Disaffiliation from the Evangelical Lutheran Churches in the Nordic Countries: Mistrust in the Churches, Changing Worldviews, or Something else?

Data description

In this work, I used the data from the 2018 round of the ISSP Religion survey, which is the only one that includes the five Nordic countries (Norway participated in the 1991, 1998, 2008, and 2018 rounds, and Denmark and Sweden in the last three rounds only). I retrieved the data from the ZA7570_v2 data file (Group 2020), which is available from the GESIS Leibniz Institute for the Social Sciences repository (Leibniz-Institute-for-Social-Sciences, 2008). The data processing was done using R (R-Core-Team 2019).

Figure 1 shows a spine plot of the religious denomination in the Nordic countries, based on the ISSP Religion 2018 data. The two main groups are Protestants and the religiously unaffiliated. The other religious denominations have small shares, but there are some interesting points to note. For example, Finland has a minority of Christian Orthodox respondents (1.91%), reflecting the historical ties with Russia and the fact that the Orthodox Church of Finland is also considered a national church (alongside the Evangelical Lutheran Church of Finland). Iceland has a minority of “Other religion,” and upon closer inspection, it turns out that about 1.66% of respondents still believe in the old Nordic Gods. Finally, Norway has the largest share of Muslim respondents (1.63%) of the five countries.

Figure 1

Figures 2 and 3 show the percentages of respondents in each country that answered “No religion” as their religious group, and “I don’t believe in God” in the variable v20 “Please indicate which statement comes closest to express what you believe about God,” respectively, in the 2018 round of the ISSP Religion survey. Clearly, the Nordic countries are not among the countries with the largest shares of the religiously unaffiliated, but the Scandinavian countries (Denmark, Norway, and Sweden) rank among those with the largest percentages of unbelievers in God. Moreover, the percentages of people that do not believe in God are significantly lower in Finland and in Iceland than in the three Scandinavian countries. This difference suggests that one should be careful not to draw sweeping generalizations when discussing the low religiosity among the Nordics. It is also important to note that in some countries, the percentages of the religiously unaffiliated in the 2018 ISSP Religion survey sample differ significantly from those reported in other sources. For Denmark and Finland, the percentages are 23% and 31% respectively, and do not differ so much from the 25% reported in the Folkkirken website (Church-of-Denmark 2022) and the 27% reported in Statistics Finland (Statistics-Finland 2022) for 2018. However, the discrepancies are larger for Iceland, Norway, and Sweden. In the latter country, the ISSP percentage is about 30%, but the Svenska Kirken (Church-of-Sweden 2022) reports 41% for 2018. The origin of these discrepancies is unclear but is probably associated with the ISSP sample. It would therefore be advisable to seek confirmation of this work’s results using other data sources or future rounds of the ISSP Religion survey.

Figure 2

Figure 3

For the reasons stated in the Introduction, I removed the respondents affiliated with minority Protestant churches to avoid possible contamination of the results. For brevity, from here on, I will designate the two groups included in the analyses as ‘Protestant|Main’ and ‘None.’ In addition, I further restricted the sample to respondents that were raised as ‘Protestant’ or ‘None’ and whose parents also belonged to one of these groups. This was necessary to correctly define the predictors related to religious socialization in the logistic regression models of disaffiliation. Table 1 shows the frequencies and percentages of each denomination in the Nordic countries in the resulting sample.

Variables’ selection

The present study consisted of two steps. In the first step, I performed a characterization of the similarities and differences between the two groups (‘Protestant|Main’ and ‘None’) in the five Nordic countries using 15 variables related to religiosity, and the levels of confidence in two secular state institutions (the Parliament, and Courts and the legal system), as shown in Table 2. In the second step, I selected another set of variables related to confidence in churches, confidence in the Parliament and in Courts and the legal system, change of belief in God, attitudes toward the importance of religion, religious or nonreligious socialization, and demographic controls, as shown in Table 3, and used these variables as predictors of disaffiliation (response variable) in two logistic regression models (one without and another with the country as a control variable).

I first analyzed the group differences in the levels of confidence in Churches and the other two secular state variables, to determine whether the association between these levels was significant. Then, I used fifteen variables related to religiosity to perform the group comparisons of the average levels of religiosity. Notice that fifteen is the maximum number that can be represented in the Chernoff faces plots in the R function aplpack::faces (Wolf 2019). These variables are the ones in Table 2 except for confidence in the Parliament and in Courts and the legal system. They describe religious beliefs, confidence in churches, attitude toward the power of the churches and the political influence of religious leaders, relevance of religion, and religious involvement (frequency of attendance at church services, frequency of praying, and self-image as an (un)religious person). Here, the variables related to confidence in churches, the perceived power of churches and the influence of religious leaders describe the ‘judgmental side’ of ‘belonging’ (Field 2014; Hoffmann 1998; 2013; Kasselstrand, Couse, and Sanchez 2017; Lemos and Puga-Gonzalez 2021; Nicolet and Tresch, 2009).

To compute the logistic regression models in the second stage of the work, I first selected the respondents that were raised Protestant (RELRIN = ‘Protestant|Main’), which corresponded to a sample size N = 4565 (86.64% of the sample described in Table 1) and created a response variable Y with value = 0 if DENOMINATION = ‘Protestant|Main’ and = 1 if DENOMINATION = ‘None.’ Thus, Y = 1 for the respondents that disaffiliated from one of the National Churches.

The covariates in the logistic regression models (Table 3) were selected as follows:

• CHG.BELIEF.GOD describes whether the respondent changed his/her belief in God (never believed, used to believe but no longer do, did not believe but believes now, and always believed). The change of worldview regarding belief in God may be an important predictor of religious disaffiliation and is important for answering the second research question of the present work;

• CONFIDENCE.CHURCHES represents the ‘judgmental side of belonging’ and its inclusion relates directly to the second research question of the present work;

• CONFIDENCE.PARLIAMENT and CONFIDENCE.COURTS.LEGSYSTEM describe the influence of trust or mistrust in these two secular state institutions on disaffiliation. The results of the first stage of the work justified the inclusion of these predictors;

• The variables REL.RELEVANCE.TODAY and REL.REPRESENTS.PAST describe the strength of the association between the perception of (ir)relevance of religion in modern life and disaffiliation from the National Churches;

• The number of affiliated parents (NUM.AFFILIATED.PARENTS) and the frequency of attendance at church services during the formative period (ATTEND.11.12) describe how strongly the respondent’s type of socialization (a causal antecedent) may influence the decision to disaffiliate;

• The control variables SEX (gender), AGE.GROUP (age group), DEGREE (educational degree), and URBRURAL (type of living community) were included in the models because they were found significant in previous works related to the present study (Lemos and Puga-Gonzalez 2021);

• Finally, country was also included as a control variable in the second logistic regression model, to compare its effect with those of the other predictors.

Variables’ transformation

I transformed some of the selected covariates to compute the logistic regression models, as shown in Table 3. Such transformations consisted of merging or altering the categories in the original variables to ensure that the cell counts in the contingency tables Y-<covariate> were at least ten (Hosmer, Lemeshow, and Sturdivant 2013). For example, in CONFIDENCE.CHURCHES, I merged the categories “Complete confidence” and “A great deal of confidence” into ‘A great deal|Complete confidence.’ CONFIDENCE. PARLIAMENT and CONFIDENCE.COURTS.LEGSYSTEM were transformed likewise. The covariates ATTEND.11.12, MOTHER.ATTEND, FATHER.ATTEND, DEGREE, and URBRURAL were transformed in a similar way.

In other cases, the information in the original variable was changed to fit the logistic regression models’ purpose. For example, to specify the covariate CHG.BELIEF.GOD, I changed the original four categories in the ZA7570_v2 variable v21 “I don’t believe in God now and I never have,” “I don’t believe in God now, but I used to,” “I believe in God now, but I didn’t use to,” and “I believe in God now and I always have” into ‘No change,’ ‘Stopped,’ ‘Started,’ and ‘No change.’ Thus, the influence of belief changes in both directions could be compared with the base category of no change (which includes both firm believers and nonbelievers).

In the ZA7570_v2 data file, the respondent’s age is stored as an integer number. I discretized this variable using the UK Office of National Statistics Harmonized Principle 3 (standard = 2) and stored the result in the AGE.GROUP covariate. All the categories of the transformed covariates in the logistic regression models will appear below in their respective coefficient tables in the Results section.

Methods

I started the first part of the study by comparing the distributions of trust in Churches with trust in Parliament and the Courts and legal system in the five countries and for both groups. I began by representing the ten distributions in the form of bar graphs, and found (not surprisingly) that trust in Churches is much lower among the unaffiliated than among the Protestants in all five countries. Then, I compared the differences between Protestants and unaffiliated using the Wilcoxon-Mann-Whitney test for all three institutions, with the alternative hypothesis that the levels of trust are lower among the unaffiliated. I found statistically significant differences for trust in Parliament and the Courts and legal system only in Finland and Sweden. These results suggest that the relationship between religious affiliation and trust in institutions may still be significant, as suggested by previous studies (Kasselstrand and Eltanani 2013). Therefore, these two variables were included as predictors in the logistic regression models of disaffiliation.

For the qualitative group comparisons of religiosity, I used Chernoff faces plots (Chernoff 1973) based on the medians of the 15 religiosity variables shown in Table 2, using the R function aplpack::faces (Wolf 2019), now excluding confidence in the Parliament and confidence in Courts and the legal system.

The Chernoff faces plots are suggestive, but only provide qualitative comparisons. Thus, they are best complemented with multiple comparisons tests, which were done in two stages. First, I tested for significant differences in the medians in at least one group using the Kruskal-Wallis H-test (Kruskal and Wallis 1952), using the R function stats::kruskal.test (R-Core-Team 2019). Upon rejection of the null hypothesis of the H-test, I performed post hoc analyses to determine whether each group was different from each other group using the Conover-Iman test (Conover and Iman 1979) using the R function PMCMRplus::kwAllPairsConoverTest (Pohlert 2021), with Holm p-value adjustment (Holm 1979) for a conservative control over the familywise-error rate (Shaffer 1995; Wright 1992). For the visualization of results of the multiple comparisons, I represented the computed matrix of adjusted p-values as a set of comparison letters using the R-function mutlcompView::multcompLetters (Graves, Piepho, and Selzer 2019). In this representation, groups sharing a letter on a comparison variable are considered similar with respect to that variable. Because it would be tedious and hardly informative to perform multiple comparison tests for the 15 variables in the faces plots, I selected a subset of these that are representative of ‘believing,’ ‘belonging,’ and ‘involvement.’

In the second part of this work, dedicated to answering the second research question, I first computed a polychoric correlation matrix for the covariates in Table 3 (Fox 2016; Revelle 2022) and removed the variables that would pose multicollinearity problems according to the criterion described in Hair et al. (2009). According to this criterion, covariates with a correlation coefficient with absolute value greater or equal to 0.7 will create multicollinearity problems and should not be included simultaneously in a regression model. Finally, I computed two logistic regression models with the response variable Y as described above, one without and another with COUNTRY as control variable. I used the probit link and analyzed the coefficients’ relative importance (or effect size) in terms of the odds ratios and their 95% confidence intervals. For assessing the models’ fit, I used the deviance D, the number of degrees of freedom df, the Nagelkerke pseudo-R² (Nagelkerke 1991), and the Akaike information criterion (AIC).

Confidence in the churches, the parliament, and courts and the legal system

Figure 4 shows barplots of the distributions of confidence in churches and religious organizations, the Parliament, and in Courts and the legal system for the five Nordic countries.

Figure 4

Clearly, the difference of confidence in churches between the religiously unaffiliated and the Protestants is much larger than the corresponding differences for the Parliament and Courts and the legal system, in all the Nordic countries. The religiously unaffiliated tend to hold a negative attitude toward churches, as shown by the large percentages of their responses in the levels “No confidence at all” and “Very little confidence.” In Iceland and Sweden, even the percentages of them with “Some confidence” is quite low. This contrasts with the distributions of confidence in the Parliament and in Courts and the legal system, for which the difference between the two groups is far smaller in the five countries.

To analyze the statistical significance of the differences between the levels of confidence in the three institutions among the two groups, I applied the Wilcoxon-Mann-Whitney test for each institution and in the five countries separately, with the alternative hypothesis that the level of confidence is lower for the unaffiliated. The obtained results are as follows:

1. Confidence in Churches is lower among the unaffiliated in all five countries (p < 0.001 in all tests). This result was evident from Figure 4.

2. In Finland and Sweden, confidence in the Parliament is lower among the unaffiliated (p < 0.001 in both cases). In Denmark, Iceland, and Norway, the differences between Protestants and the unaffiliated are not statistically significant.

3. In Finland and Sweden also, confidence in the Courts and the legal system is lower among the unaffiliated (p = 0.002 and p = 0.02, respectively), with no significant differences in the other three countries.

Apart from group comparisons, it is also useful to evaluate the strength of the association between the confidence in the three institutions, for both groups and in the five countries.

Figure 5 shows the correlation matrices of the confidence in the three institutions (churches, Parliament, and Courts and the legal system) for the two groups in the five Nordic countries. In general, the correlations between confidence in the Churches and the two secular state institutions is weaker than the correlations between the latter two, but there is a moderate association with confidence in the Parliament and the Courts, particularly among the Protestants. Hence, while confidence in the Parliament and the Courts are not directly related to religiosity, they are potentially significant predictors of disaffiliation, and so were included in the logistic regression models of disaffiliation presented below.

Figure 5

Multiple group comparisons of religiosity variables

Figure 6 shows the Chernoff faces plots of the medians of the 15 variables described in Table 2 for the two groups of interest in the five Nordic countries. Table 4 shows the correspondence between the variables and the face features.

Figure 6

The differences between the ‘None’ and ‘Protestant|Main’ in each country are far more pronounced than the differences between these two groups across the countries in many aspects of individual religiosity. In other words, the ‘None’ are all very different from the ‘Protestant|Main.’ In particular, the two groups are very distinct in relation to the perception of the relevance of religion today and in the future (size of the face), confidence in the Churches and the perception of the Churches’ power in society (size of the eyes), beliefs in afterlife and miracles (size of the nose), and belief in God (hair’s style). Neither group looks completely uniform across the five countries. For example, the ‘None’ in Denmark and Iceland appear to be somewhat different than their counterparts in Finland, Norway, and Sweden. Likewise, some religiosity levels of the ‘Protestant|Main’ in Finland and Iceland are distinct from those in the three Scandinavian countries.

The two groups are very different, yet it is interesting to find some apparent similarities between them. For example, all the ‘None’ but also the ‘Protestant|Main’ in the three Scandinavian countries share the same average levels in the variable “To me, life is meaningful only because God exists” (style of the hair). In Denmark, both the ‘None’ and the ‘Protestant|Main’ have the same (low) average frequency of praying, which is like that of the ‘None’ in Iceland (width of the mouth). Perhaps the most striking similarity is that the structure of the face (attitude towards religious leaders not influencing how people vote) is the same for both groups in Denmark and Iceland, and for Protestants in Finland and Norway.

While the faces plots provide suggestive graphical multidimensional representations, the technique of multiple comparisons provides a more rigorous way of testing for group differences in one variable within a sample. I performed multiple comparisons as follows. First, I selected a subset of the initial 15 variables, namely BELIEF.GOD and BELIEF.AFTERLIFE (for ‘religious beliefs’), ATTEND and HOW.RELIGIOUS (for ‘involvement’), CONFIDENCE.CHURCHES and POWER.CHURCHES (for the ‘judgmental side’ of ‘belonging’), and finally REL.RELEVANCE.TODAY (an overall indicator of the importance of religion). Then, I performed Kruskal-Wallis H-tests for the selected variables, all of which led to the rejection of the null hypothesis of no statistically significant difference between the group distributions. Following the rejection of the H-tests, I performed post-hoc Conover-Iman multiple comparison tests with Holm p-value adjustment for each selected variable and the ten groups (country|group).

The results of the multiple comparisons showed that no ‘None’ group was similar to a ‘Protestant|Main’ group in any of the selected variables, in agreement with the qualitative comparisons provided by the Chernoff faces plots. Therefore, I ran a second batch of multiple comparison tests in which I tested the five ‘None’ groups separately from the five ‘Protestant|Main groups. This had the advantage of reducing the number of multiple comparisons from 45 to 20 (ten for the ‘Protestant|Main’ and ten for the ‘None’), which in turn reduced the familywise-error rate, increased the power of the tests, and facilitated the interpretation of the results.

The output of the Conover-Iman test is best interpreted by assigning letters to each group based on the matrix of adjusted p-values, such that groups sharing a letter in one variable can be considered similar in that variable. Table 5 shows the results of the multiple comparisons for the five ‘None’ and the five ‘Protestant|Main’ groups, for the variables mentioned above. The most salient results shown in this table are:

• The religiosity of the two groups is indeed very different in the five Nordic countries: there are no common letters between a ‘None’ and a ‘Protestant|Main’ group in any of the variables shown in Table 5.

• For the variables considered, the religiously unaffiliated seem to be more homogeneous than the Protestants affiliated with the National Churches, in the sense that the number of country differences that are significant (equal to the number of distinct letters across the countries for each variable) is generally smaller among the religiously unaffiliated.

Logistic regression models of disaffiliation

I built two logistic regression models of disaffiliation from the National Churches in the Nordic countries, one without the country as a control variable and another including it. Recall that in these models I used the subsample of the respondents who were raised as affiliated with the National Churches, with the response variable Y = 0 if the respondent was still affiliated and Y = 1 otherwise, and the predictors shown in Table 3 above.

Figure 7 shows a graphical representation of the correlation matrix of the ordinal predictors shown in Table 3 (SEX is dichotomous but can be assumed ordinal for computing correlations). Following the criterion in Hair et al. (2009), I removed the variables on the frequency of the parents’ attendance at church services (MOTHER.ATTEND and FATHER.ATTEND), for these would introduce multicollinearity problems with respect to the frequency of the respondent’s attendance during the formative period (ATTEND.11.12).

Figure 7

Table 6 shows the results of the logistic regression model without the country as a control variable. This table includes the model coefficients (B) and their 95% confidence intervals, the odds ratios OR = Exp(B) and their 95% confidence intervals, and the coefficients’ p-values. This table also includes information on the reference levels for each predictor, the number of complete observations used for computing the model, the deviance D, the number of degrees of freedom df, the AIC, and Nagelkerke’s pseudo-R² goodness-of-fit index.

The following guidelines are useful for the interpretation of this table:

• The odds ratios show how the odds of obtaining a positive response change for the corresponding predictor level relative to the predictor’s reference level. Odds ratios greater or lesser than one mean enhancement or inhibition of the response, respectively. For example, having no confidence at all in churches and religious organizations increases the odds of disaffiliation by nearly sixfold (Exp(B) = 5.995) relative to having some confidence, but having a great deal of confidence halves the odds (Exp(B) = 0.503).

• Coefficients with p < 0.05 are significantly different from zero. Thus, to compare the effect sizes (relative influence on the odds of a positive response) of different predictors, it is necessary to consider both the coefficients’ significance and how much Exp(B) or 1/Exp(B) (if Exp(B) < 1) of the statistically significant coefficients deviate from one.

• The Nagelkerke pseudo-R² (Nagelkerke 1991) is a goodness-of-fit index often used in logistic regression models as a counterpart of the determination coefficient R² (or adjusted R²) in multiple regression models (Pregibon 1981). A value of 0.310 can be considered as acceptable for social science studies. The AIC is an index that represents a trade-off between goodness-of-fit and parsimony (number of model coefficients) which is useful for comparing and selecting statistically equivalent models (Hosmer, Lemeshow, and Sturdivant 2013).

Using these guidelines and considering the p-values and odds ratios of the coefficients, the salient aspects in Table 6 can be summarized as follows:

• Confidence in churches is the strongest predictor of disaffiliation, for having no confidence at all in them increases the odds of disaffiliation nearly sixfold and having very little confidence by more than twice relative to the (relatively neutral) reference level of having some confidence. Moreover, and conversely, having complete confidence nearly halves the odds of disaffiliating.

• Having had one or both unaffiliated parents increases the odds of disaffiliating more than twice relative to the reference level (both parents affiliated with the National Churches). It is important to notice that this effect is far more significant for just one unaffiliated parent than for both parents unaffiliated.

• The frequency of church attendance during the formative period is also a significant predictor of disaffiliation. Never going to church or doing so less than yearly increases the odds of disaffiliating by a factor Exp(B) ≅ 1.67, while having attended at least once a month nearly halves the odds (Exp(B) ≅ 0.54) relative to the reference level.

• Ceasing to believe in God increases the odds of disaffiliating by a factor Exp(B) ≅ 1.61, but the coefficient is not significant for the reverse situation (p = 0.085). Hence, disbelief in God is a weaker predictor of disaffiliation than mistrust in the Churches or nonreligious socialization during the formative period.

• Confidence in the Parliament is a relatively weak predictor of disaffiliation. Having a great deal or total confidence in the Parliament increases the odds of disaffiliating by a factor Exp(B) ≅ 1.44, but for the other levels the effect is not significant relative to having some confidence. Confidence in Courts and the legal system is not a significant predictor of disaffiliation.

• As far as the perceptions about the relevance of religion are concerned, strong disagreement with the statement that religion is relevant today as it was in the past increases the odds of disaffiliating by a factor Exp(B) ≅ 1.44 and strong disagreement that religion represents the past and not the future reduces the odds of disaffiliating by nearly two and one half (1/Exp(B) ≅ 2.49).

Considering now the socio-demographic variables, Table 6 shows that relative to the reference group 0–24 years the odds of disaffiliating increase between about 2 and 2.4 for the age groups 25–44, 45–64, and 65–74, and is not significant for the older group (+75) only. This result tells that the trend toward disaffiliation is significant in a wide age interval (25–74). Besides the age group, the other significant socio-demographic predictor of disaffiliation is living in a big city, which increases the odds of disaffiliating by a factor Exp(B) ≅ 1.36. In this model, gender and the educational degree do not come out as significant predictors of disaffiliation.

Table 7 shows the results of the logistic regression model that includes the country as a control variable. Using the same guidelines as before, this table shows that the introduction of the new control variable changed the relative importance of several coefficients. Nevertheless, the main conclusions obtained with the previous model remain essentially unchanged. Confidence in churches still comes on top in enhancing the odds of disaffiliating for a very negative attitude, or in inhibiting disaffiliation for a great deal or total confidence. However, in this second model, the country also becomes a (surprisingly) strong predictor of disaffiliation. More specifically, being Finn increases the odds of disaffiliating relative to Denmark (the reference country) by a factor Exp(B) ≅ 4.48 and being Norwegian or Swedish increases the odds by Exp(B) ≅ 2.54 and 2.64 respectively. For Iceland, the odds are still significantly higher than for Denmark (Exp(B) ≅ 1.54).

The large odds ratios associated with the country are best interpreted by considering a logistic regression model with COUNTRY as the single predictor (Y ~ COUNTRY), which would predict the odds or (equivalently) the probability of disaffiliation based only on the country’s proportion of people that did disaffiliate. This proportion reflects the cumulative effect of multiple path-dependent and country-specific factors over the recent years, in which the trends toward disaffiliation were significant but probably different across the countries (as shown by Lemos and Puga-Gonzalez (2021) for Scandinavia in the period 1998–2018). Nevertheless, the logistic regression model Y ~ COUNTRY only predicts 3.6% of pseudo-variance (using Nagelkerke’s index), whereas the model Y ~ CONFIDENCE.CHURCHES already predicts 22.4% of pseudo-variance! These and other aspects will be further discussed in the next section. In the model with the country as an additional control variable, gender comes out as a significant (albeit weak) predictor of disaffiliation, as the odds of women disaffiliating are less than one relative to men (Exp(B) ≅ 0.77).

I compared the two logistic regression models (Tables 6 and 7) using the likelihood ratio test and found that the model represented by Table 7 is statistically superior (p < 0.001).

Confidence in churches, the parliament, and courts and the legal system

Both logistic regression models showed that mistrust in the churches is clearly the strongest predictor of disaffiliation (of all the independent variables included in the models), while having a great deal or complete confidence in them significantly inhibits disaffiliation. This finding could be expected from the results of the group comparisons, because the distributions of confidence in churches of the two groups are markedly different. Moreover, a logistic regression model with confidence in churches and religious institutions already explains 22% of the pseudo-variance, while the model described in Table 7 predicts 34% of that pseudo-variance. The logistic regression models further show that the influence of confidence in the National Churches seems to by asymmetric, because rejection due to mistrust seems to be stronger than the binding effect of trust.

The results of the logistic regression models of disaffiliation confirm the recent findings by Lemos and Puga-Gonzalez (2021), who showed that confidence in churches is a key variable for explaining the dynamics of religion in Scandinavia, and probably in a wider context. It also lends support to the argument that confidence in a church is an important element of ‘belonging’ because of its judgmental and emotional dimension (Hoffman 1998; 2013; Nicolet and Tresch 2009), and needs to be considered in secularization studies. In the specific case of the Nordic countries, the results of the present study also confirm the viewpoints of Niemelä (2007) and Lüchau and Andersen (2012) that disappointment with the National Churches is a major cause of disaffiliation.

In contrast with confidence in churches, confidence in the Parliament is a weak predictor of disaffiliation, and confidence in Courts and the legal system is not significant for predicting disaffiliation. This result is not surprising, for the multiple comparisons on the confidence in institutions in the first part of the present work showed that the distributions of confidence in the Parliament were significantly different for Protestants and the unaffiliated in Finland and Sweden only, and confidence in courts and the legal system is not significantly different in any of the five Nordic countries. Although there might be a significant association between political and religious orientation, as shown by the correlation structure in Figure 7, that association is probably weak. Thus, the results of the present work do not lend much confirmation to the results by Kasselstrand and Eltanani (2013), who found evidence that people affiliated with the National Churches have higher trust in the state in a study on Denmark, Finland, Norway, and Sweden. However, the present work is based on different methodology, data, and time points, and is restricted to the two major groups (‘Protestant|Main’ and ‘None’). Thus, the results of the two studies cannot be directly compared. In summary, the relationship between religious (un)affiliation and confidence in state institutions in the Nordic countries remains a potential topic for further research.

The influence of religious or nonreligious socialization during the formative period

The importance of parental religiosity and participation for the intergenerational transmission of religiosity is well-known (Storm and Voas 2012). Lemos and Puga-Gonzalez (2021) showed that the conditions of religious or nonreligious socialization during the formative period significantly influence the level of belief in God in Scandinavia. The results in Tables 6 and 7 show that the conditions of parental socialization during the formative period also influence whether (or not) the Nordics later decide to disaffiliate from the National Churches. Recall that the reference levels for the two variables that describe the socialization conditions, NUM.AFFILIATED.PARENTS and ATTEND.11.12, were ‘2’ and ‘Yearly,’ respectively. Thus, we see that having had just one unaffiliated parent more than doubles the odds of disaffiliating. This result is consistent with the theory of Credibility Enhancing Displays (CREDs) (Heinrich 2009) because exposure to conflicting religious experiences between the parents (the most important role figures in the formative period) is conducive to breaking the bond with religion later in life (Langston, Speed, and Coleman 2020; Lanman 2012; Lanman and Burmester 2017). The statement by Norenzayan and Gervais in their article on the origins of religious disbelief (Norenzayan and Gervais 2013):

seems to also hold for disaffiliation in all the Nordic countries.

The logistic regression models showed that people that never went to church or did so less than yearly during the formative period have higher odds of disaffiliating, whereas people that attended church with monthly frequency have significantly lower odds of doing so. This result confirms the importance of ritual participation during the formative period for establishing psychological bonds with religion (Lawson and McCauley 1990; McCauley and Lawson 2002). The present results also suggest that even a low frequency of church attendance, probably on special occasions like Christmas or Easter, is enough to maintain significant bonds to the National Churches in the Nordic countries.

Ceasing to believe in God: the effect of changing worldviews?

Belief in God is of course a core element of religiosity, particularly in Christian-traditional cultures, and so we would expect that stopping to believe in God would be a strong predictor of disaffiliation. However, the results of the logistic regression models show that ceasing to believe in God increases the odds of disaffiliating, but by much less than having no confidence at all in the National Churches. Mistrust in churches is thus a much more powerful predictor of disaffiliation than disbelief in God.

Does starting to believe in God inhibit disaffiliation? The logistic regression models are inconclusive about this, because the corresponding coefficients are not statistically significant. This is probably due to the small proportion of the respondents that were raised as Protestants and did not believe in God, but acquired that belief later in life.

These results again confirm the viewpoints of Niemelä (2007) and Lüchau and Andersen (2012), according to which disaffiliation in the Nordic countries is mainly due to disappointment with the religious institutions and not to a loss of faith. But how can we interpret this fact? Intuitively, one may think that stopping to believe in God radically changes the way one fulfills the need for a coherent view of reality, thus demolishing a critical pillar of the whole edifice of religiosity, after which religion itself no longer makes sense.

I propose the following explanation. Affiliation to and confidence in a Church represent different but strongly associated forms of ‘belonging.’ The former is institutional and cultural (Demerath 2000), and the latter is judgmental (Hoffmann 1998; 2013; Nicolet and Tresch 2009). Thus, the association between variables representing two forms of the same dimension is stronger than the one between a variable representing a rejection of ‘believing’ and another representing rejection of ‘belonging.’

Notice that this interpretation does not imply or suggest that belief in God is less important than confidence in churches for studying religion in Nordic countries, because the causal paths leading to the rejection of belief in God (Norenzayan and Gervais 2013) and rejection of religious institutions (Field, 2014; Kasselstrand, Couse, and Sanchez 2017; Lüchau and Andersen, 2012) are probably distinct.

Perceptions about the relevance and timeliness of religion

People that disagree that religion is relevant today as it was in the past have higher odds of disaffiliating, but thinking that it is still as relevant today as it was in the past does not work as an inhibitor. On the other hand, disagreement with the statement that religion represents the past and not the future significantly inhibits (nearly halves) the odds of disaffiliation, but agreement with that statement has no significant effect. However, these two variables are not as influential as confidence in churches, or the conditions of religious socialization discussed above.

I propose the following interpretation of the results just mentioned. First, the two predictors in question (REL.RELEVANCE.TODAY and REL.REPRESENTS.PAST) describe relatively generic attitudes toward religion as a cultural and social phenomenon, but are not associated with any definite form of ‘believing,’ ‘belonging,’ or religious priming. Second, the effect of these two variables on the response seems to be asymmetric, in the sense that rejection of the statements in the questions’ labels (see Table 2) has a stronger impact than agreement. This type of asymmetry probably exists for other predictors. One plausible explanation for these asymmetries is that they are rooted in the very general two-system model of the mind’s working proposed by Kahneman and Tversky, according to which negativity consistently dominates positivity in people’s judgments (Kahneman 2011). This is also a plausible explanation for the asymmetric effect of trust and mistrust in the National Churches on disaffiliation.

Influence of the socio-demographic variables

The two socio-demographic variables with a strong influence on the odds of disaffiliating relative to their reference levels are the country and the age group. Living in a big city also increases the odds of disaffiliating. The results of the present work are not conclusive on whether the odds of disaffiliating are different for women and men, and (somewhat surprisingly) no evidence was found that the educational degree influences the odds of disaffiliating.