Measuring Resurgence of Religion?: Methodological Considerations in a Study of Swedish Editorials

In 2005 the German philosopher Jürgen Habermas gave a lecture for the Holberg Prize seminar with the title “Religion in the Public Sphere”. This lecture, later published as an article in European Journal of Philosophy (Habermas 2006), can be seen as the starting point of a debate about a possible resurgence or new visibility of religion in the public sphere of secularized Western, primarily European, societies. In his lecture Habermas stated:

We can hardly fail to notice the fact that religious traditions and communities of faith have gained a new, hitherto unexpected political importance. The fact is at least unexpected for those of us who followed the conventional wisdom of mainstream social science and assumed that modernization inevitably goes hand in hand with secularization in the sense of a diminishing influence of religious beliefs and practices on politics and society at large. (Habermas 2005, http://www.holbergprisen.no/en/juergen-habermas.html).

In the following, we take as our point of departure the scholarly and political discussion about a resurgence or new visibility of religion in the public sphere of contemporary society that followed Habermas's 2005 lecture (Habermas 2005, 2006, see also Micklethwait & Wooldridge 2009; Toft et al. 2011). More precisely, we address the critique within the sociology of religion that followed this debate (see Beckford 2010; Hjelm 2015), which centres on the question of whether this resurgence is supported by empirical evidence or rather has to do with a change of discursive practices within politics and philosophy (Davie 2015: 28–29). A general assumption in this discussion has been that the media is the most obvious “site” where we could see the manifestation of some sort of public resurgence of religion. However, as pointed out by Köhrsen (2012), there is in this debate a lack of clarity regarding how to measure this resurgence. For example, many studies of the resurgence of religion in the public sphere, he argues, overstate the presence and impact of religion due to an “excessively vast definition of religion” (2012:278). Our discussion in this essay is based on a study of references to religion in Swedish newspaper editorials covering a time period from 1976 to 2010.

The research project “The Resurgence of Religion?! A Study of Religion and Modernity in Sweden with the Daily Press as Case”, which was funded by the National Research Council of Sweden 2010–2014.

In line with these previous studies, one of the major research questions guiding the study that we discuss here concerns whether empirical evidence of an increased visibility of religion can be found in Swedish editorials over the time period from 1976 to 2010. Our primary aim here is, however, to address the methodological considerations regarding measuring a potential increase of public references to religion in society. We will particularly focus on the use of keywords to analyse such change over time, and the choice of unit of analysis in quantitative content analysis. As we will show in the following sections, we found during the course of the empirical work that our methodological choices had a major impact on the empirical results. Our argument in this article is that, in order to develop further our understanding of changes in the presence of religion in the media, a substantial discussion of definitions and methodological choices is crucial. This discussion is relevant not only to a study of references to religion, but also for studies researching other dimensions of cultural change as expressed in the written content of the daily press.

Our discussion of methodological issues when studying changes in references to religion in Swedish editorials will be structured by the following three headings:

Resurgent religion – the Swedish case

Our focus in the present essay is, as was stated above, not on the empirical results from our study of Swedish editorials per se, but rather on methodological questions and challenges that we had to face in the context of this study. Nevertheless, in order to contextualize this discussion, we need to introduce the research question for our study and why we found it both interesting and important.

For a study investigating whether empirical evidence for an increase of references to religion in the media discourse can be found, Sweden seems to be a particularly interesting case. Political scientist Ronald Inglehart, initiator of the World Values Survey (Inglehart 2008), together with Christian Welzel, has constructed a so-called cultural map based on data collected from a large number of countries all over the world (Figure 1). This map has frequently been referred to in the scholarly discussion about social and cultural implications of the development of modernity. Furthermore, this map has also been used as a point of reference in the media as well as in the general political discourse, and for good reasons. The World Values Survey, the context in which this cultural map was created, is a solid research endeavour involving many researchers in many different countries.

Figure 1

This map shows Sweden in a position in the upper right corner of the map. Thus, Sweden has moved further in the development towards self-expression and secular-rational values than most other countries. Since secular-rational values have replaced more traditional values, including religiously grounded value systems, one could interpret this map as illustrating that Sweden should be one of the most secularized countries in the world – and secularized here would mean that there is a reduction of the significance of religion in a general sense (cf. Habermas 2005). This, of course, is interesting in its own right. But it is also interesting for an empirical study of evidence for a resurgence or new visibility of religion. If there is empirical support for some form of growing presence of religion in Swedish society, then there are good reasons to discuss further the general assumption that growing modernity naturally will imply a reduction of the significance of religion. Such assumptions were common in the academic debate in the latter part of the twentieth century, in the sociology of religion as well as in other academic fields (Davie 2013). Assumptions of this kind are also typically part of the discursive traditions within secular modernity. This is the case in Sweden, but also in many other parts of the world – not least in Protestant Europe. Hence, it seems to be of particular interest to explore empirically if there is support for assumptions about a growing presence of references to religion in the media discourse in Sweden.

As will be discussed further below, we used a combination of qualitative and quantitative approaches in our study. The qualitative approach made it possible for us to study more closely the various ways in which references to religion were used in Swedish editorials (see Lövheim & Linderman 2015; Lövheim 2015). The assumption about a resurgence of religion in modern society implies some sort of quantitative dimension, however. It implies that there has been an increase of references to religion over the last few decades. This is a quantitative assumption which made it appropriate for us to use quantitative research methods. This would make it possible to explore quantitative changes over time. It was also in the quantitative approach that we came across methodological questions that turned out to be essential to our empirical findings. In the following, we will discuss methodological issues concerning the use of keywords to analyse cultural change, and the choice of unit of analysis. These methodological questions are not inherently related to the specific research questions in our project – even though for obvious reasons they take a shape and form related to our specific area of research. Rather, these questions are relevant to a wide range of empirical studies of cultural change over time where newspaper material, i.e., words and phrases, are the object of analysis. As pointed out by Riffe et al. (2005), it is important to consider the general character and function of the media material one aims to study, and to make relevant methodological decisions based on such considerations. The discussion below illustrates how theoretical considerations and methodological choices impact the results in computer content analysis of newspaper editorials.

A primary methodological issue for studying changes in the media presence of religion concerns the choice of texts. The choice of editorials as a genre was based on their close connection to political debate combined with the tendency to express continuity of opinions over time (Nord 2001:74). Religion in a general sense is something that has been pointed out as a precarious area for journalism, especially in the context of news reporting (Hoover 1998; Schmalzbauer 2002; Mitchell & Gower 2012). Compared with other newspaper content, editorials might in particular be regarded as upholding values of secular modernity due to their strong connection to ideals of the humanistic enlightenment tradition, such as freedom of opinion, rational arguments and a general balance between various opinions and special interests (Nord 2001). Thus, on the one hand, one could expect this area of discourse to change slowly compared with other media contexts. On the other hand, changes that emerge in references to religion in editorials are for the same reason interesting to study as expression of more profound changes of the cultural landscape and the prevailing value systems in Swedish society.

The methodological approach of this study draws on two previous research projects studying cultural indicators in the Swedish daily press. These two projects, respectively, cover the post-war period from 1945 to 1975, and the period from 1976 to 1995.

The projects “Kulturindikatorer: svensk symbolmiljö 1945–1975” (Block 1984); KUSS 1 (1945–1975) and KUSS 2 (1976–1995).

In our study, we have used the same technique for sampling as was used in the previous studies of cultural indicators. We have digitalized editorials from 1996 to 2010. Since we use the same technique for sample drawing as was done in the previous projects, we have also been able to use the digitalized material stemming from the second of the two projects mentioned above. Thus, we have a digitalized sample of opinion articles covering a period of 35 years, from 1976 to 2010.

We have chosen to analyse the largest newspapers in Sweden, newspapers with a circulation of over 80,000 copies

Aftonbladet, Arbetet, Expressen, Dagens Nyheter, Dagens Industri, Göteborgs-Posten, Göteborgs-Tidningen, Kvällsposten, Svenska Dagbladet, Sydsvenska Dagbladet, Helsingborgs Dagblad.

Our method is a content analysis based on a list of keywords. This list of keywords is made up of words used in previous studies as well as words selected through an inductive procedure related to the digital texts that we were going to analyse in the present study (cf. Knott et al. 2013; Niemelä & Christensen 2013). Thus, the presence of religion has been analysed through a quantitative content analysis using these keywords. These words have been clustered into larger categories that represent various indicators of religion. The indicators cover a broad spectrum of references to religion, including direct references to all the world religions, but also to various dimensions of religiosity and spirituality in general.

Keyword categories (translated from Swedish): religion, Islam, Christianity, Judaism, world religion (includes Hinduism and Buddhism), new forms of spirituality, religious metaphor.

The compilation of a list of keywords that could be used as search words in our computer content analysis resulted in a number of different larger categories. Thus, these categories were indicators covering different dimensions of religion. We also used a number of categories of indicators related to different dimensions of modernity since we wanted to explore the relation between the process of modernization and the presence of religion in the public sphere in Sweden (cf. Figure 1 above). The discussion of the relation between religion and modernity, however, falls outside the main scope of the present essay with its primary focus on methodology.

For all categories, there were typically a number of word stems, exact words, and/or word combinations that constituted the category. Each such category became one specific variable. Altogether, the list of words consisted of 48 different categories – thus resulting in 48 different variables. Of these 48 variables, some related to religion, some to modernity and some to other things that we wanted to be able to analyse, like the name of the specific newspaper, political profile, gender, etc.

After having constructed the list of keywords to be used as indicators of references to religion, the first obvious question concerns the frequency of these indicators in the material. To get a general picture of the development over time, all religion indicators were combined into one variable. This variable was then recoded to show only the presence of a religion indicator or not, not the actual number of appearances of various religion indicators. The analysis of this variable, however, gave very different results depending on methodological decisions. Figure 2 shows the three different results we got as to the general presence of religion indicators in editorials in the Swedish daily press from 1976 to 2010.

Figure 2

This figure vividly illustrates the differences in results. Not only is there a significant difference in the degree to which religion indicators are found in the editorials, but there is also a significant difference in the tendency over time. While one line indicates a significant increase over time, and with a specific peak in the last years of the twentieth century, the other two lines are on a much lower level and do not at all reflect the same tendency over time of religion indicators. We will briefly discuss the empirical results as such later, but the primary objective here is to explore the significance of methodology for the variances in the empirical findings.

In a general sense, it is interesting to note how methodological choices completely determine our empirical findings. This is interesting not only in relation to this particular study. If certain methodological choices can be so important for the empirical findings, it is of course essential to evaluate empirical results in the light of the methodological choices made in a certain study (cf. Riffe et al. 2005). Thus, we need to explore further how methodology can affect the results of content analysis of this kind of media text. In the following, we will describe and discuss a couple of significant methodological issues, issues explaining the differences in the results of our longitudinal quantitative analysis of religion indicators in newspaper editorials in Sweden.

The dashed line in the middle in Figure 2 was the first empirical result that we obtained in our study of religion indicators. Since we wanted to get a better picture of exactly what it was that we were measuring, we decided to explore further the indicators that actually were found in the database using this quantitative procedure. When we took a closer look at the sentences where indicators of references to religion had been found, we made an important observation. Even if all the indicators we had constructed were supposed to be valid examples of references to religion, this did not always seem to be the case. For several keywords, we had to work with word stems to make sure that we would include as much as possible of the presence of these indicators in the material. However, this also meant that words became included as indicators on completely misleading premises. One of the most illustrative examples was the Swedish word for “farmers” (bönder). The word stem of this word is also a word that means “prayer” (bön). A similar example was the word “sect”, a word that can relate to religious sects but also as a word stem appears frequently in words like “sector” and “section”. We realized that this is a significant challenge to quantitative computer content analyses of the kind that we were conducting. Even if dual meanings of many words are obvious when identified, as in the examples above, it is very easy to define indicators in a way that might lead to misleading results. We had definitely not been careless as we worked on the list of keywords to be used in this study, but when taking a closer look at the first results we still found a number of problematic cases.

The conclusion was that we had to test the reliability of all indicators. This was done by manually exploring all indicators to find the ones that represented some kind of problem. Such problems were solved either by making the indicator more specific, or by excluding the indicator altogether. After having gone through all indicators found in the material, we could again analyse the general presence of religion indicators over time. The result of this second analysis was the dotted line in Figure 2, the line that indicates the lowest presence of religion indicators in our material. In terms of the general tendency, however, which was at the core of our research questions regarding changes over time, there is almost no difference between the two lowest lines, the dashed and the dotted, in Figure 2. Thus, the reliability test did affect the degree to which we found religion indicators in our database, but it did not change the result as to changes over time in terms of the presence of religion indicators in editorial texts in the Swedish daily press.

The next methodological issue that we will touch upon here was even more significant for our empirical findings. This was whether the editorial text, each individual editorial text on a specific editorial page, should constitute a single case in our study, or if we should use the whole editorial page in a specific newspaper as a single case. The obvious assumption upon which the present study was founded was that each editorial text should be considered as one case. This strategy became natural for several reasons. The main reason was that part of the material we used, which was digitalized in the previous research project KUSS 2, was organized in such a way. The first main editorial text was indicated with number 1, and the consecutive texts were ordered accordingly. The number of editorials varied between newspapers, but there was also a striking difference in length of the texts. The first editorial text on the editorial page is typically a relatively long text with ample space to develop reasoning and arguments. Many of the other text units on the editorial page were considerably shorter. There were many text units that are no more than one to three sentences.

The difference between shorter and longer editorial texts also showed a development over time. Shorter texts are more frequent in the later years than in the earlier part of the time period, and these shorter texts also add to the total number of texts on the editorial page. Out of the 1229 newspapers included in the study, there are 113 newspapers having seven different text units on the editorial page. Eight editorial texts were found in 37 newspapers, nine text units in 9 newspapers, ten text units in 2 newspapers and finally one newspaper containing eleven text units on the editorial page.

In addition to the development of the length of editorial texts over time, there are also differences in the layout of the editorial page. Editorial pages towards the end of the period look different than they did more than 30 years ago. This is illustrated by Figure 3. The image shows two full pages from the Swedish newspaper Dagens Nyheter. The first editorial page in Figure 3 is from 7 January 1996, and the second from 12 November 2008. As the figure shows, there are significant differences between the 1996 example compared with the copy of the editorial page from 2008.

Figure 3

The differences in layout of editorial pages together with differences in length of the editorial texts do, of course, pose a problem. Is it methodologically valid to compare long texts that contain elaborate discussions with very short texts including only a statement like, “Read this or that person's blog”? There are good reasons to answer “no” to this question. The alternative is either to include only the first, or perhaps the first two or three, editorial texts in each newspaper or to consider the editorial page as a whole to be one case. Reducing the sample available to analysis does not appear to be the most attractive path forward. The option to use only one or two text units from each of the 1229 newspapers included in the sample would drastically reduce the size of the sample. As a consequence, some of the richness of the available material would be lost. Thus, the remaining strategy would be to identify each editorial page

In a few cases, more than one page was allocated to editorial material.

We can now return to our empirical findings. In Figure 2 above, the solid line at the top of the figure illustrates our empirical results when using the whole editorial page as one case. Instead of finding indicators of religion in about 10 to 15 per cent of the cases when using each editorial text unit as a single case, the use of the whole editorial page as case leads to a very different result. Now, we find indicators of religion in around 25 to 45 per cent of the cases. This, however, is not the most interesting difference. It is not so surprising that larger text units, i.e., the whole page vs. each individual text, lead to a significant increase in the relative number of cases where we find references to religion. What is more interesting is that while the use of each individual text as a case showed almost no change over time, the new approach using the whole editorial page as a case actually shows a significant increase in terms of the relative presence of religion indicators from 1976 to 2010. Given our focus on exploring changes over time in our longitudinal quantitative analysis of religion indicators in editorials in the Swedish daily press, the use of the whole editorial page as a case made our empirical results much more interesting.

Are these results, which we obviously found more interesting given our research question, actually valid results from a methodological perspective? Or, phrasing the question in a different way, are the results using editorial pages as cases more valid than using each separate text unit as a case? Given the difference in terms of empirical findings, this is of course a very important question. It is unquestionably so, however, that using single text units as cases implies great differences between the different cases. Very short texts are then on equal terms with much longer and more elaborate texts. Using each editorial page, and all editorial texts on this page, as one case makes all cases more consistent and the whole empirical quantitative study more coherent. Thus, the solid line in Figure 2 represents what we, after a thorough methodological discussion, consider as our empirical findings in this quantitative study. We will now return to our general research question and briefly discuss these empirical findings per se.

As illustrated by the solid line in Figure 2 above, there is an increase over time in references to religion on the editorial pages of the Swedish press. This, however, is not a linear development going from a lower level at the beginning of our time period (1976) to a higher level 35 years later (2010). As indicated in the figure, the period from 1996 to 2005 show the highest relative presence of religion indicators. If we compare these results with the results for each year, we actually find that the absolute peak is 2002. Even if there is a decrease over the last ten of the 35 years included in the study, it is still interesting to note that the level for the last five-year period is almost 40 per cent higher than was the case for the first five-year period. There is an increase from just over 22 per cent 1976–1980 to just over 31 per cent in 2006–2010.

If we look specifically at the various different kinds of religion indicators that we used in our quantitative analysis, such as indicators of different religious traditions, we find that there are statistically significant changes over time for three main categories of indicators: Christianity, Islam, and references to religion where religion is used in a metaphorical way. Indicators of Judaism, other world religions and new spirituality did not show any statistically significant change over time, and such indicators were also generally very infrequent in the material. Islam, Christianity and the use of religion indicators as metaphors however, all increased from 1976 to 2010. Most interesting perhaps is the development for indicators of Islam. The level for the first period of five years is very low. For the first five years, there are indicators of Islam in only about 3 per cent of our cases. There is a peak for the five-year period from 2001 to 2005. During this period, almost 12 per cent of all editorial pages included references to Islam. For Islam, as for religion in general, there is then a decrease for the last five-year period in the study. But still, the general increase of indicators related to Islam from 1976–1980 to 2006–2010 is almost 300 per cent: from 3 per cent for the first five-year period to 9 per cent for the last five years.

Some additional things are interesting to notice when looking at our findings. References to Christianity and references to religion as a metaphor are the most common references to religion. In the early 1990s, however, Islam is actually – together with references to religion as a metaphor – the most common reference to religion. For the following time period, the relation is the opposite. References to Christianity and references to religion as a metaphor peak at the same time as there is a sharp decline in references to Islam. The last five years up to the Millennium obviously sparked an increase in references to religion, but then more in line with Sweden's historical religious situation with a dominance of Christianity and religious references used as metaphors.

As described in the introduction, we also used qualitative text analysis in our study. A content analysis of the editorial texts containing one or more indicators of religion, in total 481 texts, showed three main ways in which religion was used: as a description of an individual, group, or country; as a metaphor; and as the main topic for discussion (see further Lövheim & Linderman 2015). Indicators of religion used as descriptions made up the largest category (45 per cent). A slight majority of these articles include references to Christianity, such as priest, bishop, pope, church, Protestant or Catholic, but an almost equally large group use words that refer to Islam, such as Muslim groups, Muslim countries, Islamic fundamentalism etc. Furthermore, more than half of the articles in this category describe an event or situation in an international context. Many of those are ongoing wars or conflicts between or within nations such as the war in former Yugoslavia in the 1990s.

In the second largest category of indicators (33 per cent of the 481 cases where one or more religion indicator was found) religion is used in a metaphorical way. The majority of the words used in these texts relate to a Christian, biblical tradition, such as prophecy, preaching, doomsday, martyr, sacred, hell and blessing. The smallest category (22 per cent) consists of editorials where religion actually is the main topic of the discussion. For these cases, the large majority contains indicators that refer to Christianity while a smaller fraction refers to a Muslim tradition. The qualitative content analysis of the material shows, in line with the quantitative analysis, an increase of references to religion particularly in the period 1996–2005. This tendency is most salient in the category where religion is the main topic of discussion and where, as we have seen, references to Christianity are most frequent. In the category of indicators where religion is used as a description of something, of an individual or of a group of people, the pattern is not so drastic but an increase is clearly visible: from 25 articles before 1990 to above 40 in the period 2001–2010. Here, references to Islam are also more frequent.

The aim of this article has been to discuss methodological issues that emerged from a study of whether empirical evidence of an increased presence of religion over the latest decades can be found in the media, using the texts of Swedish editorials as our case. As shown above, the results of our study show an increase in religion indicators in Swedish editorials over the period from 1976 to 2010, in particular with regard to the period from 1996 to 2005. Given our research question this result is intriguing, and also well in line with previous studies using a similar methodology. The Nordic comparative NOREL project reported a similar tendency of increase in references to religion in the daily press in 1988, 1998 and 2008, in particular between 1988 and 1998 (Niemelä & Christensen 2013). A continuing interest in religion in secular media is also reported by the UK-based project on media representations of religion and the “secular sacred” (Knott et al. 2013). Our findings of an increased diversity in the representation of religion, where references to Islam increase in the editorials over the time period even if references to Christianity still dominate the picture, is also in line with these contemporary studies from the Nordic countries and the UK.

While these results can be interpreted as supporting the argument about a resurgence or new visibility of religion in the part of the public sphere represented by editorials in the daily press, our main point in this article has been that our methodological choices had a major impact on our empirical results, and thus that the interpretation of these results is a complex matter where methodological choices have to be discussed and evaluated.

In conclusion, we would like to offer the following reflections with reference to further studies using quantitative methods for analysing indicators of religion, or other cultural dimensions, in the daily press. The first reflection concerns the importance of issues concerning reliability. The use of keywords is a common method in quantitative content analysis of newspapers. As our experiences of using keywords to form indicators of religion clearly show, the importance of a thorough reliability test that ensures that word stems and words used do not measure the wrong things is crucial. This might seem like a very basic statement, but the risk of jumping too quickly to conclusions about tendencies of cultural change from the frequency of certain keywords is evident when time and economical resources for empirical research are scarce. Our scrutiny of the way in which certain indicators of religion were used in the editorial texts also made us realize the variety of meanings conveyed with references to “religion” in one way or another.

Another insight from our study concerns the value of using qualitative content analysis in the process of attaining higher reliability of keywords used to measure cultural change in media material. The results of our qualitative content analysis of religion showed that indicators of religion in the editorial texts were used in different ways and with different purposes (Lövheim & Linderman 2015). Our analysis of when and how religious words were used as descriptions, metaphors and as the main topic made it clear that increase of the frequency of certain religious indicators do not in any simple way equal a return of religion as a topic in the political debate. The most frequent ways in which religion was referred to in the editorials was, as shown above, as a description or as a metaphor rather than as a topic for explicit debates on contemporary cultural and political issues. In line with Köhrsen's argument (2012: 278), the lack of clear definitions of religion, together with a lack of rigorous attention as to the reliability of search words and the categories used to measure religion, makes it easy to overstate the public presence and impact of religion.

Our second consideration concerns the importance of using comparable cases for the outcome of a quantitative analysis. As the differences in tendencies of changes over time between using editorial texts or editorial pages as the unit of analysis made clear, great consideration has to be taken to make sure that the analytical units are comparable. Here, knowledge about the content and genre of the media texts analysed is also crucial – as pointed out by Riffe et al. (2005). Moreover, it is important to take into consideration the character of the concepts that are analysed. Religion is a complex category, and as the qualitative study has shown, references to religion can be made in many different ways. Moreover, religion is perhaps not the most expected category to be used in the context of the editorial discourse in the Swedish secular press. This also has implications for reflections on what to consider as comparable cases in a specific empirical study. In our study of religion indicators, using the whole editorial page as case proved to be the most relevant strategy to explore changes over time.

The NOREL study referred to earlier (see Niemelä & Christensen 2013) chose to analyse a selection of major daily newspapers and covered particular shorter periods of time. Furthermore, the NOREL study covered a broader range of genres in newspaper coverage of religion, such as news articles, feature texts, cultural debate, chronicles and editorials. Although quantitative differences between various genres to some extent were accounted for in the analysis (see also Lövheim & Lundby 2013), a more profound discussion of the particularities of various genres is lacking. The material chosen for the study discussed in this essay was limited to editorial texts. This choice can, on the one hand, be seen as more limited than the range of texts analysed in the other studies. On the other hand, this choice enabled us to conduct a more focused study of changes over a longer period of time within one newspaper genre. Our choice of editorials as genre is rooted in their central role over time in Swedish political debate, which makes them an interesting case to study in terms of a possible change in the position of religion in political debates. The fact that editorials might in particular be a genre where values of secular modernity dominate also implies that the tendencies of change we have found need to be interpreted with caution. The presence of religion might be more, or less, salient in other areas of public discourse. Nevertheless, since our findings show that changes in terms of references to religion also seem to be found in editorials, these changes are, for the same reasons, important to study with regard to what they might reveal about changes in the position of religion in Swedish political debate. This also makes our findings as to the representation of religion in editorials of potential interest to media scholars researching cultural and political change as expressed in opinion material of the daily press.

Finally, our study underlines the general importance of practising methodological reflexivity, not least in the study of newspaper editorials related to the general political debates and value changes of society. In studies using quantitative content analysis to analyse these issues, the methodological problems and solutions researchers face, and the way they influence the results, are often presented in footnotes or in a methodological appendix. By bringing these issues to the fore in this essay, we have attempted to open up the discussion of the complex relation between methodological choices and empirical results. Given the methodological considerations discussed here, we have come to the general conclusion that assumptions about a new visibility of religion in the media, sometimes referred to as a resurgence of religion, do not relate to simple linear processes. The role of religion in late modern society is a complex issue. It has to be explored further, but then with great focus on the implication of methodological strategies and choices. More likely than not, the same is the case for studies of other dimensions of cultural change and development when using newspapers and other media content as empirical material for quantitative research.