Doing qualitative and interpretative research: reflecting principles and principled challenges

ABSTRACT Research in Political Science is increasingly based on qualitative and interpretative methods. Based on concrete experiences in a comprehensive qualitative interpretative study, this article discusses general challenges of interpretative methodologies and their application in Political Science. It fills a gap in the current methods literature by concretely explaining how the methodological presumptions of interpretative research are to be carried out in such a way that they lead to substantial findings, irrespective of the material, cases and method one choses. To do so, it is core to analyse not only the ‘what’, but also the ‘how’ and the ‘why’ in the material. A classical qualitative content analysis consists in analysing the ‘what’ in a text, a field, or a visual, that is, utterances, arguments, or concepts that are used in it. Beyond this, in a qualitative interpretative project, the second part of analysis targets the ‘how’ and ‘why’ of constructing meanings, narratives, arguments, topoi, or (mental) images. Analysing the ‘how’ and ‘why’ requires specific analytical and interpretative steps which are, however, barely discussed in the methods literature. Based on the experiences in a concrete research project, the article explains how to structure analytical steps for researching the ‘how’ and the ‘why’.


Introduction
Research in Political Science is increasingly based on qualitative and interpretative methods. Document analyses, discourse analyses or ethnographic studies have become more and more common (Halperin and Heath 2020;Silverman 2021). However, the application of these methods confronts researchers with a number of principled questions and challenges that concern the concrete research process and are seldom concretely discussed in the methods literature (for overviews on the proceedings in qualitative research see Bhattacharya 2017;Flick 2018aFlick , 2018bHalperin and Heath 2020;Liamputtong 2020;Taylor, Bogdan, and DeVault 2016).
Against this backdrop and based on the concrete experiences with a comprehensive qualitative interpretative study, this article discusses general challenges of interpretative methodologies and their application in research in Political Science. The goal is to fill the gap in the current methods literature by explaining what is beyond research techniques, i.e. by concretely describing how the methodological presumptions of interpretative research are to be carried out in such a way that they lead to substantial findings and how to handle challenges that may appear. Therefore both the core steps of qualitative interpretative analyses in Political Science, irrespective of the material, cases and method one choses, and the practical challenges that relate to them will be described (on the following see also Wiesner, Haapala, and Palonen 2017).
As discussed in the introduction to this exchange section, the labels qualitative and interpretative research lead into an interrelated field of approaches that work with qualitative data, ranging from qualitative content analysis over text and document analyses, discourse and narrative analyses, to the analysis of images and pictures, ethnographic approaches, and other types of fieldwork. In the following, these analyses will be labelled 'qualitative interpretative analyses'. It is not the aim of this article to disentangle this broad field into subfields, but rather to focus on commonalities: I will lay out general considerations for carrying out qualitative analyses that include interpretative components. The aim is to describe general steps and crucial decisions which scholars can be expected to be facing in any kind of qualitative interpretative research, independently of their specific topic, research material, methodology and research design, and to open up possibilities to deal with them. The following ideas thus should be valid and helpful for most qualitative and interpretative approaches.
I begin with explaining why it is important to study the How and Why in the material, not only the content, i.e. the What. Then I continue by considering the researcher's perspective, stressing that she needs to keep a distance from the material while at the same time participating in the debate, and underlining that sources and material have a 'veto power' (see Koselleck 2010). After that, a prototypical research course will be lined out, beginning with fixing the research question and ending with interpretation of the results. The description will rely on the research design that I developed in a comparative interpretative study (Wiesner 2014), a comparative discourse analysis.

What, how, and why
Qualitative interpretative research, it is important to begin with this simple constatation, does not just consist of making a plan for the study in advance and applying it to the material to be analysed. This is the classical design of the hypothetico-deductive method as it is described in Karl Popper's Logik der Forschung (1934): a hypothesisbased research design is coupled with an empirical study that leads to a 'yes or no' answer, i.e. to a falsification or verification of the hypothesis (Wiesner, Haapala, and Palonen 2017: 64). Such a design is causal, and it is not applicable for interpretative studies. They serve other purposes, as will be discussed in detail belownamely, they aim at more complex findings and do not search at answering only 'why' questions (which can be done causally) but also at explaining the 'hows'. This is where the interpretative dimensions set in.
In particular, qualitative interpretative research means to carry out several steps that require reflection beyond a mere qualitative content analysis. This part usually comes first in the analysis. It consists in in analysing the contents, i.e. 'what' in a text, a field, or a visual, that is, the utterances, arguments, or concepts that are used in it. It is argued that a second part of analysis should target the 'how ' and 'why' of constructing meanings, narratives, arguments, topoi, or (mental) images. These are often not made explicit. Answering the 'how' and 'why' questions thus requires specific analytical and interpretative steps which are, however, barely discussed in the methods literature. Based on the experiences in a concrete research project, it will thus in the following be explained how these can be structured. The 'why' can include the backgrounds and reasons for an utterancesuch as actor's interests, the political context or the strategic aims related to it. The 'how' mainly concerns the way things are put: which arguments, (mental) images, and motives are used or referred to? Which words and terms? How do they play with established narratives, with fears, with positive and negative connotations? To carry out qualitative interpretative analyses is related to a number of steps -finding, concretizing and further sharpening the research question, the cases to be studied, and the research material; defining methodology and research design; and then, in carrying out the research proper, to adjust these to one another in the course of analysis. In all this, the task to study the 'how' and 'why' crucially determines the setting of the analysis, the material selection, the research questions, and the categories and course of the analysis. Irrespective of the material, cases and method one choses, there are a number of steps and considerations that are valid for all kinds of analyses. The following considerations are to be interpreted as possible approaches and strategies, options and ideas that can be adapted to the needs of the respective research plan.
It is argued that an interpretative qualitative analysis, first, needs to be based on and follow a specific research question and a goal that extends beyond a mere qualitative content analysis. If, for instance, a researcher asks how imaginaries of Europe are constructed in discourse, or which narratives are referred to in images, analysis needs to grasp the dimensions in the research material that help answering these specific questions. The research interest and research question, second, have to target these aspects, rather than simply the contents of the text or the image. This, in return, crucially determines the setting of the analysis, the material selection, the research questions, and the categories and course of the analysis.
Third, this requires analytical and interpretative steps besides a qualitative content study. A content analysis aims at finding out a 'what' in a text, a field, or a visual, that is, the utterances, arguments, topoi or concepts that are used in it. Analysing these and fixing them lies the foundation for the further analysis. In order to answer her specific research question(s), the researcher also needs to study the 'why' and the 'how' behind the 'what' (for a related discussion see Doty 1993). To look into the 'why' means to find out the reasons for an utterance, e.g. the background, the actor's interests, the political context or the strategic aims related to it. The 'how' regards the way things are put: which arguments, (mental) images, and motives are used or referred to? Which words and terms? How do they play with established narratives, with fears, with positive and negative connotations?
In this article, different practices and possibilities for enquiring into the 'what', 'how' and 'why' in a qualitative interpretative analysis will be sketched. Again, it shall be emphasized that these reflections cannot be classified as following any definite school or methodological approach. Even if they are inspired by, conceptual history (Koselleck 2000(Koselleck , 2006Skinner 1978Skinner , 1996Richter 1995;Palonen 2003Palonen , 2005 and discourse analysis (Johnstone 2018;Boreus and Bergstrom 2017;Titscher 2000) as qualitative interpretative techniques (see in detail Wiesner, Haapala, and Palonen 2017), they are relevant for visual approaches as well and focus on aspects that all qualitative interpretative approaches have in common.
The researcher's position and the 'veto power of the sources' Some general considerations on the role of the researcher in interpretative analyses should be kept in mind (see also Bhattacharya 2017;Flick 2018aFlick , 2018bHalperin and Heath 2020;Liamputtong 2020;Taylor, Bogdan, and DeVault 2016). First, reflexivity is important. Depending on the question and the approach chosen, researchers take a position towards the research topic and the material, as well as the results of analysis. This position may be more or less strongly normatively oriented, i.e. the researcher may aim at making a normative judgement at the end of analysis, or not. The position of course is also influenced by the social position of the researcher and the connection he or she has to the field in question, and this position needs to be reflected. (As a simple example, a female researcher has a different position regarding gender inequality in academia than a male researcher).
Second, researchers should be aware of the limits of the available material and the approach and methods used in the study. This is not merely an exercise to be carried out in the 'outlook' session at the end of a thesis or a bookit is something that needs to be carried out right from the beginning. Every concept used, every research approach, every method has its advantage and blind spots (see the reflections in Wiesner 2021b, 2021a), and none is able to answer all questions related to a topic. While, for instance, quantitative text analyses provide representative overviews of large quantities of material, they will have difficulties to explain ways of argumentation, successes of certain arguments, or the exchange of arguments. On the other hand, while qualitative approaches will be able to explain these aspects, they will not be able to deliver a broad or long-term overview.
Third, besides studying carefully what is in the material and what is not, it is therefore to be recommended to study and explain the context in which the text, or the image, is embedded (see below). A detailed analysis of the context makes it easier to analyse the material, as the researcher learns about the reasons, interests and strategies behind certain arguments, actions or claims referred to in the material.
Fourth, analysis should be based primarily on what is in the materialthis may seem an unnecessary or simple directive, but indeed it is a major point of caution. It means that researchers need to base their analysis on the material, or the sources, and that they should analyse them with the help of categories. The research material (often called 'data') never speaks for itself. Both in its generation or collection and its analysis, interpretive researchers need to proceed abductively (Peirce and Deely 1994, 172), i.e. by building plausible explanations that are based on their findings (Jahoda et al. 2009), as will be discussed below.
Importantly, however, a researcher should not search only for predefined categories or assumed findings, making the material fit what she expects to find and jumping to conclusions that are not justified by the material. In particular, researchers must take account of what Reinhart Koselleck named the 'veto power of the sources' (das Vetorecht der Quellen; see Koselleck 2010). This metaphor emphasizes that qualitative interpretative research must always be strongly based on what is found in the material, even if these findings contradict previous assumptions. If that happens, i.e. if the findings contradict what was expected, the researcher should if necessary revise her research design and findings. The result seldom is to abandon the original plan, but it will entail a revision of the research design in such a way as to account for the findings drawn from the material. Research plans thus are "no Soviet-style five-years plans (although those presented to the funding agencies tend to resemble them)" (Wiesner, Haapala, and Palonen 2017: 65). Even if successfully funded, no plan should be followed as faithfully as possible, but its realization requires both improvisation and willingness to revision. A closer acquaintance with the material usually obliges researchers to leave out something of the plan, to narrow down the focus or to intensify the study of those aspects which turn out to be the main topic. This is a perfectly normal stage in interpretative research. But the researcher should neither limit analysis to a mere description or a rephrasing of the contents of the material. A close account of what each party said in commenting the last general elections alone is not enough to answer interpretative questions, and a mere analysis of a 'what', i.e. the utterances, themes, or arguments derived from analyzing the material forms only the first step of research. As has been said in the beginning and is further explained in the following, in interpretative research it is equally important to ask after the 'how' and the 'why' of an argument, an utterance, or an image.

The research question
When one analyses qualitative materialdebates, discourses, documents or other texts, or images, websites and printswith an interpretative purpose, one does it with the aim to find answers to a distinct research question. But how does one find this research question, to start with (for a reflection on the proceeding see also Booth et al. 2016;Tracy 2020;Silverman 2013)? Most studies do not begin with an explicit research question, even if this can be the case. What is more frequent is that research starts with a certain phenomenon the researcher finds interesting or intriguing, or with the idea of a certain field she would like to enquire into, or a certain case she would like to study. However, as said above, the research material does never 'speak for itself'. Therefore the researcher must approach it with his or her precise questions, and interpret the findings. A well-formulated research question therefore is the indispensable red thread guiding the analysis.
A first step to approach the task of formulating a research question is to read the existing academic literature, i.e. to take into account previous findings on the matter and hence the state of the art in research and its lacunae. A second step that is always necessary and advisable is to start at least with a short pretest, or a tentative overview of and an analytical take on part the material. Such a first overview on the potential research material and the information found in it, as well as a reflection of possible steps of analysis in relation to what is found there, will frequently indicate which research question is a fruitful oneand again, it can also happen that the material indicates that not the original questions are interesting, but something different. Researchers never know what they will find in the material until they have started analyzing it.
All this means that the process of fixing 'the research question' has multiple steps that relate to overseeing the state of the art in research in this field, sighting the material, revising one's own question, returning to the material, doing steps of analysis, and concretizing further analytical steps, and all this once again. To illustrate this process, I will in the following explain how it was carried out in an exemplary study, i.e. the comparison of the French and the German discourses prior to the ratification vote of the EU Constitutional Treaty in 2005 (Wiesner 2014).
In this case, what triggered my research interest was a brochure distributed in France that invited people to attend a meeting that informed about reasons for voting 'No' in the French referendum on the Constitutional Treaty in 2005. I was attending a conference in France, being of German origin, and I was wondering about the strangeness of this brochure: it claimed the EU to be 'ultraliberal' and strongly invited people to vote 'No' in the French referendum on the constitutional Treaty in 2005. Such a thing was not to be imagined in Germany, not only because there was no ratification referendum for the Constitutional treaty (the Treaty was ratified by a parliamentary vote alone), but also because such a kind of EU-criticism was unknown there.
As I was quite familiar with France and the French political culture, having lived and studied there for over a year, I got interested because the event strongly signaled that the French political culture with regard to the EU apparently was very different from the German one. When later the 'No' camp took on the victory in the 2005 referendum, this impression was confirmed. I decided that this case was worth a study. But this did not mean I had my research question ready, far from it. How should I put my interestwhy was it so that in France the referendum turned out with a 'No' vote?and my comparative look at Germany into a research project and a research design, then?
Finding a research question in this case required considerable work. It required to find out more about the context of the French referendum, first. There are several ways to proceed in such a case, the usual one is reading, i.e. overseeing the literature and reading about what others have found out about the topic in question. If possibleand in the case discussed here it was indeed possiblediscussions with experts on the topic can be very helpful, too.
After thus having explored the field, one core finding was that there had been a large public debate on the referendum and the Constitutional Treaty in France, i.e. a discourse. The difference between a debate and a discourse is crucial here for methodological purposes. A debate is an exchange of arguments in a certain matter. Building on Wood and Johnstone, a discourse is a setting of practices or events that constitutes meaning and that can be distinguished as according to a certain subject, or a special institutional setting or context (Johnstone 2018, 1-5;Wood and Kroger 2000, 3). It was this very aspect of construction of meaning that I was interested in. So my aim turned out to be a study of the French referendum discourse which took place in different kinds of media and also in public forums, and that apparently had strongly contributed to turning the general opinion on the EU Constitutional Treaty into the negative.
A second point to clarify was what exactly I wanted to study: it is an important point to underline that a formula like 'how the French referendum discourse went about' in this case would have defined a topic or a research field, but not yet a research question. So what was the question that I wanted to answer?
A link to earlier research proved decisive here, i.e. the academic debate on building a European Identity and an EU Demos. Studying the question of Demos building in the EU (Wiesner 2007) I had been left with the puzzle of 'EU identity construction'research results in this field claimed the development of an EU-related identity a necessary condition for democratizing the EU, but also underlined how complicated this construction process is. Results also hypothesized that discourses on the EU and EU referendums would be means of EU identity construction. It was therefore a crucial finding of this academic debate, and also of my previous work, that public discourses on the EU could potentially have a crucial impact in helping to construct an EU identity.
Thus relating to the state of art in research, I found that an analysis of the French EU referendum discourse would probably give some valuable findings regarding construction processes of European identity as they went about in reality. The research question would be, then, how different kinds and courses of a discourse would influence the construction of an EU identity and an EU-related Demos in different ways? This was already closer to a proper research question, but not yet there.
Third, for further clarifying the research question and the research design it was helpful to decide whether I wanted to do a comparative study, and what I wanted to compare. I decided to compare the French referendum discourse with the German discourse that preceded the German Treaty ratification process in spring 2005. The German case was different from the French one in numerous respects: in Germany, the Constitutional treaty was only ratified in parliament and not by referendum, the discourse was much less intensive, and it was also much less controversial. But there was a discourse and there was a ratification, which was the common denominator of the two cases that permitted the comparative perspective.
Defining the final research question required a number of further steps and back-andforth between the first parts of the analyses, the material and the project, until it was finally settled: To what extent do national EU discourses function as media for the formation of European identity and the democratization of the EU?
This overarching question lead to two specific research questions: (1) How do the national EU discourses by political, academic, and economic elites, mediated via national quality newspapers construct the EU and Europe? (2) How are national EU discoursesand thus also the formation of European identityshaped by specific national contexts and references? A problem that occurred at this point was how to shape my comparative research design. The study aimed a comparative examination of national EU discourses as media and processes that construct European identity, as well as of the structures and contextual factors that shape or influence them. Investigating discourses means to decide on a discourse-analytical research design, but this does not imply a specific or even generalizable discourse analysis procedure. Rather, a wide variety of discourse-analytical approaches can be distinguished, all of which merely share certain epistemological and methodological premises (cf. Johnstone 2018, 1-15;Wood and Kroger 2000, 3, 95).
For the study, a comparative discourse analytical research design was developed independently. It differs from most other discourse-analytical approaches in one core respect: usually, discourse analysis is not seen as a path to an understanding of events in the world outside of discourse (Wood and Kroger 2000;Johnstone 2018;Boreus and Bergstrom 2017;Titscher 2000). In contrast, the study was based on the methodological premise that discourses should not be regarded as self-contained, with meaning construction taking place without notable external influences. Rather, it is assumed that discourses are not only potentially open to outside influences, but can be fundamentally influenced by prior discourses, socioeconomic factors, and social and political structures. Therefore, the contexts of national EU discourses are also considered and operationalized in five areas: (1) political system, (2) political parties and European integration, (3) citizenries and their attitudes towards the EU, (4) the patterns of national identity, (5) previous EU discourses.

Determining the research material
The first step of any interpretative analysis always consists in selecting a corpus of theoretically relevant material. As briefly sketched above, depending on the research design, material of any kind can be of interest if it is useful for answering the research question. But how can the researcher determine which kind of material is the good one for her purposes? The main criterion for material selection in any case is not representativity, but theoretical relevance: is the material selected theoretically relevant for answering the research question?
There are three classical types of material selection: (a) case contrasting (Kelle and Kluge 1999, 40), (b) Theoretical Sampling (Glaser and Strauss 1967;Khan 2014;Vollstedt and Rezat 2019), (c) selection plans (Kelle and Kluge 1999, 40). These types that will be exemplified below can also be used during the analysis to reduce the material basis for detailed analysis. The first step of material determination will usually be closely linked to the determination of the research question as it was just sketched. As said above, it is imminently important that while determining what is the exact research question she wants to answer, the researcher also starts reflecting upon, and also trying out, what kind of research material will help answer that question.
In the exemplary study, to analyse a public discourse related to an EU referendum, several different kinds of material could have been used: booklets, pamphlets, and flyers that underline the political claims made in the discourse; posters, websites, blogs or other more visual campaign material; party protocols and information material; TV programmes; discussion protocols of public meetings; newspaper articles; books on the matter; expert interviews; research articles; parliamentary protocols and government documents.
Therefore it is important to determine what exactly the research interest is in connection to selecting the material, as not any kind of the possible material will help answer the question in an equal way. Moreover, it is decisive to know that the material in question will contain the information that is sought forthis is why, as repeatedly said, an overview of the material and a short prestest with the analytical tools is strongly recommended. If this has proven promising, the researcher should start sampling the material in question, starting by seemingly basic, but elementary matters: can I find the material? Is it accessible? And can it be retrieved? If for instance only part of the relevant TV material is accessible for a researcher, because the usage of the other is prohibited by the TV networks, it might be unwise to use them as research material. If the websites in question have not been archived (which is very rarely the case) it might not even possible to do a proper analysis based on websites. Party protocols and materials usually are more systematically stored, but how can they be accessed? Does the researcher have to dive into the archives herself and dig out what she can find? Can someone help her with that? Or are the documents in question maybe even stored online and she can simply download them? Even in cases where the material is well stored, more or less complete (from what can be said at that point of the research) and easily accessible, it can be a problem how it can be retrieved: can it be downloaded? And in what kind of data format? In case the researcher wants to use a software tool that only uses pdf or Word format, for instance, it might represent not a major problem, but some considerable work, to transform websites into the appropriate form.
The next question the researcher should ask herself is whether there is enough material to answer the questionor whether there is maybe rather too much of it? If, for instance, she finds only 10 or so texts on her topic of the public discourse, it will be hard to use it for answering the research question properly. A discourse of several months can hardly be analysed when there are only ten texts. At this point, the researcher can either change her question and her approachor her choice of material.
If, instead, she finds over 8000 texts in her material sampling as it has been the case in the exemplary study, she can be sure on the one hand that it will most probably be enough to analyse her discourse of several monthsbut she will also know that she will never seriously be able to analyse in detail all those 8000 texts. She rather will have to find a way of limiting the number of texts that are finally analysed in detail.
The next step that should be carried out might help with this purpose. The researcher should ask herself whether the material she has tested contains theoretically relevant information. Does it, in other words, contain the information she looks for and that helps answering the question? This step usually entails probing more intensively into the material, i.e. doing some readings, establish overviews, and start analyzing the texts. It might get obvious in such a reading that the material does not contain the relevant information, but also that it contains information that so far had been unknown to the researcher and will be important to be includedhere again we encounter the 'veto power of the sources' (Koselleck 2010). How the researcher at this point proceeds with her interim findings on and in the materialwhether she writes protocols, or memos as in classical Grounded Theory research (Glaser and Strauss 1967), is secondary, the important aspect being that she proceeds this way and somehow makes her interim findings valid for (a) the further research process and (b) the final findings of the research.
In case a comparative analysis is planned, it is important to make sure whether the material can be compared, i.e. whether it shows similar features at least with regard to a certain extent. For instance, it can be complicated to compare very different types of protocols or different types of media. It is worthwile noting at this point that comparisons in interpretative analyses so far are seldom discussed in the academic debate (but see Abramson and Gong 2020).
The last two decisive questions for material selection are how long it takes to assemble the material and how it can be stored (and eventually processed). Especially in cases where the material is difficult to access (paper archives) it is important to estimate how long it might take to assemble the necessary corpus, but this is also the case when extensive online searches have to be carried out. Storage is a question especially in case of paper copies, but also with regard to electronic documents: shall they be just assembled in a computer file? Or shall they be stored with the help of the analytical software?
In brief, all the information that has just been sketched can be found out in the pretest, and it can also be tested how long each step will take. On this basis it is possible to establish a realist work plan.
In the research design described here, after some pre-tests with the material, two selection types have been used.
A selection plan has been used to define the cases to be analysed as well as the material. This was possible and useful because of prior knowledge in the field which served at determining cases and material that was theoretically relevant. For a number of reasons the material selection was limited to the quality press. The first reason was comparability: as the study was aiming at comparing the German and the French discourse, first it would have been difficult to analyse very different types of materials. If, for instance, leaflets and brochures were analysed, those would only have existed for the French case, as in Germany there was no public campaign. Comparability, second, regarded the similarities within the material corpus. Visual elements, as they are contained in TV programmes, must be analysed with different strategies than texts (see the contribution by Freistein and Gadinger in this exchange forum), which would have added at least a different layer of analysis. For this reason, the corpus was determined as according to a text-based strategy of analysis.
This could, however, have spoken in favour of including different types of texts, which was my original idea. At first I was aiming at including various daily and weekly newspapers as well as books. A decisive information that changed the selection parameters in this respect came from French colleagues. In the beginning of material sampling I carried out a guest stay in France. I discussed the project with numerous French colleagues who made clear one important factor I had previously not noticed, because to be aware of it, one definitely needed to be in the heart of French political discourse: a decisive split in the referendum campaign went through the centre-left, i.e. the socialist and the green parties and their adherents. Had I, as I originally intended, focused on a selection of various weekly and daily newspapers that were more or less to be situated in the political centre, i.e. the biggest newspapers, I would not have found this decisive split in the material. In these discussions with the French experts, I therefore obtained a crucial information about the French case and the criteria of theoretical relevance for it: if I wanted to understand the dynamics of the French discourse, I needed the crucial conflict lines to be present in my material, so I needed to sample the material in such a way that it would do so.
Such discussions with experts on the subject are always to be recommended, especially if the researcher is not herself a full-fledged expert of the field, which is mostly the case if she deals with specific details of a specific political culture she is not fully familiar with. Key issues like deeply-rooted and unspoken national traditions or historical conflicts are difficult to discover if one is not a total insider. But even if a researcher feels that she has considered everything that possibly might be important, a discussion with experts is to be recommended just to make sure this is true. A stay in the country in question is always recommendable, too, if this is possible.
As a consequence to the decisive insight into the French Conflict lines, I decided to change my plans on the corpus and to concentrate on daily quality newspapers alone, leaving out the books and the weekly papers. In order to reach the goal to enable a perspective on the discourse in its broadness and including all the relevant conflicts, I decided to include four quality newspapers per country in the corpus, covering a broad range of the political spectrum from conservative-right to liberal left and extreme left. The newspapers I chose were thus: Le Figaro, Le Monde, Libération and L'Humanité for France Frankfurter Allgemeine Zeitung, Süddeutsche Zeitung, taz and Neues Deutschland for Germany.
The relevant time period also had to be determined. The decision here was taken in relation to crucial dates in the Treaty ratification process. On 1 January 2005, Jacques Chirac, then French president, had announced he would hold a referendum on the EU constitutional Treaty. The referendum took place on May 29th. Just before, on May 12th and 27th, the German Bundestag and Bundesrat had ratified the Treaty. To include these events and also the discussion on the French referendum and its consequences, the analysis was extended until one week after the meeting of the European Council on June 16th and 17th. The period of analysis hence was set from 1 January 2005 to 25 June 2005.
Having thus determined the material and the time period to be studied, the material needed to be retrieved. I used web-based press databases for this purposes. For the French case, I assembled all articles that contained the search word 'referendum' in the four newspapers. For the German case, the search words had to be varied, as the constitutional treaty was ratified in parliament. Here the search terms "EU UND Verfassung* (EU AND Constitution*) and "EU UND Referendum' (EU AND referendum) were chosen. Other possible search words were tested, too, but the tests showed that they did not lead to different material. In total 6373 articles were sampled for France and 2152 for Germany. After a control of the material, part of it had to be corrected, as the search had brought about some double and fake findings. Finally, 8145 relevant articles were included in the analysis.
This was obviously too much for a detail analysis, even for a determined researcher. I therefore took some time to reflect upon and test sampling strategies I could use to reduce the corpus, but they all proved unsatisfactory: a random selection would have produced a random outcome, mixing relevant articles and ones that contained only marginal information such as short notes of 10 lines. A search using topics or new search words was not satisfying because without actually some detail insight into the discourse it was not possible to know what had been the decisive topics. The same was true for key events.
The successful strategy, despite the big number of articles, proved to be theoretical sampling, which I used throughout the coding of the material: based on findings of the previous steps of the analysis and the coding, the next relevant material unit has been determined. The selection process then ended when no further theoretically relevant differences and findings could be detected in the material, i.e. when theoretical saturation was obtained (Glaser and Strauss 1967;Vollstedt and Rezat 2019;Khan 2014;Kelle and Kluge 1999, 44-46).
In the concrete case this meant that I did a first overview of all the of material oncethis is possible as newspaper articles have headlines and it can be quickly grasped what topics they cover. The results of this first overview were handled in a twofold way: first, a sample of articles that potentially were relevant for detail analysis was assembled, and second, I noted findings in an overview document on the course of the discourse.
I then proceeded on a month-per-month basis with these pre-assembled relevant articles, reading and coding them one by one. Thus I found out more and more, and step by step, about the themes and arguments, the actors and their strategies in the discourse. On this basis at some point a theoretical saturation (Glaser and Strauss 1967;Khan 2014; Vollstedt and Rezat 2019) was obtained: when an argument popped up for the 37 th time and I had already analysed it in three possible shapes, it was not necessary to do this for the 37 th time all over again. I thus was able to limit material analysis to those texts that added new information. Nevertheless, all in all 2247 articles were hand-coded, using a software for qualitative data analysis (MaxQDA).
At this point, some words shall be said about the usefulness of computer-supported analysis in the analysis of textual material in general. In a nutshell, the help of such tools is very useful, but they should be regarded exactly as this: as more or less sophisticated tools. Most studies that start with retrieving their material in electronic form will rely on computer-support from the beginning, by using search engines, search words, databases, or the internet; and also most of the research material will then belong to the field of the new catchword of 'Digital humanities': it is digitally registered and stored in large, often freely accessible databases for the biggest part. If textual material is analysed, to use a software like MaxQDA or Atlas.ti for the further steps of analysis may be helpful for a number of reasons (Gibbs 2013;Lewins and Silver 2014;Kuckartz and Rädiker 2019;Maricut-Akbik 2021). It is to be noted that several different kinds of analyses are possible, and most of the softwares in question offer possibilities for electronic analyses to be run, i.e. computer-based ways of text or data mining. These represent a different kind of approach that is not in the focus here. In interpretative qualitative analyses, such computerized analyses can be useful at a certain stage, especially in the beginning when the researcher is interested in an overview on keywords or topics. However, they do not replace an interpretative analysis proper that is to be carried out by the researcher. The software is then used as an electronic tool. It registers and documents the work, but it does not carry out the analysis itself. In the following, I will hence describe the steps of an interpretative qualitative analyses that is based on interpretation and coding done by the researcher.
To use softwares to document and register the proceedings of the interpretative qualitative analysis and the results is to be differentiated not only from the computerized analyses sketched above, but also from working with paper sources and without a softwarethis was the classical way of doing qualitative interpretative analyses until the softwares appeared. In the days before these software tools were developed, research was often carried out on paper, using pens, scissors and glue to mark, cut out and play with the intermediate findings. While this is perfectly possible today, the softwares offer some easy possibilities to order and sort the material and to register the proceeding of her research as it develops (Gibbs 2013;Lewins and Silver 2014;Kuckartz and Rädiker 2019). The softwares are designed for registering and supporting analyses of large corpora of material: a corpus of several hundreds or thousands of texts and documents can easily be handled and interpretatively analysed. The advantage especially in the coding phase is then that the software registers the coding system as it is developing, and it offers different possibilities to sort out the coded parts of the texts and hence to play with them at a later stage.
All in all, there are a number of arguments that speak in favour of using a software, but again, the limits of what it can do shall be underlined: the software helps to register the researchers own interpretative analysis, it does not carry out the analysis itself. The software is always only as clever as the questions that are made the basis of the analysis. The usage of a software, then, is a useful tool in supporting the researcher's political and methodological literacy, but it cannot replace it. Or, in other words: neither the software, nor the sources just speak for themselves, but it takes the interpretative analysis by the researcher.

Analysing the material
When analyzing the material, the researcher needs to follow an individualized proceeding that is custom-tailored for the needs of her research. However, some generalizations can be made here as well.
First, the material needs to be treated and/or prepared for analysis. This concerns some points previously mentioned: the format of the texts when they are retrieved, the format they need to have for analysis, and the type of material (on paper or in electronic form). In any case the material needs to have a form that the researcher can use it for her analyses. If they are computer-based, this means she will probably need word-, rtfor pdf files. If she uses interviews, they need to be fully transcribed in order to be used as a written text. Once the material is thus prepared, the researcher can start her analysis.
It is important to underline once again here that there are some practices and techniques that should be common to all kinds of interpretative analyses, but that there is by far not a single way of carrying them out. The following considerations therefore are to be interpreted as possible approaches and strategies, options and ideas that can be adapted to the needs of the respective research plan.
One possibility is to 'just start' with an overview on all or part of the material, reading through some text(s), looking at some images, noting impressions and ideas, and develop first theses, on categories and possible findingsagain, this is very much what is suggested by Grounded Theory (Glaser and Strauss 1967). An opposite strategy would be to proceed with a predetermined plan for analysis, by defining, say, the order of material analysis, the steps of analysis, and the categories as much in advance as possible this means to try and follow the principles of a deductive research design as far as this is possible in an interpretative design. In reality, most researchers will combine parts of both approaches in an abductive design (Peirce and Deely 1994). Independent from how the researcher proceeds, it is, again, essential that she is open for what she finds out and learns in her analysis, and also to unexpected findings.
The next step in text-based analyses then is coding (Adu 2019;Elliott 2018;Gibbs 2013). Coding means that after the material has been selected, the relevant information will be marked in the texts, or extracted from the texts; they are coded. Codes can be understood as intermediate categories of analysis that build a basis for the final results, but that do not themselves represent the final results.
Coding systems can be either defined before coding the material or they can be developed throughout the coding, largely following Grounded theory premises (Vollstedt and Rezat 2019;Kelle and Kluge 1999). This distinction is decisive, albeit somewhat idealtypical. In the first case, when a coding system is defined before analyzing the material, the researcher puts together a list of categories that she supposes to be of importance in the analysis. The coding scheme hence is based on an expectation of what will be found in the material and is relevant for answering the research question. The expectation is usually well founded on what is called 'prior knowledge' in interpretative analysis (Kelle and Kluge 1999, 14-17). Prior knowledge here means any kind of prior knowledge, ranging from everyday knowledge (i.e. the researcher knows what is decisive in the French discourse because she is French and was present in France and read the newspapers) to previous research results (i.e. most studies on a topic agree on certain core findings).
The other strategy is to proceed in a way that is purely inspired by theoretical sampling (Glaser and Strauss 1967;Kelle and Kluge 1999), i.e. strictly based on the information and patterns that the researcher finds in the material. The strong advantage of this approach is that it guarantees the highest level of openness, while the other approach runs a higher risk of closedness. This distinction can again best be explained using an example from the press discourse analysed.
In the exemplary case, at first, I started with a predefined coding scheme in analyzing the French referendum discourse. I had read a lot of secondary literature on the French discourse and its context, and I had discussed the research design with colleagues -I seemingly had all information together to know what I was looking for and what I could expect to find. But when I started coding, I saw that some of my previously defined categoriesthat had been presented as well-established findings in the literaturedid not work. In other words: there were no parts in the texts that could be assembled to some of the codes that I definitely would have expected to be decisive after reading through other research findings.
This is a strong case for the need to respect the 'veto power of the sources' (Koselleck 2010). The most crucial example in my material was the theme of 'le plombier polonais', the Polish Plumber. It was well-received and well-discussed in public and academic debates, both in France and outside France, that this stereotype of a Polish Plumber who came to France and took away jobs that could have been done by French had been decisive in the referendum discourse. Now, the material analysed definitely showed that this was not the case, at least not in what had happened in the press discourse. The Polish Plumber barely appeared, and where it did appear, it did not have any decisive effect in the discourse, i.e. people did not react to the motif, it did not trigger conflict, or anything near to that. The analysis proved, on the contrary, that a number of other motifs and arguments decided the discourse that had not previously been known to me, and neither had been mentioned as results in previous research on the topic. This experience underlines why it is essential in interpretative analyses to include an element of deliberate openness and the possibility to learn from the material otherwise we might overlook the most telling findings.
This does not mean that a predefined category is altogether a bad thing. It rather is an invitation for a mixed strategy. In practice it is often recommendable that both approaches are combined. Researchers will start their analyses with a certain set of predefined codes or categories, and add to them in the course of their analysis what they newly learned. The only important remark in this case is that it is then necessary that the complete coding system which is finally developed is applied to the whole material.
In sum, in coding the material, I proceeded as follows: Preparation of coding (1) definition and sampling of corpus Coding (2) definition of a basic coding system (prior knowledge) (3) first phases of coding and further development of coding system (grounded theory approach) (4) further selection of texts to be coded (grounded theory approach) (5) after first phase: fully developed code system (6) application of fully deveoped coding system to the whole coded material Analysis (7) Interpretation and further analysis of the results Source: (Wiesner 2014, 113).
Categories for analysing debates and documents: finding about the 'how' and the 'why' It has been outlined above why it is essential in an interpretative analysis not only to search for the 'what' in a debate or a document, but also for the 'how' and the 'why' (for a similar discussion see Doty 1993). The 'what' in a very general sense can be lined out by the analytical steps presented above. When the researcher is thus done with her coding, she is more or less done with the 'what' questionsshe will know which arguments, motives or (mental) images she finds in her material. At this point, a remark on numbers and quantities is relevant. Numbers and quantities of utterances or arguments (or frames, or the salience of a word, etc.) are useful indicators that give an overview of what happens e.g. in a discourse. But the researcher needs to be aware of the limits of counting: Quantitative analyses cannot give complete answers to 'how' and 'why' questions, since quantities of a 'what' do not explain a 'why'. Changes in quantities of words, themes, or topoi do hint at possible changes and developments that invite to further study, even if they do not tell us what happened and why. Despite an insight into these limits of quantitative content analysis, there is no reason to strictly avoid counts and numbers altogether. The researcher only needs to be aware that numbers and frequencies only give an average idea on what was happening in the debate and indicate phenomena worth further study. In the exemplary discourse-analytical study that was already explained above, numbers were mainly used to follow the course of the discourse. The articles that had been retrieved in the database were counted on a daily basis, and the count showed several crucial results of the course of the discourse.
It is important to highlight that researching the 'what', 'how' and 'why' cannot be clearly separated into completely different steps. On the contrary, these tasks and steps are interrelated, so while analysing the 'what', i.e. the contents in the material, the researcher will usually also already develop several theses and ideas that open up a way to the further analysis of her findings. As said above, to then focus on the 'how' and 'why', she will add further steps to the analysis after studying the content and doing the coding proper which explore interrelations of the 'what' findings with the interests, strategies, actors, their interests, and the contexts. This enables to explain the 'how' and 'why' behind the texts (see also Doty 1993).
In the following, some useful possible research dimensions for getting behind the 'how' and 'why' will be sketched. The dimensions and approaches that are explained help linking the what (is being expressed) to the how (it is expressed) and the why (it is expressed). In particular, to understand the 'how' and 'why' behind political and rhetorical moves and actions, it is often helpful or even necessary to be familiar with the context, or to include it systematically into the analysis.
Finding out about the 'how' and 'why in the exemplary study For the purposes of the comparative study of the two discourses, I developed three key questions on the what, how and why: (1) what happens in the discourse (contents)? (2) How is meaning constructed? and (3) Why is meaning constructed in a certain way? Now, how did I proceed in answering them?
As I said, the phases of analysis are interwoven, and so a number of analytical steps that served at answering these three questions have already been described above. They referred to eight analytical dimensions that I used as key categories:

Course
The course of the discourse with regard to topics, intensity/number of contributions, significant events 2. Actors The central persons or institutional actors shaping the discourse 3. Rules They structure the course of discourse and the sayability of utterances 4. Reference level Political levels (EU, foreign, domestic) or thematic fields to which the discourse relates 5. Topics Content areas touched upon by the discourse 6. Motifs Types of attributions of meaning in the sense of attributed characteristics and motives for action 7. Arguments Typifying the course of meaning attributions or argumentation processes 8. Cross-References Relationships between motifs, subject areas, reference levels, rules, actors or contextual factors constructed in discourse Own representation.
Based on this typology, I studied the course of the discourses, namely, the development of its intensity (this was done by counting all articles that appeared per day), and did an overview on course, actors, rules, reference levels, topics, motifs, arguments and cross-references of the discourse. This also influenced the theoretical sampling of the most relevant articles. Based on this ongoing selection of the theoretically most relevant articles, the discourses were then coded with regard to the eight discourse dimensions sketched above.
The following A crucial step in the study has not yet been sketched: a systematic context analysis of both cases. Context for the purpose of this research was systematically studied with regard to five dimensions: (1) the political systems, (2) the political parties and their reaction to European integration, (3) the citizens and their views on EU integration, (4) concepts of national identities and (5) prior discourses on the EU. The core findings were (see Wiesner 2014, 398-415): (1) France's political system is a presidential system, showing a traditionally strong role for protest movements and a weak parliament with a majority voting system: It therefore sets other conditions for a discourse than the German system, which is parliamentary, with a proportional voting system and a strong culture of consensus. France experienced decisive changes in the political system due to Europeanization, but Germany less so. Moreover, France held a referendum, and this means that in France the chances for an intensive discourse were higher, and an occasion was created for interaction of elite discourses and opinions and attitudes of the population. For six months, the European Union became a central topic of public debate across nearly all social classes and groups. In Germany, on the contrary,there was only little public debate, and this debate rarely cut across the limits of the level of political and media elites.
(2) France's political parties were strongly influenced by European integration, or more exactly: by the fact that they had to take a stand with regard to it. In particular following the Maastricht debate in 1992, diverging actors left mainstream parties and founded new, often EU-critical parties or movements, while the official positions of the centre actors and parties converged. In Germany, most mainstream parties except the Left party (in 2005 this was the case for the then existing PDS) agree in an elite consensus in favour of EU integration. (3) The citizens had a different role in both discourses: in France they voted on the Constitutional Treaty, hence they were decisively concerned by the referendum discourse, and they also were actors in the discourse. In Germany, due to the parliamentary ratification, citizens intervened much less in the discourse. The citizens' opinions with regard to the EU, on the other hand, are rather similar in both countriesin both there is an EU-critical potential of up to 50%but only in France this played a role in the discourse. (4) The narratives of national identity are different, as is well known, in particular with regard to EU integration: In Germany, EU integration was part of the raison d'état of the new federal republic, it was a means to become sovereign again, and it became an integral base for the new narrative of national identity. The French national identity narrative on the other hand is based on specific interpretations of the state, the republic, the nation and sovereignty (unified, impartible, special) that are rather contradictory to European integration. (5) In France, finally, the attitudes of national elites regarding the EU are traditionally conflicting, and so were most prior EU discourses. In Germany, both the attitudes of national elites to the EU and prior EU discourses are much more harmonious.
There is a broad national consensus of the political elites in favour of the EU, German EU membership and Germany's role in the EU.
All of those context factors influenced the discourses, sometimes directly, when the context was referred to in the discourse, and sometimes indirectly, when a context setting e.g. influenced an actor's behaviour. The context factors thus were extremely useful to explain the 'how' and especially the 'why' of the discoursewhy did the discourse take a certain turn, to which context factors did an argument relate, and out of which strategic interests did an action take place?
Building typologies, drawing conclusions, and answering questions Towards the end of her study, the researcher will have analysed her material with regard to her research dimensions in several steps of analysis, going back and forth between her material, her questions, and the intermediate findings. Her ultimate goal is to answer her research questions by drawing conclusions from her results. Intermediate steps that will be helpful in this task are the building of typologies, the ordering of the findings, and the systematic linkages to the context.
A typology in this context can refer to any kind of order that structures the findings into groups as according to shared characteristics. The researcher might for instance build typologies of arguments, terms, (mental) images, or rhetorical figures. The criteria that guide building these typologies always refer to the characteristics that are relevant with regard to the respective research interest, and, once again, not to any criteria that are objective or always the same. On the contrary, the building of typologies is a highly individual step of the research.
In the synthesis phase of the exemplary study, the results of the first three phases were synthesized and analysed with regard to the 'how and why'. Based on the synthesized findings, (a) the research questions were answered, (b) the results were summed up, (c) the contexts were compared, (d) the discourses were compared and conclusions drawn. Typologies were one result. They were summed up in overview tables that allowed for a compact answer to the question. The following tables show the main crucial findings of the discourse analysis:

Key motifs, topics and rules of the discourses in the exemplary study
To illustrate what the categories above mean, one example will now briefly be explained. It concerns the most decisive motif in the French discourse, the antiliberal motif, which can be summed up like this 'The EU threatens France's welfare state, and it is ultraliberaland we are fed up with this!' This motif definitely was dominant in the discourse. It was coined on the political left and influenced actors and the debate on the right. It used traditional motifs of the French political culture like services publiques, Égalité, a strong state, protest, elite criticism, and France's special role. It was used strategically by left-wing and centre-left actors (Trotzkyists, communists, dissident socialists like Laurent Fabius), and it met with a mood of the citizens which was shaped by a disenchantment with politics and a strong feeling of social insecurity.
The main motifs, topics and rules of the German discourse, on the other hand, rather differed into motives related to the outside of Germany and those directed to the inside. As an example, a key rule of the German discourse was the Silencing strategy. It was used to silence EU criticism issued both from actors in the mainstream big parties and in the far-left PDS. The silencing strategy proceeds in three steps: (1) EU-criticism is muted in the discourse (it is simply not further discussed when issued) or it is marginalized (discussion only in small articles or remarks) (2) EU-criticism and the critics are trivialized ("anyway, it is impossible to take him seriously') (3) The last Step is to threaten the critics with sanctions ("obviously, the party might remind such events when the lists for the next elections are set up').

In conclusion
In sum and in conclusion, doing interpretative qualitative research requires the researcher to follow an analytical path between formulating the research question, finetuning the research design, defining the research material, and analysing it in several steps, focusing on the 'what', 'how' and 'why' dimensions. All interpretative qualitative analyses require to combine an analysis of the 'what' found in the text with an analysis of the 'how' meaning is constructed or political action is taking place, and the 'why' arguments are used, actions are taken, and certain moves are made. Proceeding thus requires to keep the research design open. This means to follow an abductive approach that leads to building plausible explanations (Peirce and Deely 1994;Jahoda et al. 2009), relying at least in parts on Grounded Theory premises. Quantifications of codings, arguments or topoi as well as automated analyses can deliver additional information in such analyses, but the interpretative work needs to be done by the researcher herself. Importantly, the ideas and suggestions presented above can serve as an example or toolbox for an interpretative qualitative study. However, this does not mean they can always be directly transferred without adaption. There is no fully predefined course in research, as it is of decisive importance that the researcher develops or at least adapts the categories of her analysis to her specific research question and her material. It might be the case, for instance, that the researcher is interested in other dimensions than the the ones sketched above, or that she wants to focus rather on the ways and moves in a debate that use and shape the rules and structures. Research dimensions thus need to be defined and varied in accordance with the research interests and the research question. In all this, it is indispensable to respect the 'veto power of the sources' and to be ready to accept both the research design and the findings accordingly.

Disclosure statement
No potential conflict of interest was reported by the author(s).