Is France exceptionally irreligious? A comparative test of the cohort replacement theory

Abstract

While scholars have often pointed to the fact that France might be an exceptionally irreligious country, this hypothesis has not yet been tested with longitudinal data; and nor have researchers tried to account for this alleged irreligiosity. The present article tries to fill this gap in the literature by comparing France to other Catholic countries in Western Europe. To do so, we use the Church Attendance and Religious change Pooled European (CARPE) dataset, which to date is the most extensive dataset of church affiliation and church attendance in European countries, as well as International Social Survey Program (ISSP) data that allow us to measure religious beliefs and make retrospective estimations as far back as the 1910s. We find that France shows significantly lower aggregate religiosity than other Catholic countries in Western Europe, although this difference has diminished in the last few decades; and this low level of religiosity cannot be explained by France-specific period effects operating in the last 40 years. Rather, secularization in France takes the form of cohort replacement, which has led to differences in religiosity between the countries being reproduced from one generation to the other. In other words, France is so irreligious because it began on the path of secular transition earlier or from a lower level than comparable countries.

Full text

1
Is France exceptionally irreligious? A comparative test of the cohort
replacement theory
Please cite from:
Stolz, Jörg, Ferruccio Biolcati-Rinaldi and Francesco Molteni. 2021. "Is France
Exceptionally Irreligious? A Comparative Test of the Cohort Replacement Theory ".
L’Année sociologique (71):337-67.
Jörg Stolz, Ferruccio Biolcati, Francesco Molteni
i
Abstract
While scholars have often pointed to the fact that France might be an exceptionally
irreligious country, this hypothesis has not yet been tested with longitudinal data; and nor
have researchers tried to account for this alleged irreligiosity. The present article tries to fill
this gap in the literature by comparing France to other Catholic countries in Western Europe.
To do so, we use the Church Attendance and Religious change Pooled European (CARPE)
dataset, which to date is the most extensive dataset of church affiliation and church
attendance in European countries, as well as International Social Survey Programme (ISSP)
data that allow us to measure religious beliefs and make retrospective estimations as far back
as the 1910s. We find that (1) France shows significantly lower aggregate religiosity than
other Catholic countries in Western Europe, although this difference has diminished in the
last few decades; and (2) this low level of religiosity cannot be explained by France-specific
period effects operating in the last 40 years. Rather, secularization in France takes the form of
cohort replacement, which has led to differences in religiosity between the countries being
reproduced from one generation to the other. In other words, France is so irreligious because
it began on the path of secular transition earlier or from a lower level than comparable
countries.
1. Introduction
For many observers, France, with its “laïcité à la française”, is a unique and special case of
how a country may regulate religion and religious diversity institutionally (Baubérot, 2004b,
Baubérot, 2014, Baubérot, 2017, Beckford, 2004, Portier, 2016, Willaime, 2009, Zuber,
2017). What is less widely known – although specialists have remarked on it several times –
is that France is also a particularly irreligious country (Bréchon, 2000, Dargent, 2010,
Stoetzel, 1983).
ii
This is the case even though France is traditionally Catholic, with only
small Protestant minorities (Fath, 2005, Fath and Willaime, 2011). The hypothesis that
France might be an especially irreligious country has not yet been tested systematically with
longitudinal data, and this is precisely what we intend to do in our article.
Recent research has shown that secularization – understood as a decline in religiosity – takes
the form of cohort replacement in many Western countries (Crockett and Voas, 2006, Voas
and Chaves, 2016).
iii
This means that it is not the case that individuals change their religiosity
during their journey through life, either because of lifecycle or period effects; rather, what
happens is that more religious cohorts are replaced by less religious cohorts.
iv
While cohort

2
replacement seems to be the major factor in most Western countries, we also find deviating
cases, where period effects clearly play a role and individuals become more secular over time
(e.g. New Zealand) (Voas and Chaves, 2016), or where countries succeed in either
accelerating or decelerating the secularization process (Stolz et al., 2020).
The goal of this article is to see whether France is indeed more irreligious than comparable
countries, and whether a possibly “exceptional” French irreligiosity also has “exceptional”
causes that deviate from the more normal mechanism of cohort replacement. What might
such an exceptional cause be? It has been suggested that, in the wake of new religious
movements and the arrival of Muslims in France since the 1960s, both the French
government and actors in civil society have reinterpreted “laïcité” in a particularly anti-
religious way (Baubérot, 2004b: 262f., Baubérot, 2014). This in turn could have had specific
effects on the general religiosity of individuals that differ from the situation in other
European countries. If, however, France’s secularization relies, as in other countries, on
cohort replacement, this would mean that French irreligiosity has to be explained by the fact
that France began on its secularization path earlier or from a lower level than other countries.
Building on and extending research especially by Bréchon (2000, 2014, 2018, 2019),
Lambert (1994, 1995), Willaime (1998), and Dargent (2010), this article therefore addresses
two interrelated questions:
(1) Is France really significantly less religious (measured in terms of church attendance,
belief in God, and religious affiliation) than traditionally Catholic countries in the West and
its neighbouring countries?
(2) Are the mechanisms of secularization (namely, cohort replacement) the same in France as
in comparable Western countries, or are there period or lifecycle effects specific to France?
Studying a seemingly "exceptional" case (Ermakoff, 2014), our paper is a contribution both
to the sociological literature on the religiosity in France, and to the theory of the secular
transition and the cohort replacement mechanism in general. Since France is often seen as
religiously "exceptional", the fact that its religious evolution is – as we show – perfectly
compatible with the framework of the secular transition considerably strengthens this theory.
To answer our questions, we draw on two datasets. First, the Church Attendance and
Religious change Pooled European (CARPE) dataset (Biolcati et al., 2020), which to date is
the most extensive harmonized dataset of church attendance in European countries, spanning
observations from 1973 to 2016. We will use the CARPE dataset to analyse trends in church
affiliation as well. The complete dataset is based on five survey programmes (Eurobarometer,
European Social Survey (ESS), European Value Survey (EVS), International Social Survey
Programme (ISSP), World Value Survey (WVS)), and combines 1665 surveys with
information on 1,784,825 individuals. It ranges from 1973 to 2016. Second, we also use a
harmonized ISSP dataset (with rounds from 1991, 1998, 2008, and 2018) that permits us to
analyze religious belief as well as the retrospective church attendance of mothers, fathers, and
respondents (when the latter were children), thus allowing us to observe aggregate church
attendance as far back as the 1910s.
v
2. Theory: France, secularization, and cohort replacement

3
2.1 The (ir)religiosity of France
That there might be a “French religious exceptionalism” in the way that the French state
regulates and perceives religion has often been noted both by French and non-French
scholars, and can be summed up in the formula “laïcité à la française” (Amir-Moazami, 2007,
Baubérot, 2017, Baudouin and Portier, 2001, Beckford, 2004, Bobineau, 2012, Hunter-Henin,
2012, Michel, 2010, Portier, 2016, Willaime, 2009). Very much open to debate, however, is
the extent to which French laïcité is really exceptional. A closer look at the history of church-
state developments in France and comparable countries also shows a surprising number of
similarities (Baubérot, 2013, Zuber, 2017).
A less well-known proposition is that France might also be an exceptionally irreligious
country. When looking at survey data, several sociological observers have pointed to the
relatively low level of French religiosity (Bréchon, 2018, Norris and Inglehart, 2012 (2004),
Pollack and Rosta, 2017, Stoetzel, 1983).
vi
Thus, Stoetzel (1983: 265), using EVS data,
comments on the fact that France is a country “peu religieux”, while Willaime (1998: 158),
drawing on EVS data from 1990, claims that “France belongs to the group of countries with
the highest percentages of ‘non-religious’ people”. The most explicit recent argument for a
“French religious exceptionalism” regarding aggregate religiosity has been made by Claude
Dargent (2010), who demonstrates with ESS data that France has an exceptionally low level
of religiosity compared to other European countries, and that the correlation between the left-
right variable and religiosity is especially high in France (together with Spain).
Obviously, if we want to show a French specificity, then we have to decide which other
countries we wish to compare France with. As is well known, comparative analysis is useful
when comparing units that are as similar as possible regarding possible confounding
variables. The most important dimension for our purposes is the confessional tradition of the
country. As we know, countries with a traditional Catholic majority seem to follow different
secularization trajectories than confessionally mixed or Protestant countries (Martin, 1978).
While Protestant state churches in majoritarian Protestant countries enter into a close union
with the state (e.g. Sweden), and while traditionally biconfessional countries tend towards
pillarization (e.g. Switzerland, the Netherlands), majoritarian Catholic countries are often
characterized by conflictual relationships between the state and the Catholic church (Lipset
and Rokkan, 1967, Martin, 1978, Martin, 1991). Since France is clearly a country with a
majoritarian Catholic tradition, we have to compare it to other countries in this group:
Belgium, Portugal, Spain, Italy, Luxembourg, Austria, and Ireland. Note that we also limit
ourselves here to Catholic countries in Western Europe.
While there have been a number of reports of specific French irreligiosity, there has so far
been no comprehensive test using longitudinal data, which means that we do not know either
when France began showing less aggregate religiosity or whether differences between France
and comparable countries have varied over time. This leads us to our first hypothesis:
Hypothesis 1: France has shown in the last few decades significantly less aggregate
religiosity than other Western and traditionally Catholic countries
2.2 Mechanisms of secularization
Research on secularization in Western countries has made much progress in the last two
decades (for overviews see De Graaf 2013, Stolz, 2020a). One of the central insights is that

4
there are a number of factors that seem to secularize society (such as increased education,
pluralism, secular competition, or welfare options), but that all of these seem to work almost
exclusively by hindering religious transmission from one generation to the next.
(for overviews see De Graaf 2013, Molteni, 2021, Stolz, 2020a)). This can be seen in the fact
that secularization in Western countries mainly takes the form of cohort replacement
(Crockett and Voas, 2006, Voas and Chaves, 2016, Voas and Crockett, 2005, Wolf, 2008).
On aggregate, cohorts remain quite stable in most Western countries regarding their religious
beliefs and practices. The shift towards secularization that we can see in such countries is,
then, mainly due not to the fact that individuals have lost their faith, but rather to the fact that
every new generation is a little less religious than the previous, and that older generations are
replaced by younger.
vii
Demonstrating cohort effects is a more difficult enterprise than one might expect. To make
this point clear, we have to make a short note on the famous lifecycle - cohort - period
problem (Ryder, 1965). Individual lifecycle effects occur when individuals change their
religiosity during their life course, for example by becoming more religious when they grow
older or have children, or when they lose their faith as a result of an incurable illness. Cohort
effects occur when changes in a given society mean that specific cohorts are affected in
specific ways, with these cohorts then taking their changed attributes with them through time.
For example, a specific cohort of draft-age men may be affected due to the experience of war.
We speak of birth-cohort effects, when cohorts differ in how they have been socialized.
Finally, period effects occur when all individuals in a society are affected in a similar way at
a given period of time (regardless of age). Thus, a general lockdown due to a virus is a period
effect since it may restrict the mobility of everybody in society. It is important for the
argument that we will make below to understand that both period and lifecycle effects
assume that individual religiosity changes during a person’s life course, while cohort effects
stress that religiosity is mainly a matter of socialization and therefore does not change during
the individual’s life. Lifecycle (or age), birth-year cohort, and period effects are logically
connected in that birthyear + age = period. This means that we cannot estimate the effects
independently of one another (Bell and Jones, 2014, Glenn, 1976). While these effects
(lifecycle, cohort, period) can never by separated from one another completely, researchers
may still gain a good sense of what is probably going on in their data by using subject-
specific theory and common sense.
viii
Our investigation into the causes of France's secularization now takes the form of a decision
between two mutually exclusive hypotheses. On the one hand, it may be that secularization in
France is similar to that in comparable countries in that it is a matter of constant cohort
replacement. If this is indeed the case, factors like pluralism, education, secular alternatives
etc. are hindering religious transmission in France just as in other western countries. The
amount of religiosity found in France at a given point t in time would then principally depend
on the amount of religiosity found in t-1. This leads to hypothesis 2(a):
Hypothesis 2(a) France’s secularization mainly takes the form of cohort replacement.
France shows less aggregate religiosity because it began from a lower
level in the early 20th century and has maintained its lead in the
secular transition through cohort replacement
On the other hand, France might be an "exception" regarding not just its level of
(ir)religiosity, but also the mechanism of secularization. In other words, it may be that we
find period effects in the last 40 years or so that are specific to France and that cannot be

5
observed in comparable countries. An obvious candidate for a possible cause of such a period
effect can be found the re-actualization of laïcité from the 1980s onward. While the
relationship between the Catholic church and the French state became less problematic
between the 1950s and 70s, the arrival of new religious movements and religious
pluralization in the 1980s and 1990s, and then the political resurgence of Islam that became
especially visible in the “affaire du foulard”, have led to a re-actualization and hardening of
the ideology of laïcité in France (Baubérot, 2017, Beckford, 2004, Portier, 2016, Zaretsky,
2016). Thus, Beckford (2004: 28) argues that, in recent decades "(...) the State's reluctance to
regard religious movements outside the mainstream as acceptable, or even permissible, is
hardening." Portier and Willaime (2021: 230) note that the paradigm of laïcité has changed:
"In the 1970s, it was thought that social harmony could emerge from the free play of
singularities; it is now asserted that the unity of a people presupposes regulation from above."
(Translation ours). And Zaretsky (2016) goes so far as to write: "For nearly a century, laïcité
worked well enough. It ensured public space for both those who believed — not just
Catholics and Protestants, but Jews as well — and those who did not. But with the 1980s and
1990s came a growing number of immigrants, most of whom were Muslim, from North
Africa. And so a different kind of conflict between the French state and established religion
began to take shape." This, according to him, led to an "aggressive, fundamentalist version of
laïcité" among many French intellectuals and politicians. Hence our hypothesis 2(b):
Hypothesis 2(b) France’s secularization has the form of France-specific period effects
(the "reactualization of laïcité"). These effects explain France’s
especially low aggregate religiosity
3. Method
3.1 Using the CARPE dataset to measure church attendance
The Church Attendance and Religious change Pooled European (CARPE) dataset is to date
the most extensive harmonized dataset of religious affiliation and church attendance in
European countries, spanning observations from 1973 to 2016. The complete dataset consists
of a harmonization of five survey programmes (Eurobarometer, ESS, EVS, ISSP, WVS), and
combines 1665 surveys with information on 1,784,825 individuals. It ranges from 1973 to
2016.
ix
The CARPE harmonization project does not include weights (Biolcati, Molteni,
Quandt and Vezzoni, 2020).
Using the CARPE dataset instead of only one of the different surveys that are part of CARPE
(e.g. EVS or ISSP) has a crucial advantage: by increasing the number of observations, we can
greatly increase the reliability of the estimates, especially when trying to measure change
over a large number of cohorts and years. This comes at a price, however. The CARPE
dataset only has church attendance and religious affiliation as its dependent variables, and
only includes a limited number of independent and control variables. When analyzing
religious belief and retrospective church attendance on the part of respondents and their
parents, we therefore supplement our analysis with ISSP data. We chose ISSP, a high-quality,
multi-round survey, for our complementary analysis, because it allows analysis not just of
current, but also of retrospective, religiosity.
x
We select in the CARPE dataset only the Catholic countries in Western Europe: France,
Austria, Belgium, Ireland, Italy, Luxembourg, Portugal, and Spain. We exclude individuals
with a non-Christian religion, since the items that we use are not adapted to non-Christians.

6
We select individuals in the age range of 20 to 90. This leaves us with a subset of 287
harmonized surveys and 452,760 individuals (see Table A1, Appendix).
Our first dependent variable is religious affiliation (yes/no), and our second is implied
probability of church attendance. The latter is calculated as the ratio of the number of weeks
that a person goes to a religious service per year, divided by the number of weeks in a year.
Thus, a person who claims to go to church on a monthly basis receives a score of 12/52 =
0.23, while a person who claims to go two to three times a year receives a score of 2.5/52 =
0.05.
xi
Implied probability of church attendance is an excellent way to harmonize measures of
church attendance across surveys (Table A1, Appendix) (Biolcati, Molteni, Quandt and
Vezzoni, 2020, Hout and Greeley, 1998).
Our first independent variable is survey year. The range of survey years differs for different
Catholic countries, and ranges from a minimum of 27 years (Portugal) to 43 years (Belgium,
Spain). For France, the range is 42 years. Our second independent variable is birthyear
cohort. We distinguish nine different birthyear cohorts, the first consisting of individuals born
up to, and including, 1920. Each following cohort spans 10 years, thus giving us a cohort
born in 1921-1930, a cohort born in 1931-1940, and up until a cohort born in 1991-2000. Our
third independent variable is country.
The descriptive information in Table A1 (Appendix) shows the mean church attendance of
the different countries, with France at the low end (0.154), and Ireland at the high end (0.586)
of the spectrum.
xii
The CARPE data show an overall mean age of 47.1 and an overall
percentage of women of 54.2%.
In the extremely rare cases where data are missing in the CARPE dataset, we use listwise
deletion. We control for the effect of the different studies (ESS, ISSP, etc.).
xiii
3.2 Using a harmonized ISSP dataset to measure church attendance and belief
We use a cumulated ISSP dataset harmonizing the rounds from 1991, 1998, 2008, and
2018.
xiv
While the ISSP dataset has fewer observations than the CARPE dataset, it has two
advantages. First, we can investigate religious beliefs (such as belief in God). Second, it has
retrospective questions about church attendance of respondents and their parents when the
former were children, which allows us to estimate rates of church attendance as far back as
the 1910s.
We select the Catholic countries in Western Europe that are available in this dataset: namely,
France, Austria, Ireland, Italy, Portugal, and Spain (Belgium and Luxembourg did not
participate). As in the CARPE dataset, we exclude individuals with a non-Christian religion.
We select individuals in the age range of 20 to 90. The dataset thus includes 19 surveys and
23,071 individuals (Table A2, Appendix). We use weighted data.
The dependent variable in our ISSP analysis is belief in God. The response categories were 1
= “I don’t believe in God”; 2 = “Don’t know whether there is a God, don’t believe there is a
way to find out”; 3 = “Don’t believe in a personal God, but I do believe in a Higher Power”;
4 = “I find myself believing in God some of the time, but not at others”; 5 = “While I have
doubts, I feel that I do believe in God”; 6 = “I know God really exists and have no doubts
about it”.
xv

7
Our other dependent variable is implied probability of church attendance of mother, father,
and respondent when the latter was a child. The values of this variable range from 0 to 1. The
wording of the question is: “When you were a child, how often did your mother attend
religious services?” We again transformed the nine response categories that ranged from 1 =
“Never” to 9 = “Several times a week” into implied probabilities as explained above.
As our main independent variable, we here use survey years. For Austria, Ireland, and Italy,
we have data for all four time points (1991, 1998, 2008, 2018); for France, Portugal, and
Spain, we only have data for 1998, 2008, and 2018. The second independent variable is
birthyear cohort. As with the CARPE dataset, we use birthyear cohorts with a ten-year span.
The oldest cohort comprises people born before, and including, 1920, while the youngest
cohort comprises those born from 1981 to 1990. Our third independent variable is country.
We used weighted data and imputed missing values.
xvi
We also used the ISSP dataset for robustness analyses, replicating the results for
affiliation/attendance and cohort while controlling for other variables
In the Appendix, we give some descriptive information on the different variables used in our
two datasets.
In a first step, we will present our results in the form of graphs with the help of LOESS-
smoothing (LOcally Estimated Scatterplot Smoothing), a method close, but not similar to,
moving averages.
xvii
The method uses locally weighted regression (Cleveland, 1979, Fox and
Weisberg, 2018). The basic idea of loess-smoothing is to create an estimate for every value of
the dependent variable with a regression that uses only a local subset of the y-values and
gives closer neighbours a higher weight than more distant neighbours. Changing the span of
the local neighbourhood that is considered will change the nature of the smoothing: larger
spans create smoother curves.
xviii
The advantage of using a smoother in contrast to a
parametric function (i.e. a linear or quadratic function) is that we do not have to specify a
function in advance. We are thus open to possible unexpected variability, such as created by
country-specific shocks. The advantage of LOESS compared to a simple moving average is
that it assigns more weight to closer neighbours than more distant neighbours in the window
of observation; moving averages, however, assign an equal weight to all observations in the
window of observation.
xix
The higher flexibility of LOESS comes with a price: it does not
produce a regression function that is easily represented by a mathematical formula and there
are no "coefficients" to report. We use LOESS here simply for descriptive purposes.
It is useful to show the importance of cohort replacement by means of graphs, but they should
be complemented with statistical models estimating the size of the effects. There is an
important discussion on how this should best be done. We use a method proposed by
Firebaugh (1989) that Voas/Chaves (2016) applied to the same problem.
Some authors have suggested using cross-classified models to solve the problem of the
logical interconnectedness of age effects, period effects, and cohort effects (Yang and Land,
2013), but the results of these methods have been strongly criticized (Bell and Jones, 2014,
Pelzer et al., 2015). We therefore adhere to the conventional method of linear decomposition
only of age and cohort proposed by Firebaugh (1989: 253) and used recently by Voas and
Chaves (2016). The method uses a simple OLS regression:
y = b0 + b1 survey year + b2 cohort + e

8
The goal is to partition the overall change in y into a part that is caused by the passage of
time and a part that is caused by birthyear cohort. In our formula, the coefficient b1 shows the
average change of y with every additional year over time (that can be attribute to either a
lifecycle or a period effect), controlling for cohort. The b2 coefficient, on the other hand,
shows the effect of an additional birthyear (a birthyear cohort effect), controlling for the
survey year in which the individual was interviewed. b0 is the intercept and e is noise. The
degree of change due to the two mechanisms can then be calculated as
Intracohort change = b1 (tT – t1)
Cohort replacement = b2 (CT – C1)
Where tT – t1 equals the range of survey years covered and CT denotes mean birth year at time
T, C1 denotes mean birth year at time 1.
The models were calculated with R, version 3.6.3. The syntax files that permit the replication
of all results can be obtained from the authors.
4. Results
4.1 Is France exceptionally irreligious?
Turning to our first question, we find that, when compared to other Catholic countries in
Western Europe, France has indeed shown very low aggregate religiosity in recent times.
We can see in Figure 1 that France’s aggregate percentage of affiliation to a religion is
significantly lower than it is in all other Catholic countries except Belgium. Belgium shows
an even lower percentage of affiliation, but the differences between France and Belgium are
slight and not significant. The percentage of religious affiliation diminishes over time in all
these countries, but at very different levels. While the affiliation rate in all other Catholic
countries except Belgium hovers above 80% and sometimes even reaches 90%, in France it
fell from around 70% in 1997 to around 50% in 2015.
Figure 2 shows that France has had significantly lower aggregate church attendance than all
other Catholic countries in Western Europe (including Belgium) since the early 1970s. As a
matter of fact, the difference was greater then than now, and seems to have diminished
somewhat. Thus, aggregate church attendance in France may have reached the lowest level
possible, with other Catholic countries in Western Europe “catching up” with this level.
France thus indeed has an “exceptionally” low aggregate level of church attendance
compared to other Catholic countries. Note that Ireland is another “exception”, in that it
began from a very high level with regard to the implied probability of church attendance, and
has witnessed in the last few decades a very strong drop; it nonetheless still has today the
highest aggregate level of church attendance of all Catholic countries in Western Europe.
Unfortunately, the CARPE data permit us to observe aggregate church attendance back only
to the 1970s. To look further back, we can use retrospective data created with ISSP data. In
the ISSP religion module, respondents are asked how often their mother, father, and they
themselves attended religious service “when they were a child”. Since respondents are from
different birth cohorts, their childhood took place in different decades, permitting us to

9
estimate the aggregate attendance of mothers, fathers, and children in different decades as far
back as the 1930s. To be able to do this, we attribute the levels of church attendance of the
mother, the father, and the respondent as a child to the year when the respondent was 10.
xx
In
Figure 3, we use this methodological tool to show that church attendance on the part of the
mother, the father, and the respondent herself when she was a child has been significantly
lower in France than in comparable Catholic countries in Western Europe since at least the
1910s. It also shows a slow, relatively steady, decline in church attendance all through the
20th century. One interesting fact that shows up is that church attendance seems to decline
rather steadily for the mothers and fathers, and relatively sharply somewhere in the 1960s for
the children, which we take as evidence of a “religious crisis in the 1960s” (Brown, 2001,
Gärtner, 2018, McLeod, 2007). This crisis shows up less clearly in the parents’ indicators,
because the parents of children of a certain age do not all belong to the same cohorts.
Figure 1 Percentage of religiously affiliated individuals in Catholic countries in
Western Europe
Note: CARPE data.

10
Figure 2 Mean church attendance in Catholic countries in Western Europe
Note: CARPE data.
Figure 3 Implied probability of church attendance of the mother, the father, and the
respondent when the respondent was a child, in Catholic countries in Western
Europe (retrospective data)
Note: Pooled ISSP data.

11
4.2 Is France an exception to the cohort-replacement mechanism?
Having established that France has a significantly lower level of religiosity than all other
Catholic countries in Western Europe (except Belgium in terms of religious affiliation), we
now turn to our second question regarding the main mechanism responsible for this low level.
Can secularization in France be ascribed mainly to cohort replacement, as it can be in the
case of most other Western countries?
One of the simplest but most effective ways of judging the importance of cohort replacement
is to examine the question by means of graphs and by looking at the variable aggregates of
interest in religion that different cohorts show over time. Figure 4 depicts the aggregate
religious affiliation of seven cohorts over time. For example, the likelihood that those
belonging to the cohort born between 1921 and 1930 were as children affiliated to a religion
was about 0.7, a figure that remained roughly the same over time until 2015 (0.68), when our
observation window for this cohort ended – because the members of this cohort had either
died or were too old to take part in the survey. To take another example, those belonging to
the cohort born in 1981-1990 appear in our dataset in 2003, and have an aggregate
probability of church attendance of about 0.3, a figure that seems to remain constant until the
end of our period of observation in 2016.
Figure 4 shows two things quite clearly. First, every new generation has a lower aggregate
level of religious affiliation than the previous. Much of the overall decline in religious
affiliation is therefore due simply to cohort replacement. Second, there seems to have been a
period effect somewhere around 2000 that caused a “fanning-out” of religious affiliation.
While older cohorts largely retained their affiliation, the aggregate level of religious
affiliation among the younger generations fell.
Figure 5 depicts church attendance. Again, we can see a very strong effect of cohorts. Every
new cohort starts out from a slightly lower level than the previous one, and on aggregate
stays at the same level over time. We can also see that the last four cohorts – those born in the
1950s and after – become increasingly close to each other over time. This might be
interpreted as a “bottoming-out” in the sense that church attendance might have reached a
plateau that it is not likely to go below in the future (Burkimsher, 2014). Note also that
France was already at a relatively low level in the 1970s, as the following comparisons will
show. It might be tempting to see a lifecycle effect in the data for the cohorts born in 1921-
1930 and 1931-1940, since their church attendance fluctuates somewhat. However, we
should be circumspect here, since these fluctuations are probably due to sampling errors.
The cohort effects are not just a matter of affiliation and attendance, but also show up in
almost all other indicators for Christian religiosity. To show this, we have to turn from our
harmonized CARPE dataset (that only included affiliation and attendance) to one of the
smaller original datasets. We chose ISSP. For lack of space, we can only show the effects in
Figure 6 when it comes to belief in God, but the same is true for practices such as prayer or
beliefs such as the belief in the holy status of the Bible (shown in the Appendix, Figures A1
and A2).
Demonstrating the cohort replacement effect only with ISSP data is useful for a second
reason: here, we have many more control variables. We can thus test whether or not the
cohort replacement mechanism is robust when we control for variables such as education,
sex, or political preference. Might it be that the cohort-replacement mechanism plays out

12
differently for different groups created by these variables? A test with only ISSP data for
France shows that this is not the case. Just as in other countries, cohort replacement with
regards to religiosity in France is a very robust phenomenon.
Figure 4 Religious affiliation in France according to survey year and cohort
Note: CARPE data (only ISSP and ESS).
Figure 5 Church attendance in France according to survey year and cohort

13
Note: CARPE data.
Figure 6 Belief in God in France according to survey year and cohort
Note: ISSP data.
How does France compare with other Catholic countries in Western Europe regarding the
cohort effect? From Figure 7, which depicts religious affiliation and cohorts over time, we
can see that most other Catholic countries show a cohort effect in that every younger cohort
shows a slightly lower aggregate level of religious affiliation than the previous, although the
effect is still rather small. It is strongest in Belgium, where we can clearly observe not just a
cohort effect (which would imply horizontal lines), but also a period effect (visible in
declining lines). Thus, in Belgium, individuals lose their affiliation even during their adult
lives. In several countries (Belgium, Ireland, Portugal), we can see again instances of the
“fanning-out effect” that we already observed for France. In other words, a situation in which
everybody has a religious affiliation turns into a situation where the cohorts are clearly
differentiated, with younger cohorts having a lower probability of religious affiliation.
Figure 8 depicts church attendance and cohorts over time in other Catholic countries in
Western Europe. The importance of the cohort effect is striking. The lines representing the
probability of implied church attendance are virtually straight in all countries, showing again
the expected order of cohorts. While the overall level of church attendance of the different
cohorts in different countries varies, the pattern of the cohorts is identical. Declining church
attendance in these countries since 1997 is clearly almost entirely a question of cohort
replacement. In comparison to France, we can see that the bottoming-out effect is not yet as
strongly advanced in most Catholic countries, but has clearly already started everywhere
except for Ireland.
Figure 7 Religious affiliation in other Catholic countries in Western Europe according
to survey year and cohort

14
Note: CARPE data. Lines are created with local regressions (“LOESS”).
Figure 8 Church attendance in other Catholic countries in Western Europe according to
survey year and cohort
Note: CARPE data. Lines are created with local regressions (“LOESS”).
Having graphically shown the importance of cohort replacement, we now turn to statistical
models estimating the size of the effects. Table 3 shows the results of a decomposition

15
approach as described in the methods section for church attendance in Catholic countries.
The aggregate change in church attendance is given in columns (5) and is partitioned into an
individual-change component in column (10) and a cohort-replacement component in column
(11). Ideally, columns (10) and (11) would add up to the aggregate change (5). As we can
see, this works relatively well in most cases, but there are deviations, suggesting some
nonlinearities in the data. For France, we can see an aggregate change of -20.0% in church
attendance, partitioned into a large cohort-replacement effect of -14.1% and a very small
individual-change component of -0.4%. It also means that the decrease in aggregate church
attendance in France is attributed wholly to cohort effects, and that there are certain
nonlinearities in the data, since our effects do not add up nicely to the overall aggregate
change. In Austria, an aggregate effect of -13.0% is partitioned into a cohort-replacement
effect of -14.6% and an individual-change effect of 0.5%. In all countries, the cohort-
replacement effect is stronger than the individual effect. However, the individual effect is
quite strong in Belgium, Ireland, Italy, and Luxembourg. Notably, in Italy we find a cohort
effect (-20.9) and an individual effect (+11.6) that do not add up to the aggregate change (-
20.0). It likely depends on the linear assumption of the decomposition model that is not met
by the Italian case. In fact, the overall decline in church attendance rate that we observe for
this country since the 1970s was broken by a temporary period of stability in the 1990s
(Vezzoni and Biolcati, 2015).Our results therefore show that declining church attendance in
some countries not only takes the form of cohort replacement, but may also be caused in part
by individual change.

16
Table 3 Decomposition of aggregate trends regarding church attendance: individual change and cohort replacement
1
2
3
4
5
6
7
8
9
10
11
Country
Start
year
End
year
Start
percent
End
percent
Aggre-
gate
change
Length
of
period
Difference
mean
birthyear
Within
slope
Between
slope
Individual
change
Cohort-
replace-
ment
change
France
1973
2015
28.0
8.0
-20.0
42
29.09161
-0.00009
-0.00486
-0.4
-14.1
Austria
1986
2016
32.0
19.0
-13.0
30
27.20271
0.00018
-0.00536
0.5
-14.6
Belgium
1973
2016
50.0
10.0
-40.0
43
37.13597
-0.00356
-0.00559
-15.3
-20.8
Ireland
1973
2014
96.0
46.0
-50.0
41
31.71093
-0.00505
-0.00825
-20.7
-26.1
Italy
1973
2013
55.0
35.0
-20.0
40
31.6942
0.00291
-0.0066
11.6
-20.9
Luxemb
ourg
1973
2010
52.0
21.0
-31.0
37
26.53746
-0.00372
-0.00624
-13.8
-16.6
Portugal
1988
2015
42.0
31.0
-12.0
27
16.66346
-0.00089
-0.00592
-2.4
-9.9
Spain
1981
2016
43.0
16.0
-27.0
35
25.75033
-0.00157
-0.00764
-5.5
-19.7

17
5. Discussion
In this article, we have asked two simple questions: To what extent is France less religious
than other Catholic countries in Western Europe when we take a longitudinal perspective?
And is France secularizing in a similar way to Western countries in general, i.e. by means of
cohort replacement? Our answers are straightforward. First, France is indeed “exceptional” in
the sense that it is significantly less religious in terms of aggregate church attendance, belief
in God, and religious belonging, than other Catholic countries in Western Europe (with the
exception of Belgium when it comes to religious affiliation). This difference is not new but
can be observed by means of retrospective data since the 1910s and with self-reports on items
measuring religious belief and practice since the 1970s. We can therefore verify hypothesis
1. One interesting fact with regard to aggregate church attendance is that the gap between the
other Catholic countries and France has narrowed in the last few decades, since irreligiosity
in France might be “bottoming out”.
As to our second question, the mechanism of secularization observable in France is not
exceptional, but very similar to almost all other Catholic countries in Western Europe (with
the possible exception of Ireland, and, to a lesser extent, Italy). France is losing its aggregate
religiosity mainly through cohort replacement. In other words, we cannot explain France’s
irreligiosity by pointing to specific causes linked to France in the last 40 years (such as
specific French policies or the specific enactment of the ideology of laïcité). If we assume
that the stable decline shown in our retrospective data is also caused by cohort replacement,
then we can even say that France’s secularization over the last 100 years or so has not been
exceptional. Thus, France’s low aggregate level of irreligiosity today should be explained by
the fact that France began the secular transition earlier or from a lower level than other
Catholic countries in Western Europe. We can therefore verify hypothesis 2a (cohort
replacement), but falsify hypothesis 2b (period effect).
Of course, this article has certain limitations. We have looked only at indicators of Christian
religiosity and have omitted an analysis of more individualized beliefs, alternative
spirituality, and non-Christian groups (Champion, 2001, Hervieu-Léger, 1999, Hervieu-
Léger, 2001). On the basis of research on religion both in France and in other Western
countries, however, we can say that non-Christian religions and alternative spirituality
certainly do not compensate for the strong secularization of Western countries (Bréchon,
2014, Stolz, Könemann, Schneuwly Purdie, Englberger and Krüggeler, 2016, Voas and
Bruce, 2007).
xxi
Our findings are also limited by the available data: our assertions about
church attendance before the 1970s are based only on retrospective data that may be more
prone to bias (for a discussion, see Franck and Iannaccone, 2014). Finally, while we have
shown that the current French irreligiosity is mainly a product of the past, it remains an open
question whether France started the secular transition earlier or from a lower level than other
countries (or both) .
xxii
One of the functions of the analysis of "critical" or "exceptional" cases in the social sciences
is to test whether a given theory allows to explain even these seemingly difficult instances of
observations. Scientific progress of a research programme often takes the form of researchers
showing that their theory is able to explain also what had hitherto been seen as "exceptions"
or "anomalies"(Lakatos, 1978). Alternatively, exceptional cases may reveal new facts and
relationships that force researchers to revise their theories (Ermakoff, 2014). In this paper, we
have shown that the "exceptional irreligiosity" of France is perfectly compatible with the

18
cohort replacement theory. Our article is another strong confirmation of the idea of a “secular
transition” (Brauer, 2018, Stolz, Pollack and De Graaf 2020, Voas, 2008), which claims that
countries or regions enter a secular transition (much like the demographic transition) and then
slowly secularize through cohort replacement. France is indeed very – and perhaps even
“exceptionally” – irreligious, but the roots of this phenomenon are not due to causes specific
only to France, at least not in the last century. Rather, the cause for the current differences in
religiosity are just transpositions – by cohort replacement – of similar differences in
religiosity at a higher level in the past. In this sense, France is a country whose exceptional
irreligiosity has non-exceptional causes.
Literature
Amir-Moazami, S. (2007).Laïcité à la française: Die verschiedenen Gesichter eines politisch-
sozialen Begriffs In Amir-Moazami, S. (Ed.), Politisierte Religion. Bielefeld: Transcript Verlag,
89-101.
Baubérot, J. (2004b). Laïcité 1905-2005, Entre Passion et Raison. Paris: Seuil.
Baubérot, J. (2013). Sécularisation, Laïcité, Laïcisation, ERES “Empan”, 90, 31-38.
Baubérot, J. (2014).Laïcité and freedom of conscience in pluricultural France In Berlinerblau,
J., Fainberg, S. and Nou, A. (Ed.), Secularism on the Edge. Rethinking Church Stat Relations in
the United States, France and Israel. Palgrave: MacMillan, 103-111.
Baubérot, J. (2017). Histoire de la laïcité en France. 7ème édition. Paris: PUF.
Baudouin, J. and Portier, P. (2001). La laïcité, une valeur d’aujourd’hui? Contestations et
renégociations du modèle français. Presses universitaires de Rennes.
Beckford, J. A. (2004).“Laïcité,” “Dystopia,” and the Reaction to New Religious Movements
in France In Richardson, J. T. (Ed.), Regulating Religion. Boston, MA: Springer, 27-40.
Bell, A. and Jones, K. (2014). Another “futile quest”? A simulation study of Yang and Land’s
Hierarchical Age-Period-Cohort model, Demographic Research, 30, 333-360.
Biolcati, F., Molteni, F., Quandt, M. and Vezzoni, C. (2020). Church Attendance and Religious
change Pooled European dataset (CARPE): a survey harmonization project for the
comparative analysis of long-term trends in individual religiosity, Quality & Quantity.
Bobineau, O. (2012). La spécificité du régime français de laïcité. Grilles de lecture à partir
des sciences humaines, Revue d’’éthique et de théologie morale, 269, 49-74.
Boulard, F. and Rémy, J. (1968). Pratique religieuse urbaine et régions culturelles. Paris: Ed.
ouvrières.
Brauer, S. (2018). The Surprising Predictable Decline of Religion in the United States, Journal
for the scientific study of religion, 57, 654-675.
Bréchon, P. (2000). Les attitudes religieuses en France: quelles recompositions en cours?,
Archives de sciences sociales des religions, 109, 11-30.
Bréchon, P. (2014).Le croire religieux des Français d’après les enquêtes ISSP In Aubin-
Boltanski, E., Lamine, A.-S. and Luca, N. (Ed.), Croire en actes. Distance, intensité ou excès?
Paris: L’Harmattan, 99-115.
Bréchon, P. (2018). La transmission des pratiques et croyances religieuses d'une génération
à l'autre, Revue de l'OFCE, Presses de Sciences Po, Aspects des transmissions familiales entre
générations, 2, 11-27.
Bréchon, P., Gonthier, F. and Astor, S. (2019). La France des valeurs. Quarante ans
d’évolutions. Grenoble: Presses universitaires de Grenoble.
Brown, C. G. (2001). The Death of Christian Britain. Understanding Secularisation 1800-2000.
London: Routledge.
Bruce, S. (2002). God is Dead: Secularization in the West. Oxford: Blackwell.

19
Burkimsher, M. (2014). Is Religious Attendance Bottoming Out? An Examination of Current
Trends Across Europe, Journal for the scientific study of religion, 53, 432-445.
Champion, F. (2001). Univers mystique-ésotérique et croyances parallèles, Futuribles, 260,
49-60.
Chenu, A. (2011). Les enquêteurs du dimanche. Revisiter les statistiques françaises de
pratique du catholicisme (1930-1980), Histoire & Mesure, XXVI, 177-221.
Cleveland, W. S. (1979). Robust Locally Weighted Regression and Smoothing Scatterplots,
Journal of the American Statistical Association, 74, 829-836.
Crockett, A. and Voas, D. (2006). Generations of Decline: Religious Change in 20th-Century
Britain, Journal for the scientific study of religion, 45, 567-584.
Dargent, C. (2010).La fille aînée de l’église dans l’isoloir. Religion et politique en France et
en Europe In Boy, D. (Ed.), Les Français, des Européens comme les autres? Paris: Presses de
Sciences Po, 181-214.
De Graaf , N. D. (2013).Secularization: Theoretical Controversies Generating Empirical
Research In Wittek, R., Nee, V. and Snijders, T. A. B. (Ed.), Handbook of Rational Choice
Social Research. Stanford: Stanford University Press, 322-354.
Dobbelaere, K. (2002). Secularization: An Analysis at Three Levels. Bruxelles: Peter Lang.
Ermakoff, I. (2014). Exceptional Cases: Epistemic Contributions and Normative Expectations,
European Journal of Sociology, 55, 223-243.
Fath, S. (2005). Du ghetto au réseau. Le protestantisme évangélique en France 1800-2005.
Genève: Labor et Fides.
Fath, S. and Willaime, -. J.-P. (2011). La nouvelle France protestante. Essor et recomposition
au XXIe siècle. Genève: Labor et Fides.
Firebaugh, G. (1989). Methods for Estimating Cohort Replacement Effects, Sociological
Methodology, 19, 243-262.
Fox, J. and Weisberg, S. (2018). Nonparametric Regression in R. An Appendix to An R
Companion to Applied Regression, third edition (PDF). SAGE. ISBN 978-1-5443-3645-9.
Franck, R. and Iannaccone, L. R. (2014). Religious decline in the 20th century West: Testing
alternative explanations, Public Choice, 159, 385-414.
Gärtner, C. (2018).Die normative Krise der 1960er Jahre: Der Wandel von Familie,
Geschlechterverhältnis und Sexualität In Gärtner, C., Gutmann, T., Mesch, W. and Meyer, T.
(Ed.), Normative Krisen. Tübingen: Mohr Siebeck, 161-193.
Glenn, N. D. (1976). Cohort Analysts´ Futile Quest: Statistical Attempts to Separate Age,
Period and Cohort Effects, American Sociological Review, 41, 900-904.
Hébel, P. and Recours, F. (2007). Effets d’âge et de génération: transformation du modèle
alimentaire, Cahiers de Nutrition et de Diététique, 42, 297-303.
Hervieu-Léger, D. (1999). Le pèlerin et le converti. Paris: Flammarion.
Hervieu-Léger, D. (2001). La religion en miettes ou la question des sectes. Paris: Calmann-
Lévy.
Hout, M. and Greeley, A. (1998). What church officials’ reports don’t show: Another look at
church attendance data, American Sociological Review, 63, 113-119.
Hunter-Henin, M. (2012). Why the French don’t like the burqa: laïcité, national identity and
religious freedom, The International and Comparative Law Quarterly, 61, 613-639.
Isambert, F.-A., Terrenoire, J.-P. and Boulard, F. (1980). Atlas de la pratique religieuse des
catholiques en France. Paris: CNRS.
Lakatos, I. (1978). The methodology of scientific research programmes. Philosophical Papers
Vol. 1. Cambridge: Cambridge University Press.

20
Lambert, Y. (1994).La religion: un paysage en profonde mutation In Riffaut, H. (Ed.), Les
Valeurs des Français. Paris: PUF, 123-162.
Lambert, Y. (1995). Vers une ère postchrétienne?, Futuribles, 200 Juillet-Août, 85-111.
Le Bras, G. (1955). Etudes de sociologie religieuse. Paris: PUF.
Le Bras, G. (1976). L’Eglise et le village. Paris: Flammarion.
Lipset, M. S. and Rokkan, S. (1967). Party Systems and Voter Alignments: Cross-National
Perspectives. The Free Press.
Maître, J. (1961). Les sondages sur les attitudes religieuses des Français, Revue française de
Sociologie, 2-1, 14-29.
Martin, D. (1978). A General Theory of Secularization. New York: Harper & Row.
Martin, D. (1991). The secularization issue: prospect and retrospect, British Journal of
Sociology, 42, 465-474.
McClendon, D. and Hackett, C. (2014). When people shed religious identity in Ireland and
Austria: Evidence from censuses, Demographic Research, 31, 1297-1310.
McLeod, H. (2000). Secularisation in Western Europe, 1848-1914. New York: St. Martin's
Press.
McLeod, H. (2007). The Religious Crisis of the 1960s. Oxford: Oxford University Press.
Michel, P. (2010). La laïcité française, Amérique Latine Histoire et Mémoire. Les Cahiers
ALHIM, 20, 1-10.
Molteni, F. (2021). A Need for Religion. Insecurity and Religiosity in the Contemporary World.
Leiden/Boston: Brill.
Norris, P. and Inglehart, R. (2012 (2004)). Sacred and Secular. Religion and Politics
Worldwide. Second edition. Cambridge: Cambridge University Press.
Pelletier, D. (2019). Les catholiques en France de 1789 à nos jours. Paris: Albin Michel.
Pelzer, B., Grotenhuis, M. t., Eisinga, R. and Schmidt-Catran, A. W. (2015). The Non-
uniqueness Property of the Intrinsic Estimator in APC Models, Demography, 52, 315-327.
Pollack, D. and Rosta, G. (2017). Religion and Modernity. An International Comparison.
Oxford: Oxford University Press.
Portier, P. (2016). L'Etat et les religions en France. Une sociologie historique de la laïcité.
Rennes: Presses Universitaires de Rennes.
Portier, P. and Willaime, J.-P. (2021). La religion dans la France contemporaine. Paris:
Armand Colin.
Ryder, N. B. (1965). The Cohort as a Concept in the Study of Social Change, American
Sociological Review, 30, 843-861.
Stoetzel, J. (1983). Les Valeurs du temps présent. Paris: PUF.
Stolz, J. (2020a). Secularization theories in the 21st century: Ideas, evidence, problems.
Presidential address, Social Compass, 67, 282-308.
Stolz, J., Könemann, J., Schneuwly Purdie, M., Englberger, T. and Krüggeler, M. (2016).
(Un)Believing in modern society. Religion, spirituality, and religious-secular competition.
London: Routledge.
Stolz, J., Pollack, D. and De Graaf , N. D. (2020). Can the State accelerate the Secular
Transitition? Secularization in East- and West Germany as a Natural Experiment, European
Sociological Review, 36.
Tocqueville, A. d. (1835/40 (1981)). De la démocratie en amérique. Paris: Garner-
Flammarion.

21
Vezzoni, C. and Biolcati, F. (2015). Church Attendance and Religious Change in Italy, 1968-
2010: A Multilevel Analysis of Pooled Datasets, Journal for the scientific study of religion, 54,
100-118.
Voas, D. (2008).The Continuing Secular Transition In Pollack, D. and Olson, D. V. A. (Ed.), The
Role of Religion in Modern Societies. New York: Routledge, 25-48.
Voas, D. and Bruce, S. (Eds.) (2007).The Spiritual Revolution: Another False Dawn for the
Sacred In Flanagan, K. and Jupp, P. C. (Ed.), A Sociology of Spirituality: Ashgate, 43-62.
Voas, D. and Chaves, M. (2016). Is the United States a Counterexample to the Secularization
Thesis?, American Journal of Sociology, 121, 1517-1556.
Voas, D. and Crockett, A. (2005). Religion in Britain: Neither Believing nor Belonging,
Sociology, 39, 11-28.
Willaime, -. J.-P. (2009). Pour une sociologie transnationale de la laïcité dans
l’ultramodernité contemporaine, Archives de sciences sociales des religions, 54e Année,
201-218.
Willaime, J.-P. (1998). Religious and Secular France Between Northern and Southern Europe,
Social Compass, 45, 155-174.
Wolf, C. (2008). How secularized is Germany? Cohort and Comparative Perspectives, Social
Compass, 55, 111-126.
Yang, Y. and Land, K. C. (2013). Age-period-cohort analysis: New models, methods, and
empirical applications. Boca Raton: CRC Press.
Zaretsky, R. (2016). How French Secularism Became Fundamentalist, Foreign Policy.
Zuber, V. (2017). La Laïcité en débat: au-delà des idées reçues. Le cavalier bleu.
Appendix

22
1. Descriptive information
Table A1 Descriptive information for CARPE dataset
Country
Affiliation
Mean(sd)
Attendance
Mean(sd)
Age
Mean(sd)
Female
Mean(sd)
Range survey years
Number of surveys
n
France
0.64( 0.48 )
0.12( 0.27 )
46.58( 17.71 )
0.53( 0.5 )
1973-2015
47
78,857
Austria
0.84( 0.36 )
0.24( 0.36 )
46.28( 17.21 )
0.55( 0.5 )
1986-2016
27
39,193
Belgium
0.64( 0.48 )
0.2( 0.37 )
44.89( 17.67 )
0.51( 0.5 )
1973-2016
34
57,287
Ireland
0.92( 0.26 )
0.69( 0.43 )
43.53( 17.6 )
0.53( 0.5 )
1973-2014
42
66,806
Italy
0.91( 0.28 )
0.41( 0.44 )
42.8( 16.92 )
0.53( 0.5 )
1973-2013
37
57,508
Luxembourg
0.84( 0.37 )
0.26( 0.39 )
42.85( 17.41 )
0.5( 0.5 )
1973-2010
23
19,235
Portugal
0.91( 0.29 )
0.37( 0.43 )
47.84( 18.89 )
0.57( 0.5 )
1988-2015
31
53,185
Spain
0.85( 0.36 )
0.26( 0.39 )
46.03( 18.54 )
0.53( 0.5 )
1981-2016
46
80,689
Total
0.81( 0.39 )
0.32( 0.42 )
45.3( 17.91 )
0.53( 0.5 )
1973-2016
287
452,760

23
Table A2 Descriptive information for ISSP dataset
Country
Belief
in
God
Mean(sd)
Church
attendance
mother
Mean(sd)
Church
attendance
father
Mean(sd)
Church
attendance
child
Mean(sd)
Age
Mean(sd)
Female
Mean(sd)
Range
survey
years
Number
of
surveys
n
Austria
4( 1.68 )
4.25( 1.61 )
3.79( 1.71
)
4.51( 1.45 )
47.59( 17.54 )
0.56( 0.5 )
1991-2018
4
3516
France
3.39( 1.83 )
3.36( 1.79 )
2.82( 1.75
)
4.01( 1.82 )
51.16( 17.23 )
0.54( 0.5 )
1998-2018
3
3959
Ireland
5.01( 1.32 )
5.71( 0.86 )
5.6( 1.04 )
5.71( 0.81 )
45.53( 17.42 )
0.53( 0.5 )
1991-2008
3
3878
Italy
4.76( 1.48 )
4.84( 1.49 )
3.98( 1.75
)
5.19( 1.23 )
48.17( 17.31 )
0.51( 0.5 )
1991-2018
4
3854
Portugal
5.05( 1.37 )
4.65( 1.56 )
4.07( 1.81
)
4.88( 1.5 )
47.16( 17.35 )
0.59( 0.49 )
1998-2008
2
1997
Spain
4.34( 1.77 )
4.39( 1.77 )
3.6( 1.92 )
4.61( 1.73 )
47.05( 18.04 )
0.51( 0.5 )
1998-2018
3
5867
Total
4.37( 1.71 )
4.51( 1.72 )
3.93( 1.9 )
4.8( 1.58 )
47.78( 17.62 )
0.53( 0.5 )
1998-2018
19
23,071

24
2. Prayer and belief in the sacred status of the Bible
Frequency of prayer was recoded as 4 = daily; 3 = regularly, but not daily; 2 =
a few times per year; 1 = never. Figure A1 shows the expected cohort effects.
Figure A1 Frequency of prayer in France according to survey year and cohorts
Note: ISSP data
Belief in the sacred status of the Bible was operationalized in the following way: 3 = the
Bible is the actual word of God, to be taken literally: 2 = the Bible is the inspired word of
God, and not everything should be taken literally; 1 = the Bible is an ancient book of legends
and moral precepts recorded by man. Figure A1 shows the usual cohort differences, although
they are quite small and the youngest generation shows higher values than would be
expected. In general, the French population is on average only prepared to see the Bible as a
text that should not be interpreted literally.

25
Figure A2 Belief in the sacred status of the Bible in France according to survey year and cohorts
Note: ISSP data
i
We thank Jean Baubérot, Pierre Bréchon, Claude Dargent, Christophe Monnot, Philippe Portier, and Jean-Paul
Willaime as well as the reviewers of L'Année sociologique for their very useful comments and suggestions. All
remaining errors are of course ours.
ii
Tocqueville (1835/40 (1981)) already observed that Americans went to church much more often than the
French when he travelled around the colonies.
iii
The term “secularization” can have different meanings and may be applied to different levels of social reality.
Obviously, it presupposes a certain definition of religion and religiosity (Bruce, 2002, De Graaf 2013,
Dobbelaere, 2002). In this article, we take secularization to mean simply a decline in religiosity. We define
religion as combining (1) an ideology referring to a transcendent (i.e. supernatural) reality, (2) a social group or
groups producing and transmitting this ideology, (3) the individual experiences, beliefs, and actions referring to
the ideology. Religiosity subsumes individual experiences, beliefs, and actions belonging to one or several
religion(s). Examples of individual religiosity as defined here include attending church service or a meditation
course, praying, going on a pilgrimage, and believing in angels (Stolz, 2020a).
iv
We use “cohort” as a shorthand for “birthyear cohort”. For this article, we use the terms "cohort" and
“generation” interchangeably
v
The high n makes measurement unprecedentedly precise even in a longitudinal perspective, leading to a
robustness of results that would not be possible otherwise. However, these advantages come at a price: namely,
the only indicator of religiosity that we can consider is church attendance, since this is the only dependent
variable that can be harmonized in such a wide range of surveys. Having acknowledged that, we think of course
that it is interesting to use church attendance to answer our questions, while leaving it to further studies to
replicate the analysis with further indicators, such as religious belief, importance of religion, and spirituality.
vi
Quantitative research on religiosity in France began earlier than international programmes such as EVS and
ISSP. There are first surveys on Catholics in the 1940s, and there is the well-known survey project by Gabriel le
Bras and Fernand Boulard (Boulard and Rémy, 1968, Isambert et al., 1980, Le Bras, 1955, Le Bras, 1976) that
began in 1945 and lasted until the end of the 1960s (Chenu, 2011, Maître, 1961).

26
vii
However, not all religious change takes the form of cohort replacement. Researchers have found a number of
exceptions. For one thing, cohort replacement does not explain all aspects of the decline in religiosity, but leaves
a more or less important place for individual “secular drift” (Stolz et al., 2016). This becomes very obvious in
some countries, such as New Zealand (Voas and Chaves, 2016). Furthermore, the state may be able either to
accelerate or decelerate secularization. For example, many individuals in East Germany were forced during the
socialist regime to disaffiliate from the church. Likewise, church scandals and church tax seem to have led
adults in Austria to disaffiliate at very different ages (McClendon and Hackett, 2014).
viii
There are researchers who believe that the APC problem may be solved statistically. One of the latest and
most well-known attempts is the cross-classified multi-level model by Yang and Land (2013). With Bell and
Jones (2014) however, we think that the search for such statistical solutions is a futile quest. For an application
of the APC model in a French context, see Hébel and Recours (2007).
ix
For the first article using the CARPE dataset applied to the Italian case, see Vezzoni and Biolcati-Rinaldi
(2015).
x
The ISSP data can be downloaded at http://w.issp.org/menu-top/home/.
xi
For mathematical reasons, individuals who go to church service every week are given 0.99 (instead of 1), and
individuals who never go to church are given 0.01 (instead of 0) (Biolcati, Molteni, Quandt and Vezzoni, 2020).
xii
These means should not be directly compared yet, since they may refer to somewhat different time periods
depending on the countries. The graphs and multivariate analyses below will allow for a useful comparison.
xiii
We imputed 8.3% of values in the attendance of the father, 6.0% in the attendance of the mother, and 2.0% in
the attendance of the child.
xiv
The data can be downloaded at https://www.gesis.org/issp/modules/issp-modules-by-topic/religion. The
harmonized dataset that also included the data from 2018 was not yet available when this article was written.
See, for specificities, the “Guide for the ISSP ‘Religion’ cumulation of the years 1991, 1998 and 2008 (ZA5070
and ZA5071)”, downloadable at the same address (last accessed 28.4.2020).
xv
We assume this variable with its range from 1 - 6 as being numerical and calculate means.
xvi
We used the hotdeck function in R, as well as a linear imputation model for the church attendance of mother,
father, and child (when the respondent was a child).
xvii
See for the details as implemented in R: https://stat.ethz.ch/R-manual/R-devel/library/stats/html/loess.html
(last accessed on 28.4.2021)
xviii
We use the default span of 0.75.
xix
For the mathematical background and an example of how loess-smoothing works in practice in R, see Fox
and Weisberg (2018: 3f.).
xx
This is, of course, arbitrary, but other choices would not substantially change the outcome.
xxi
For a representative mixed methods study that includes both individualized or fuzzy religion and alternative
spirituality in the case of Switzerland, see Stolz et al. (2016).
xxii
There is an important historical literature on secularization in France and other western countries, see for
example Baubérot (2017), Pelletier (2019), McLeod (2000), Portier (2016). We are not sure, however, whether
the historical record and the availability of data from the past permit to well answer the very specific question
we are posing here and leave the question to future research.