffiliated individuals report believing in a god or higher power (Smith et al. 2024; Esquivel 2021; Wilkins-Laflamme 2023). Adding the requirement of never attending results in 25.1% being classified as nonreligious. Finally, if the requirement of self-identifying as not being religious at all is included, just 20.7% of our survey participants would be classified as nonreligious, meeting all four criteria. The second set of bars on the right of Figure 3 shows the results of a similar analysis but on the other end of the spectrum with religious individuals. Using just religious affiliation, 55.9% of our participants would be considered religious. That drops to 33.4% when we add the requirement that they believe in a monotheistic god. The next criterion – weekly religious service attendance – results in a dramatic reduction in the religious; just 8.6% would be classified as religious. If they have to meet all four criteria, including self-reporting as very religious, just 3.6% of our survey participants would be considered religious.

Although the patterns of decline in how many people would be classified as nonreligious as criteria are added are similar among all eight countries, there are some differences between countries. In most countries, the percentage of the population who are unaffiliated nonbelievers is between 18.5% (USA) and 45.2% (Finland), but in Brazil, only 9.4% of the population are unaffiliated individuals who do not believe in a god. While in most of the countries, the overlapping of the four items still results in a nonreligious population above 20%, the percentages are lower for Argentina (12.7%), USA (11.8%), and Brazil (5.0%). Brazil stands out as being somewhat unique in our dataset. The growth of the nonreligious has been slower than in the other contexts analyzed. Additionally, a sizable percentage of the unaffiliated are arguably ‘religiously unaffiliated with beliefs’ (sem religião com crenças; Ritz and Senra, 2022; Fernandes, 2018). Importantly, we are not arguing that large numbers of people actually have intrasubject religious congruence but rather agree with Chaves (2010) that the vast majority of people do not.

As another relatively straightforward test of overlap, we also correlated the four variables we are analyzing. However, with two of the variables – attendance and self-reported religiosity – we correlated the variables in two ways. With both of these ordinal variables, we ran a correlation with both a simple dummy code for whether they were nonreligious or not using our most stringent criteria, and we ran the correlation with the full variable. The results are shown in Table 1. The coefficients not in brackets or braces are Pearson correlation coefficients. When correlating two dichotomous variables, we also calculated phi coefficients, which are shown in brackets. When correlating a continuous variable with a dichotomous variable, we also calculated point-biserial correlations, which are shown in braces. The strongest correlations in the table are, not surprisingly, with the two ways of measuring a variable: never attending religious services is strongly correlated with the full measure of religious attendance (r = –0.798; point-biserial = .79) and identifying as not at all religious is strongly correlated with the full self-reported religiosity measure (r = –0.784; point-biserial = .784). The other correlations vary from a low of 0.319 between never attending religious services and not believing in a god to a high of 0.614 between the full frequency of religious service attendance measure and the full self-reported religiosity measure.

Our final test was to see how well each of these six variables – the four primary variables but with two of them measured in two different ways – predicted two outcome variables that should clearly be related to nonreligion. There are no other variables in these regression models (i.e., no control variables), just the independent (i.e., absence measures) and dependent/outcome variables. The two outcome variables are “Religion is good for those who believe in it, but it’s not for me,” and “Religion is bad and only leads to violence, hatred, prejudice, and delusion.” Readers should note that (1) is Strongly Agree and (5) is Strongly Disagree in these models. The results are shown in Table 2. For both variables, the full self-reported religiosity measure is the strongest predictor, accounting for 39.9% of the variation in the first variable and 21.7% of the variation of the second variable. The weakest variable is the dummy code for never attending religious services.

Discussion

Recognizing that scholars who study nonreligion and the nonreligious are turning away from absence measures and instead focusing on what the nonreligious do (Cragun and Smith 2024), believe (Taves et al. 2018; Schnell 2015), and value (Thiessen and Wilkins-Laflamme 2020), much of the existing research on the nonreligious uses measures of religiosity to classify individuals as either religious or nonreligious. We have labeled these efforts “absence measures.” Our paper had multiple goals. First, we wanted to determine how many people would be considered nonreligious based on the absence measure used. Second, we wanted to see to what extent the measures overlap. And third, we wanted to determine how well each of the measures predicted or correlated with other variables. Examining these questions will help other scholars as they consider which measures they want to use to classify individuals as religious or nonreligious.

Our first analysis indicated that more people would be classified as nonreligious using a single-item measure of religious attendance (i.e., those who never attend religious services) than with the other three measures – lack of a religious affiliation, not believing in a monotheistic deity, and identifying as not religious at all. In line with Stolz et al. (2025), it appears that the first dimension of religiosity to decline as a result of secularization is religious attendance. This is likely because it is more “costly” in terms of time and resources than belief, identity, or self-reported religiosity. Self-reported religiosity typically resulted in the lowest percentage classified as nonreligious, but only when using the most extreme value of “not religious at all.” As noted in our results (see Table A2), if both those who reported they were “not religious at all” and were “a little religious” (1 and 2) are categorized as nonreligious, the percentage rises to roughly that of religious attendance in most countries, suggesting that self-reported religiosity is likely tied closely to frequency of religious attendance, which is also illustrated in the correlations we performed.

Our second analysis examined what percentage of individuals would be considered nonreligious if they had to meet all four criteria – no religious affiliation, never attending religious services, not believing in a monotheistic god, and reporting they were not religious at all. Not surprisingly, as additional criteria are added, the percentage of nonreligious people decreases. But the same is also true for the religious – as one adds criteria to classify someone as religious, the percentage who meet all the criteria drops off precipitously. These analyses illustrate that scholars can manipulate/control the percentage of nonreligious individuals in a country or study based on the measures and criteria they use to classify someone as nonreligious (cf. Putnam and Campbell 2012 for a study that manipulates these groups). Our findings suggest that the decisions involved in classifying someone as religious or nonreligious need to be justified. Whether the justification is the limitation of available questions or a desire to follow some strict criteria, that justification should be clear and recognize that these decisions can dramatically alter who is and who is not included among the nonreligious and religious.

Our final set of analyses looked at the ability of each of the measures to predict some outcome variable. We are not arguing that our OLS models are causal, rather we wanted to examine the coefficients of determination to see how much of the variation they explain in the dependent variables. These analyses help illustrate the strength of variables that range along a continuum – attendance and self-reported religiosity – as contrasted with variables that are dichotomous – religious affiliation and belief in a monotheistic god. Variables that can be treated as interval-like are stronger predictors of other outcomes, likely because they are mathematically richer but also because they reflect the fuzziness that exists in the real world. As our earlier analyses examining the overlap of the variables illustrated, only a minority of individuals are (non)religious using all four criteria. The vast majority of people are somewhere in the middle of the two extremes – infrequently attending religious services, considering themselves somewhat religious, and maybe believing. This supports the idea that most humans are not logically consistent in their beliefs, behaviors, identities, and values (Evans 2018), or what Chaves termed “intrasubject religious congruence” (Chaves 2010). This is not a criticism of most humans but rather an important lesson for social scientists. We need to recognize that people fall along a continuum. As a result, continuous measures tend to have more explanatory power than do dichotomous variables (Ritchey 2008).

If researchers are interested in the explanatory power of nonreligiosity rather than classifying individuals as nonreligious, our results suggest they may want to give greater consideration to a measure of self-reported religiosity. Other disciplines, like health research and psychology of religion, have realized that self-reported measures can be accurate reflections of more objective measures (Idler and Benyamini 1997; Svob, et al. 2019; Abdel-Khalek, 2007). Individuals may be surprisingly good at reflecting their overall religiosity in a single-item, self-report measure of religiosity, just like they are at evaluating their overall health. Measures of self-reported religiosity are often included in surveys but do not appear to be widely used to classify individuals as nonreligious, perhaps because it is not clear what the cutoff point should be. Another added advantage of the self-reported religiosity measure is that it had the lowest non-response rates of any of the four variables we examined, suggesting participants were comfortable answering this question. Future research should also consider the ramifications of using different cutoff points on a self-reported religiosity item for classifying individuals as nonreligious.

What are the implications of our analyses and findings? We want to be very clear that we are not arguing that there is one right way to classify individuals as nonreligious using absence measures. To the contrary, how people are classified as nonreligious should be a careful decision that is based on the aims of the study and the available measures. And, importantly, every decision should be clearly and carefully justified and will always be context-dependent. As our study illustrates, it would be quite easy to increase or decrease the size of the nonreligious population by choosing to use one variable compared to another or by combining variables. Scholars need to be aware of the consequences of the decisions they make in this regard. Moreover, those evaluating other people’s scholarship should critically evaluate these decisions to make sure they are defensible.

There are some clear limitations to our study. First, readers will note that all eight of the countries we included in our study have a Christian history, even if some of the countries are no longer majority Christian. It is unclear whether these four measures would function the same in countries that do not have a history of being predominantly Christian but rather have been historically Jewish, Muslim, Buddhist, Hindu, or some other religion or religious culture. Scholars should consider replicating our analyses in such countries. Of particular interest to us and presumably other scholars would be to what extent results are similar in polytheistic contexts: Does belief in any god or not result in the same outcomes as belief in a monotheistic god? Additionally, as noted above, our data are matched quota samples and not true probability samples. While this method is widely utilized today, there are limitations to this approach (Kennedy et al. 2016). There is also a possibility that these measures of religiosity are subject to acquiescence bias (Hill and Roberts 2023). Survey participants are more likely to agree with positively-keyed items. Of these four measures, religious affiliation cannot be positively-keyed but could still suffer from social desirability bias. The other four measures, while not worded in such a fashion as to elicit agreement, may suffer from acquiescence bias. We did not include negatively-keyed versions of the same items to counteract this effect and encourage future researchers to consider such studies. Finally, there are a number of scale measures that combine a variety of questions to estimate religiosity (Hill and Hood 1999). There are also some scale measures to capture nonreligiosity (Coleman and Jong, 2021; Cragun et al. 2015; Schnell 2015). It is possible that these scale measures are more robust and better than the single-item measures we have examined. Future research should investigate this question.

Conclusion

While we are involved in several projects that focus on the substantive content of nonreligion, most research on the nonreligious utilizes some method to determine who should be considered nonreligious and who should not. In this article, we examine four common measures of (non)religiosity – religious affiliation, frequency of religious service attendance, belief in a monotheistic god, and self-reported religiosity – using data from representative surveys in eight countries. We find that each of these measures has strengths and weaknesses. Religious affiliation results in the highest estimates of the nonreligious while self-reported religiosity is the strongest predictor of the two dependent variables we examined. Our findings suggest that scholars can use any of these measures to classify individuals as nonreligious but they should always justify their decisions and be aware of the consequences of choosing one variable over another.

Additional Files

The additional files for this article can be found as follows:

Competing Interests

The authors have no competing interests to declare.