Mining PIGS. A structural topic model analysis of Southern Europe based on the German newspaper Die Zeit (1946-2009)

ABSTRACT

The onset of the Eurozone crisis popularized the PIGS acronym, contrasting Portugal, Italy, Greece, and Spain unfavorably with their Northern European neighbors. In order to investigate the historic formation of this country group heuristic, we apply a Structural Topic Model (STM) to all 2,443 articles published between 1946 and 2009 in the German newspaper Die Zeit that mention Spain. The estimated topics and their evolution over time confirm a strong association of Spain with other Southern European countries as an ongoing characteristic of the German medial discourse for more than sixty years. Moreover, the STM allows us to distinguish and characterize a cultural, a political, and an economic dimension of PIGS countries’ perception in this newspaper. We find that the initial PIGS heuristic shaped by socio-cultural attributes that mainly reflected the experiences of German tourists was only later supplemented with economic ascriptions.

KEYWORDS:

• PIGS
• Eurozone crisis
• topic modeling
• text mining
• media

1. Introduction

In the years of the Eurozone crisis, peripheral European countries (Portugal, Italy, Ireland, Greece and Spain) were often identified in the international press as the PIIGS group. This acronym aimed to condense, in a single term, the notion of economic vulnerability and risk factors common to peripheral economies of the Euro area. PIIGS was an extension of PIGS, an earlier acronym of the four Southern countries that is the focus of this article. This had emerged in the mid-1990s, during the harsh negotiations over the terms and conditions for the accession to the Economic and Monetary Union (Gros 2000). Apparently, the first use of the term was a Wall Street Journal article published on November 6, 1996. However, the idea of a Southern European country cluster (occasionally including France) with a pejorative connotation – e.g. the popular concept of ‘Club Med’ countries – had been around since the early 1990s and had been used in the context of the process of monetary integration (Brazys and Hardiman 2015).

The acronym PIGS is an example of country grouping heuristic, a cognitive shortcut based on the categorization of countries into peer groups with (alleged) common characteristics. Country grouping has come to the attention of researchers interested in the social construction of ideas that provide a framework for the understanding of the world. This literature sees ideas working as templates that enhance the interpretation of economic and political realities and may possibly influence the choices and decisions made by actors – whether investors, voters or policymakers (Blyth 2003; Brazys and Hardiman 2015). In this perspective, the term PIGS has special characteristics that make it an especially powerful cultural template, as it conveys a derogative judgement of value that converts the economy into a morality play with all its burden of identities and stereotypes (Fourcade 2013).

The PIGS heuristic reflects a widespread Northern perception of a cultural and historical divide with the South. As Jürgen Stark, a former German member of the European Central Bank board, wrote recently: ‘Different economic points of view stem from historical experience and cultural peculiarities,’ pointing explicitly to the typically German value of individual responsibility as the determinant of a fundamental divergence with other European countries (Stark 2015). This is not an isolated case. The discourse of German policymakers on the Eurozone crisis was dominated by the ideas of Ordoliberalism, a socio-economic theory with deep roots in Protestant ethics. A key notion here is that one-way solidarity only creates moral hazard; implying that the transfer of resources and the mutualization of debt would be acceptable only if debtor countries stop ‘living beyond their means’ (Hien 2019; Akaliyski 2018; also Elliott 2000). Across the entire spectrum of the German press, the policy conflict between the core and the periphery of the Eurozone was rationalized as a ‘clash of cultures’ (Brunnermeier, James, and Landau 2016). By emphasizing diverging histories and identities, the public discourse on the Eurozone crisis revived essentialist topoi that degraded Southern countries as champions of backwardness, laziness, irrationality, corruption, inefficiency and wasteful spending – a reductionist and culturalist discourse leading to a ‘calibanization’ (Ervedosa 2017) or ‘racialization’ (Van Vossole 2016) of the Südländer. Now, in the wake of the Covid-19 outbreak and debates about Coronabonds, these perceptions prove their stickiness once again.

In this paper, we aim to explore when and how the idea of a ‘European South’ emerged historically as a cultural construct in German public discourse. Was the South always perceived as an economic, social and cultural ‘whole’? Or is this a relatively novel notion constructed around some specific events and characteristics? The emergence of a European South heuristic can be empirically observed and analyzed through Topic Models, algorithms that allow for a machine-assisted reading of large collection of documents (known as ‘corpora’) in order to automatically infer content from them (Blei, Ng, and Jordan 2003; Blei and Lafferty 2007). We apply a specific type of Topic Model, a so-called Structural Topic Model (STM), to a corpus composed of the entire collection of articles published in the German weekly magazine, Die Zeit, from 1946 to 2009.¹ STM has been recently used to model the framing of international newspapers (Roberts, Stewart, and Airoldi 2016), as well as other forms of modern media communication such as Twitter feeds (Lucas et al. 2015), but to our knowledge, this paper is the first one to apply STM to articles published in Die Zeit. In contrast to standard Topic Modeling exercises, STM permits users to incorporate specific metadata, defined as information about each text contained in the corpus (e.g. year of publication), into the model. Based on this model, the method allows us to discover key medial discourses and their evolution over time.

Simple collocation analysis applied to the corpus of Die Zeit has shown that the strong association of Spain with other Southern European countries has been an ongoing characteristic of the German medial discourse for more than sixty years and that this association has cultural, political, and economic dimensions (Garrido 2018). All three discourses (cultural, political, economic) regarding PIGS countries can be captured empirically by the estimated STM presented in this paper. We find that the initial PIGS heuristic shaped by socio-cultural attributes that mainly reflected the experiences of German tourists was only later supplemented with economic ascriptions, which in turn provided the basis for the moralizing medial discourse accompanying the Eurozone crisis.

The outline of this paper is as follows: In Section 2 we introduce the new dataset on Spain’s image in the German press, including descriptive statistics and word correlations that allow us to characterize the underlying discourses related to PIGS countries. In Section 3 we provide an introduction to STM and details of the specific implementation followed in this paper. Sections 4 and 5 cover our main results related to STM, accompanied by several visualizations that allow for an understanding of the temporal dimension of the cultural, political and economic discourses regarding PIGS countries as expressed in Zeit articles. Section 6 concludes.

2. The ZEIT corpus: construction and descriptive statistics

To conduct Text Mining analysis, including the estimation of a STM, we use the whole population of Zeit newspaper articles about Spain that were written between 1946 and 2009 (2,443 articles in total). The articles were gathered via the open-source software DiaCollo, which allows to search through the Deutsches Textarchiv, a large collection of German texts published in different fields, including the entire collection of articles published in Die Zeit (when the present analysis was conducted, Zeit articles were accessible up until 2009). Preliminary exploration of DiaCollo suggests that Spain is the most significant country for analyzing the PIGS heuristic in Zeit articles, which is why this country is used as ‘anchor’ for creating the corpus.² The distribution of the individual articles within this corpus is plotted in Figure 1. As can be seen, there has been a steady increase in Zeit articles dealing with Spain, with a particular strong rise starting in the early 1990s. Since this distribution is highly unequal, we have to normalize measures such as frequency counts and utilize text mining methods such as STM that can statistically accommodate unequal distributions. Since we were able to access Zeit articles through DiaCollo only until 2009, the resulting corpus does not contain any articles published during the peak of the Eurozone crisis. We will thus concentrate predominantly on the initial PIGS heuristic associated with the Southern countries (instead of the broader PIIGS heuristic that became relevant only recently).

Figure 1. Distribution of selected Die Zeit (1946–2009) articles over time

Each article touches upon different topics and frequently mentions other PIGS countries besides Spain. In order to capture this country grouping heuristic, we create an additional variable that records if another PIGS country was mentioned in addition to Spain, which is, by definition, always mentioned. By comparison, Ireland was mentioned only very infrequently, which provides another reason why we focus on the initial PIGS countries in our analysis.³ The year when each article was written was also recorded. Thus, for each Zeit article, there is metadata on the year it was written in and on associated PIGS countries.⁴ As part of the pre-processing of the corpus, we convert all words of the Zeit articles to lower case, remove punctuation, remove stop-words, and remove numbers. After removing infrequent terms in order to decrease the computational time and to ensure that the results are not affected by outliers, the final corpus of Zeit articles consists of 2,443 articles, 14,521 terms and 807,072 tokens.

We start by plotting the relative frequency with which Spain and the PIG countries appear within the Zeit corpus (Figure 2). Spain is initially particularly associated with its geographical neighbor Portugal. Besides the geographical proximity, this association in the early periods was probably driven by the same political climate in these two countries at that time (both featured dictatorial systems, with Salazar in Portugal and Franco in Spain). From the 1970s onwards, Spain is increasingly associated with Greece. Again, this could be driven by political developments, namely the military coup in Greece (1967–1974). Over the whole period up until today, however, the most constant and strong association of Spain with another country was with Italy, the other large Southern economy in Europe. Given certain prominent discourses in the Zeit articles discussed later, such as football and European integration, this grouping of Spain and Italy is not surprising.⁵

Figure 2. Relative frequency of ‘spanien [Spain],’ ‘portugal [Portugal],’ ‘italien [Italy],’ and ‘griechenland [Greece]’ in the Zeit corpus

How important a word is in a Zeit article depends not only on its frequency in that article but also on how often it is commonly used. This is taken into account in the tf-idf method, with tf denoting ‘term frequency’ and idf denoting ‘inverse document frequency.’ When combining tf and idf, a term’s importance is adjusted for how rarely it is used. This means a text is analyzed based on its position within the overall corpus (Silge and Robinson 2017). A word in one of the Zeit articles is therefore important for this article if it appears frequently therein, but less frequently in other articles by way of comparison. Figure 3 plots the ten most characteristic words, as measured by tf-idf, for each PIGS country.

Figure 3. Most characteristic words for each PIGS country (as measured by tf-idf)

Interestingly, Figure 3 reveals that the three PIG countries are highly correlated with ‘Brüssel,’⁶ pointing towards the role of the EU in these countries’ recent history and foreshadowing the difficulties that Southern European countries would experience during the Eurozone crisis when they repeatedly clashed with EU institutions over their debt management strategies. Moreover, we find evidence for the hypothesis that Germans’ perception of these countries was driven by ‘soft’ cultural factors such as car-enabled tourism or sport, given that the country profiles feature words related to these socio-cultural issues (‘urlaub [holiday]’ in the case of Greece, ‘piech [Ferdinand Piëch, the long-time chairman of Volkswagen],’ ‘toyota [Toyota]’ and ‘fußball [football]’ in the case of Italy). Finally, it should be noted that the rather curious word profile of Spain (prominently featuring medical terms such as ‘psychotherapie [psychotherapy]’ and ‘therapeuten [therapists]’) is a statistical artefact, arising from the fact that our measures does not capture the most important words (as measured by tf-idf) for all articles dealing with Spain (since this would equal the whole corpus), but captures the most important words that distinguish the Zeit articles that feature only Spain in contrast with all other articles (that include PIG countries).

Finally, we can investigate the country group heuristic by analyzing correlations between the PIGS countries’ names and other words used in the Zeit articles. While simply calculating the most common co-occurring words (such as ‘Spanien [Spain]’ and ‘spanische [spanish]’) is not particularly meaningful since they are often also the most common individual words, examining correlation among words reveals how often they appear together relative to how often they appear separately. In particular, we investigate word correlations via the ‘phi coefficient,’ a common measure for binary correlation. The focus of the phi coefficient is how much more likely it is that either both word X and Y appear or neither do, than that one appears without the other (Silge and Robinson 2017). We calculate the phi coefficient between words based on how often they appear in the same Zeit article. In particular, we pick the four PIGS countries and find the other words most associated with them (Figure 4). All correlation values are listed in Table 1.

Figure 4. Correlations between the PIGS countries’ names and other words in the Zeit corpus

Table 1. Top 20-word correlations for each PIGS country

Word 1	Word 2	Correlation	Word 1	Word 2	Correlation
spanien [Spain]	spanischen [Spanish]	0.55123149	spanien [Spain]	von [from]	0.14220115
spanien [Spain]	spanische [Spanish]	0.53478972	portugal [Portugal]	spanischen [Spanish]	0.14165065
spanien [Spain]	madrid [Madrid]	0.52195139	spanien [Spain]	blieb [remained]	0.14149496
spanien [Spain]	spaniens [Spain]	0.47830791	spanien [Spain]	hatte [had]	0.14069080
italien [Italy]	italienischen [Italian]	0.40274299	portugal [Portugal]	land [country]	0.14062520
portugal [Portugal]	spanien [Spain]	0.29316142	griechenland [Greece]	jugoslawien [Yugoslavia]	0.14031144
spanien [Spain]	portugal [Portugal]	0.29316142	italien [Italy]	prozent [percent]	0.13931162
italien [Italy]	rom [Rom]	0.28638436	italien [Italy]	bei [at]	0.13884018
portugal [Portugal]	april [April]	0.27969935	portugal [Portugal]	der [the]	0.13821314
italien [Italy]	deutschland [Germany]	0.21205840	portugal [Portugal]	den [the]	0.13799480
italien [Italy]	frankreich [France]	0.20422301	portugal [Portugal]	die [the]	0.13748452
portugal [Portugal]	gemeinschaft [community]	0.20255639	italien [Italy]	für [for]	0.13650519
griechenland [Greece]	portugal [Portugal]	0.20085360	portugal [Portugal]	dessen [whose]	0.13615900
portugal [Portugal]	griechenland [Greece]	0.20085360	portugal [Portugal]	politisch [political]	0.13536919
italien [Italy]	spanien [Spain]	0.17553164	portugal [Portugal]	hatte [had]	0.13519505
spanien [Spain]	italien [Italy]	0.17553164	italien [Italy]	deutschen [German]	0.13477129
griechenland [Greece]	tourismus [tourism]	0.17521422	italien [Italy]	bis [to]	0.13185489
portugal [Portugal]	hoffnung [hope]	0.17481122	griechenland [Greece]	türkei [Turkey]	0.12974305
griechenland [Greece]	beitritt [accession]	0.17127321	italien [Italy]	nach [to]	0.12875464
portugal [Portugal]	beitritt [accession]	0.16708490	griechenland [Greece]	probleme [problems]	0.12524959
griechenland [Greece]	gemeinschaft [community]	0.16697462	italien [Italy]	im [at]	0.12140637
spanien [Spain]	jene [that]	0.16486451	griechenland [Greece]	irland [Ireland]	0.12105895
portugal [Portugal]	25 [25]	0.16414523	italien [Italy]	zwei [two]	0.12054864
spanien [Spain]	als [as]	0.16020204	griechenland [Greece]	neun [nine]	0.12004579
griechenland [Greece]	spanien [Spain]	0.16017453	italien [Italy]	nur [only]	0.11988533
spanien [Spain]	griechenland [Greece]	0.16017453	italien [Italy]	sind [are]	0.11929763
griechenland [Greece]	länder [countries]	0.15832020	italien [Italy]	aus [from]	0.11883855
portugal [Portugal]	revolution [revolution]	0.15759666	griechenland [Greece]	hotels [hotels]	0.11882674
spanien [Spain]	und [and]	0.15154477	italien [Italy]	jugoslawien [Yugoslavia]	0.11776483
portugal [Portugal]	zur [to]	0.15120866	italien [Italy]	österreich [Austria]	0.11758125
spanien [Spain]	bach [to]	0.14957962	italien [Italy]	vier [four]	0.11713370
spanien [Spain]	die [the]	0.14698504	italien [Italy]	die [the]	0.11480179
spanien [Spain]	wirtschaftlichen [economic]	0.14698414	griechenland [Greece]	zwölf [twelve]	0.11352440
spanien [Spain]	sich [itself]	0.14627691	griechenland [Greece]	nato [NATO]	0.10355148
spanien [Spain]	dem [the]	0.14600613	griechenland [Greece]	touristen [tourists]	0.10018001
spanien [Spain]	der [the]	0.14584419	griechenland [Greece]	ländern [countries]	0.09806431
portugal [Portugal]	demokratie [democracy]	0.14474468	griechenland [Greece]	brüssel [Brussels]	0.09794863
portugal [Portugal]	wirtschaftlichen [economic]	0.14442181	griechenland [Greece]	ausländer [foreigners]	0.09495108
portugal [Portugal]	erreichen [reach]	0.14351946	griechenland [Greece]	dänemark [Denmark]	0.09331392
spanien [Spain]	begann [began]	0.14241627	griechenland [Greece]	wirtschaftlichen [economic]	0.08963311

From Figure 4, we see that Greece is associated with tourism and economic problems, Italy with Rome and Germany, Portugal with the European Union and revolution, and Spain with Madrid and economic aspects. Most interesting for our purposes is the fact that all PIGS countries have high correlations with each other: Greece is strongly correlated with Spain and Ireland (and also Turkey, Yugoslavia and Denmark), Italy with Spain (as well as Germany, France and Yugoslavia), Portugal with Spain and Greece, and Spain with all three PIG countries (Table 1). This implies that German media, as approximated through the Zeit corpus, often referred explicitly to other Southern European countries when discussing one of the PIGS countries, and particularly when discussing Spain, thereby perpetuating this country group heuristic. Having surveyed the information that can be obtained through standard descriptive statistics and correlation analysis, we are ready to estimate a STM.

3. Structural topic modeling

We begin by providing a brief overview of the STM model (cf. Roberts, Stewart, and Tingley 2019).⁷ Just as the initial Topic Models designed by Blei, Ng, and Jordan (2003), STM is a generative model of word counts. In short, this means that the algorithm defines a data generating process for each document and then uses the semantic data, as captured in our Zeit articles, to find the most likely values for the parameters within the model. Given the number of topics that should be produced, the model places together terms that appear in the same Zeit article more frequently than one would expect by chance. In an iterative process, each word is assigned to a topic, which is therefore defined as a mixture over words, with each word having a probability of belonging to a topic. Thus, a Zeit article is understood as a mixture over topics, meaning that a single article can consist of several different topics to a varying degree.

So far, this resembles the typical structure of a classic Topic Model. The additional advantage introduced by STM is the fact that it allows to incorporate metadata into the estimation framework. In particular, metadata covariates for topical prevalence allow the observed metadata to influence the frequency with which a topic is discussed.⁸ Since different Zeit articles are associated with different cultural, political, and economic contexts, it is natural to want to allow this prevalence to vary with the metadata that we have about our articles. We will let topic prevalence be a function of the country variable, which codes references to other PIG countries, and the variable year, which is an integer measure of years running. We enter in the variables additively, by allowing for the year variable to have a non-linear relationship in the topic estimation stage. Moreover, we allow the country variable to also affect topical content, i.e., the word rate use within a given topic.

Before the algorithm can start its work, we have to specify the number of topics to be estimated. Which number is ideal? Recently, several metrics have been proposed with which to identify the correct number of topics.⁹ However, they provide the optimal number of topics only in a statistical sense. Most researchers in the social sciences therefore simply run the algorithm several times and compare the respective outputs (cf., Wehrheim 2019; Ferri, Lusiani, and Pareschi 2018). In short, there is not a ‘right’ answer to the number of topics that is appropriate for a given corpus and ultimately, one should decide for the most coherent and helpful output given the initial research question. Following several estimations with different topic numbers (5, 10, 15, 20, 50), we opt for a 15 topic STM model.¹⁰ The model is set to run for a maximum of 75 iterations using a self-selected seed. We use so-called ‘spectral initialization’ that guarantees that irrespective of the seed that is set, the same output will be generated (Arora et al. 2013).

To sum up, we estimate a structural topic model for the Zeit corpus that allows us to discover the central topics discussed in these articles. By incorporating data about individual articles, such as year of publication and PIG country being addressed, we can analyze these topics in relation to both their development over time and their role within the emerging PIGS heuristic. In our specification of the model,¹¹ we estimate 15 topics, with topic prevalence being influenced additively by year and country, and topical content being influenced by country. Through ‘spectral initialization,’ we ensure that our findings are replicable and transparent and can thus serve as a starting point for further research.

4. Results I: topics

We explore the topics that have been estimated via two approaches: first, we aim to understand the content of the fifteen topics, and then, we visualize their respective proportion within the corpus of Zeit articles. As stated earlier, Topic Models treat topics as distributions over words. Accordingly, the computational output consists of the highest probability words associated with each topic. However, these groups lack any kind of label. Historians usually assign a single label to each topic in order to facilitate an aggregate analysis of the corpus’ content and focus particularly on the topics that can be coherently labelled (cf., Küsters, Volkind, and Wagner 2019). For instance, in mining newspaper articles of the Richmond Daily Dispatch, Nelson looks at a topic with words like ‘treasury,’ ‘bonds,’ and ‘interest,’ and identifies it with the label War bonds (Nelson 2019). Likewise, by reading through the estimated word lists as well as examining actual articles that are estimated to be highly associated with each topic, we are able summarize each Zeit topic with a manually specified label. The results are given in Table 2.

Table 2. A topic model with 15 topics, 2,443 documents and a 14,521 word dictionary

Topic	Topic words (high probability)	Label
1	buchmesse, weine, übersetzungen, flüssigkeit, flaschen, jahrgang, kompetenzen [book fair, wines, translations, liquid, bottles, vintage, competence]	Culture I (books and wine)
2	erklärung, regierungen, angeblich, empört [declaration, governments, alleged, indignant]	Rhetoric terms
3	ezb, geldpolitik, ews, währungsunion, euro-währung, trichet, konvergenzkriterien [ecb, monetary policy, EMS, monetary union, euro currency, trichet, convergence criteria]	European monetary policy
4	viertelfinale, mönchengladbach, spielern, stürmer, borussia, brasilianer, dfb [quarter-final, mönchengladbach, players, striker, borussia, brazilian, DFB]	Football
5	gedichten, heiratete, poetischen, romane, verse, michelangelo, kloster [poetry, married, poetic, novels, verses, Michelangelo, monastery]	Culture II (literature and art)
6	arbeitslosenzahl, rückgang, anleger, arbeitskosten, sozialprodukts, jahresbeginn, unangefochten [unemployment rate, decline, investor, labour costs, social product, beginning of year, unchallenged]	Economic competitiveness
7	putsch, fidel, islamisten, hussein, anarchisten, militanten, kardinal [putsch, fidel, islamists, Hussein, anarchists, militants, cardinal]	Political turmoil
8	fluggäste, ltu, lufthansa, verkehrsminister, airways, condor, autofahrer [passengers, LTU, lufthansa, minister of transport, airways, Condor, car driver]	Travels
9	philosophische, großfamilie, training, wahrnehmen, jakob, dramen, lyrik [philosophical, extended family, training, perception, Jakob, drama, poetry]	Diverse terms I
10	steinbrück, angela, kohlendioxid, potenzial, ökonom, passt, internet [Steinbrück, Angela, carbon dioxide, economy, potential, fits, internet]	German politics
11	vollversammlung, truman, eisenhower, blei, uno, imperialismus, großmächten [assembly, Truman, Eisenhower, lead, UNO, imperialism, great powers]	International Relations
12	bungalows, trip, kreuzfahrt, halbpension, postfach, kreuzfahrten, hummel [bungalows, cruise, trip, half-board, post office box, cruises, Hummel]	Tourism
13	eg-kommission, eurokraten, schutzmaßnahmen, eg-präsidentschaft, bangemann, eg-kommissar, thatcher [EC Commission, Eurocrats, protective measures, EC Presidency, Bangemann, EC Commissioner, Thatcher]	European integration
14	chruschtschows, pluralismus, politbüro, imperialismus, großmächten, supermächte, chruschtschow [Khrushchev’s, pluralism, politburo, imperialism, great powers, superpowers, Khrushchev]	Cold War
15	vorhin, bruttosozialprodukts, dahrendorf, labour, technologische, unsere, schmidt [earlier, gross national product, Dahrendorf, Labour, technological, ours, Schmidt]	Diverse terms II

For each estimated topic, the table gives a list of the most important words (i.e. the words with the highest probability of being named within that topic) as well as the manually added label. The words are ranked by statistical importance.

Surveying the topic labels given in the table, we can identify three discourses within the Zeit newspaper corpus, spanning 2,443 articles and almost seven decades. Most importantly for our purpose, we can identify a large cultural discourse encompassing topics mostly related to Germans travelling abroad. To begin with, the Tourism topic (topic 12) encompasses revealing terms such as ‘trip,’ ‘kreuzfahrten [cruises]’ and ‘halbpension [half-board].’ More narrowly, the related topic Travels (topic 8) focuses on the act of transportation, most commonly via plane (‘fluggäste [air passengers],’ ‘Lufthansa,’ ‘airways’), but also via car (‘autofahrer [car drivers]’). The two topics labelled Culture I and II (topics 1 and 5) deal with culture in the narrow sense, that is art (‘buchmesse [book fair],’ ‘gedichte [poems],’ ‘romane [novels],’ ‘michelangelo [Michelangelo]’) and food (‘weine [wines]’, ‘jahrgang [vintage]’). As might be expected, the German view of Southern Europe, and particularly Spain and Italy, was also shaped by these countries’ great Football (topic 4) teams, which often clashed with the German national team (‘dfb [DFB, the German Football Association]’) or national clubs (‘borussia,’ ‘mönchengladbach’) in the different competitions (‘viertelfinale [quarter-final]’).

A closer look into the corpus shows that the estimated culture topics indeed accurately capture the content of many articles.¹² The economic upswing, which began with the establishment of the social market economy by Ludwig Erhard, led to a tourism euphoria in Germany, which found it clear expression in many Zeit articles: ‘Everyone is hungry for sun; this summer more than ever!,’ the director of a Hamburg travel agency is quoted in a 1962 article, explaining the fact that special trains and ships, charter planes and coaches to the South were occupied to the last place, and that of approximately 320,000 holiday travelers in that year almost two-thirds went abroad, especially to Italy and Spain (Die Zeit 1962).¹³ A good example of how this euphoric mood was linked to the PIGS countries is a 1960 article that declares that ‘even in December we do not have to turn our backs on Europe if we want to romp around in the water,’ followed by a survey of potential destinations that prominently put together the PIGS countries’ well-known tourism destinations (Die Zeit 1960). Similar tourism-related articles can be found in many other subsequent years as well, particularly during the 1960s and early 1970s (in line with the estimated temporal pattern shown later) and often written by the journalist Horst Hachmann (Die Zeit 1964; 1965; 1966; 1967; 1970; 1971; 1973).

In the literature on tourism, it is well acknowledged that certain nationalities have preferences for individual places, particularly amongst age migrants (Breuer 2002: 22). Historically, besides the Costa Blanca around Alicante, the Balearic Islands and the Canary Islands, the Costa del Sol’s coastal areas became the most popular Spanish destinations for German tourists. A good example is the site Torrox on the Costa del Sol that was opened up for tourism in the 1970s when, on the initiative of a German construction company, blocks of flats were built that were sold exclusively to Germans. Since then, the town has offered generations of German tourists and older German migrants the opportunity to pursue a leisure-oriented lifestyle (Kordel 2011). As the above-mentioned 1960 Zeit article illustrates, these touristic experiences were well reflected in German media, which initially grouped these countries together due to their geographical suitability for sunny holidays.

Of particular interest is the fact that some of these tourism-centered Zeit articles disprove the common image of the young German holidaymaker who, in the upcoming age of mass tourism, only aimed to achieve a tan as symbol of the ‘successful’ holiday, and instead point towards cultural exchanges that might have enriched the German travelers’ perception of Southern Europe. As the writer and Zeit journalist Ferdinand Ranft noted in 1966: ‘A not inconsiderable part of the young participants in organized holidays wants activities of a sporting, cultural and foreign nature,’ and in his article, he lists several examples of how German travelers could realize these experiences (Die Zeit 1966). This perception of the German tourist travelling abroad in order to enrich his cultural image of other countries also featured in a later article that described study trips to Spain, Italy, and Greece: ‘The trips are not lazy fun, but exhausting educational trips. Travelers not only want to see the world, they also want to understand it’ (Die Zeit 1971). During these trips, Germans compared the cultural images of Southern Europe that they had acquired during their childhood through hearsay or literature with the contemporary realities: ‘All the cities on the Mediterranean are the scenes of images and books that we, as children, have seen and – more and more demandingly – devoured with our torches in bed. There are accordingly many conceptions’ (Die Zeit 1972). While this confrontation with reality could sometimes lead to disappointments, as the author describes in the remainder of this Zeit article, it definitely confirms a mental process of forming country group heuristics based on cultural features.

Besides this cultural discourse, the topics estimated on the basis of our Zeit corpus reflect the most important geo-political developments during the period of investigation, thus forming a large political discourse. Starting with the national level, several themes debated within German politics (topic 10) were featured in the Zeit articles, explicitly naming German politicians (‘angela [Angela Merkel],’ ‘steinbrück [Peer Steinbrück]’). Political turmoil (topic 7) was associated with developments in more distant countries (‘fidel [Fidel Castro],’ ‘hussein [Saddam Hussein]’) and a diverse set of extreme political goals (‘islamisten [Islamists],’ ‘anarchisten [anarchists],’ ‘militanten [militants]’), which nevertheless bear some relevance for the PIGS heuristic given that Eurocommunism influenced especially the communist parties of Spain and Italy. The Cold War (topic 14) centering around the Soviet-Union (‘chruschtschow [Nikita Chruschtschow],’ ‘politbüro [politburo]’) and the US (‘Truman [Harry Truman],’ ‘Eisenhower [Dwight D. Eisenhower]’ in topic 11) was, of course, an important topic for several decades and also shaped the way that International Relations (topic 11) were framed (‘imperialismus [imperialism],’ ‘großmächten [great powers]’). Throughout our period, the development of European political institutions (‘eg-kommission [EC Commission]’, ‘eg-kommisar [EC Commissioner]’, ‘eg-präsidentschaft [EC Presidency]’) that were related to the process of European integration (topic 13) became increasingly relevant, but was also received partly critical as too technocratic and elite-centered (‘eurokraten [Eurocrats]’).

Anticipating the results from plotting these topics’ proportion over time, we focus on the two most relevant topics in the Zeit’s political discourse, namely International Relations (topic 11) and European integration (topic 13). Reading through the original Zeit articles shows that the first topic’s link to the PIGS countries seems to be rather coincidental. Many articles deal with the danger for peace posed by the Soviet Union and simply describe the similar role played by communist parties in Southern European countries, particularly Spain, Portugal, and Greece (e.g. Die Zeit 1968; 1971).

More revealing are articles dealing with European integration (topic 13), since they surprisingly anticipate many of the arguments that were later advanced in the German public discourse on PIGS countries during the Eurozone crisis. Already at the early stage of the European Community, an article in 1975 by Rolf Zundel explains to the readers of the Zeit that ‘European policy is by no means just about whether Bonn finds the right tone for its partners,’ but primarily about the ‘dosage and conditioning of Bonn payments’ in order to avoid that Southern inefficiency ‘will be made permanent’ (Die Zeit 1975). This resembles the moral hazard argument put forward by German economists and politicians during the Eurozone crisis, who worried that bailouts would set a bad example and encourage inadequate behavior among other actors (James 2017: 26).

Back in the 1970s, Zeit authors also understood that the heterogenous nature of the different European economies made them very different from the standard model of an optimal currency area. In particular, one author emphasized the ‘steep gap between the powerful and the underdeveloped members’ (Die Zeit 1977). It is exactly this economic divergence that nowadays is seen as the origin of the Eurozone’s problems (Tokarski 2019). The Germans’ later distaste for so-called Eurobonds is likewise echoed in the Zeit articles paralleling the course of European integration. When discussing the potential membership of Portugal, the Zeit author concludes: ‘Nobody wants to be outnumbered, not even Bonn; because then, to cite a particularly deterrent example, the euro would threaten the socialization of our foreign exchange stocks’ (Die Zeit 1977).

Lastly, the estimated topics allow us to identify an economic discourse in the Zeit articles. At first sight, this discourse seems surprisingly small, at least in comparison with the larger number of topics concerning cultural and political issues, but as we have seen, many political articles also touch upon economic issues. More specifically, the economic discourse encompasses two topics, namely Economic competitiveness (topic 3), consisting of words such as ‘arbeitskosten [labor costs]’ and ‘sozialprodukt [national product],’ and European monetary policy (topic 6), defined for instance by ‘geldpolitik [monetary policy]’ and ‘währungsunion [monetary union].’ The latter topic is strongly characterized by its European Community dimension (‘ezb [ECB]’) and the convergence criteria associated with membership in this community (‘konvergenzkriterien [convergence criteria]’). The wish to promote convergence has always played a central role in the historical development of European monetary integration (Franks et al. 2018). Already the Delors Report discussed the importance of greater convergence in economic performance and living standards, as well as in economic policies in order to clear the path towards a monetary union. However, with the creation of the Eurozone, the PIGS countries lost a crucial element of their options for steering economic policy, namely the option of periodically devaluing their own currency, which made it much more difficult to achieve this convergence in the long run (Tokarski 2019). Thereby, the common currency created an economic environment that advantaged the corporatist labor markets of the Northern countries and destabilized the integration of the South (Johnston 2016).

While the Zeit articles contain several references to the economic dilemma posed by the loss of the devaluation option, the adverse effects on European labor markets went largely unnoticed. The context to most articles was given by the 1992 Maastricht treaty that established convergence criteria for countries to join the euro, focusing on nominal and fiscal indicators of harmonization, including inflation, long-term interest rates, exchange rate stability, the fiscal deficit, and the government debt-to-GDP ratio. Between 1991 and 1998, several Zeit articles described and analyzed these Euro convergence criteria (e.g. Die Zeit 1991; 1993; 1994; 1995; 1996; 1998). In this reporting, the authors generally applauded the criteria, which they interpreted as ‘mainly strict requirements in terms of price and exchange rate stability and budgetary discipline,’ concluding that the ‘Federal Government and the Bundesbank were largely able to assert themselves with their high demands on the soundness of their partners’ (Die Zeit 1991). From the start, however, Zeit authors were skeptical regarding the PIGS countries’ readiness for fulfilling these criteria and joining the Euro-area: ‘Even in the longer run, some countries already seem to have no chance. The most striking examples: Belgium, Greece and Italy have such high debts that it will be impossible for them to reduce them to the required level in a few years’ time. […] Portugal, Ireland and Spain have little chance. They would have to make huge savings and thus risk enormous social tensions’ (Die Zeit 1993). In short: the transformation from culturally acclaimed tourist destination to a negative country group heuristic based on economic attributes was completed by the late 1990s. The rise of this new image of the PIGS ultimately led to a rather pessimistic reporting on the ‘specter of monetary union,’ as one Zeit journalist called it (Die Zeit 1995).

5. Results II: developments

After having gained a sufficient understanding of the topics’ content, we can turn the effects of the year variable on topical prevalence. Figure 5 plots the relationship between the respective year and the set of cultural topics, that is, topics 12, 5, 8, 1, and 4, thereby visualizing these topics’ temporal dimension. Significantly, the Tourism topic peaks during the 1960s, that is during the Golden Age of Germany’s post-war reconstruction, which allowed many Germans for the first time to travel to Southern Europe for an annual summer holiday. Still, the Tourism topic remains relevant within the Zeit corpus and even increases its share of articles again around the year 2000.

Figure 5. Cultural topics over time (as estimated by the STM)

The STM, therefore, identifies Tourism as central cultural force in the media coverage of Spain. Such subcultural encounters and the resulting mutual relationships are by no means trivial. Tourism and identity are closely related: after exploring other countries the knowledge of the differences makes more present the characters of one’s own country. This has repercussions on one’s sense of national identity and how this identity might differ from others (Lanfant, Allcock, and Bruner 1995). Especially in the process of constructing a post-war Europe, history and shared cultural heritage has played an important role in reinforcing national identities (Tunbridge and Ashworth 1996; Hall 1995). It is interesting to note that the change of the PIGS countries’ image towards a more negative economic dimension was not caused by a decline in German tourism: With 13.5% of all holidays, Spain remains among the most popular foreign destinations for Germans (Mau and Mewes 2007: 209).

Next, we turn towards the set of political topics, that is, topics 13, 14, 7, 11, and 10, and estimate their development over time. The resulting Figure 6 accurately resembles the shift in priorities from Cold War politics to European integration and finally to a medial dominance of national political issues. The latter finding, i.e., the increased mixing of articles about Spain with discussions about German politics in the last years of our corpus, might indicate the increased relevance of discussions about PIGS countries in the German domestic political discourse during the Eurozone crisis. The accompanying medial discourse portrayed the crisis as one of ‘lazy’ Southern Europeans now punished for their ‘profligate’ lives, which justified austerity as the German government’s proposed crisis solution (Petry 2013). Commentators have stressed the medial pressure experienced by Merkel’s government due to important state elections that recurrently came up during the crisis (Mahnkopf 2012: 480; Young and Semmler 2011: 8).

Figure 6. Political topics over time (as estimated by the STM)

Finally, we turn to the economic discourse, which can be approximated with the estimated developments of topic 6 (Economic competitiveness) and topic 3 (European monetary policy). The resulting visualization shows clearly that the initial focus on cultural and political topics in the Zeit’s reporting on Spain was only lately replaced by economic issues (Figure 7). This trend can be explained with the deepening of European integration that started in the 1990s, culminating in the creation of the common currency area, and the heightened concerns over the price stability and economic competitiveness of the Southern European economies during this period. Thus, the figure indicates that between 1990 and 2000, the earlier PIGS heuristic shaped by socio-cultural attributes as experienced for instance by German tourists became supplemented with economic attributions in Germany’s medial discourse. The economic attributions inherent in this adapted country group heuristic, in turn, could be later moralized in the German medial discourse accompanying the Eurozone crisis (Fourcade et al. 2013).

However, while this turning-point in the overall discourse can be clearly located quantitatively, the question whether this negative economic association is fundamentally a product of the 1990s or whether the seeds have already been sown before-hand can be answered only qualitatively, namely through a cursory reading of the Zeit articles that involve topic 3 and especially topic 6. On this basis, three phases can be distinguished: In the 1950s and 1960s, economic references to PIGS countries were very rare, and we find only one polemic article, which noted that the ‘economic and social condition’ of the populations of Spain, Italy and France ‘is largely based on discontent,’ implying for the author that they ‘very easily turn to political movements that promise to rain manna from heaven’ (Die Zeit 1956). During the 1970s and 1980s, economic associations with these countries increased significantly in the aftermath of the two oil price shocks (in line with the first peak of topic 6’s trend line) and commentators noted that, by comparison, the more market-based variety of capitalism present in Germany makes it more resilient to these shocks than the Southern countries (Die Zeit 1977). However, these comparisons were still relatively ‘neutral’ in tone (e.g. Die Zeit 1972; 1988). This changed only with the discussions surrounding the adaption of the common market and the plans for a common monetary union, as the examples from the 1990s cited earlier underline. Thus, the evidence from the Zeit newspaper suggests that the PIGS heuristic in its present understanding is a child of European economic integration.

Figure 7. Economic topics over time (as estimated by the STM)

To sum up, estimating a Topic Model such as the STM presented in this paper has three crucial advantages, especially when compared to the classic close reading of texts. First, it is an objective method which does not require the imposition of pre-defined categories. This is especially helpful when dealing with a contested, emotional issue such as the PIGS country group heuristic: by avoiding the ex-ante imposition of a certain picture or definition of the PIGS countries, the researcher can leave it to the sources, that is the corpus, to indicate which words have been associated empirically with this country. Secondly, the estimated topics are explicit, so other researchers can reproduce the analysis. Thirdly, the computational power allows to understand and structure large corpuses of texts, thereby facilitating the complementing qualitative analysis via traditional close reading. In this case, STM allowed us to survey 2,443 articles published between 1946 and 2009 in the German newspaper Die Zeit. As has become clear as well, these quantitative techniques still depend on the researcher’s judgment and thus should be seen as a helpful complement, and not substitute, in historic research. Moreover, since our corpus was constructed on the basis of one newspaper, future research should take this evidence as a starting point for further analysis of other media in other countries.

6. Conclusion

The onset of the Eurozone crisis popularized the PIGS acronym, echoing allegedly ‘lax’ Southern European attitudes towards inflation and state deficit and contrasting Portugal, Italy, Greece, and Spain unfavorably with their Northern European neighbors. In order to investigate the historic formation of this country group heuristic, we apply a Structural Topic Model to all 2,443 articles published between 1946 and 2009 in the German newspaper Die Zeit that mention Spain. The estimated topics and their evolution over time confirm a strong association of Spain with other Southern European countries as an ongoing characteristic of the German medial discourse for more than sixty years. Moreover, the STM allows us to distinguish and characterize a cultural, a political, and an economic dimension of PIGS countries’ picture in German media. We find that the initial PIGS heuristic shaped by socio-cultural attributes that mainly reflected the experiences of German tourists was later supplemented with economic ascriptions, which in turn provided the basis for the moralizing medial discourse accompanying the Eurozone crisis.

This conclusion has been derived by analyzing quantitatively and qualitatively a single newspaper in a single, albeit important, member state. It is therefore up to future research to complement this picture by tracing the emergence and development of the PIGS heuristic in the media of other Northern countries. During the final revision of this article, this research agenda has gained tragic importance: Like the Eurozone crisis beforehand, the Covid-19 outbreak is again straining solidarity within the European Union and perpetuating dangerous misunderstanding among citizens. In such times, acknowledging and reflecting the historical stickiness of certain perceptions of Southern Europe is highly relevant for finding a common path forward.

Acknowledgments

We would like to thank all researchers associated with the HERA-funded project “Uses of the Past in International Economic Relations” (UPIER) for comments on our work, especially Stefano Battilossi and Catherine Schenk. We would also like to acknowledge funding from HERA, which allowed Elisa to work on the project as a postgraduate researcher at Universidad Carlos III. The views expressed in this paper, and all errors and omissions, should be regarded as those solely of the authors and are not necessarily the views of the affiliated institutions. For more information on UPIER visit: www.upier.web.ox.ac.uk.

Disclosure statement

No potential conflict of interest was reported by the authors.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

1. This is done via the R language stm package.

2. DiaCollo can be accessed via: https://www.dwds.de/d/k-zeitung#zeit (accessed: 06.04.2020). Choosing Spain as ‘anchor’ when creating the corpus in June 2019 had three reasons. First, one of us had already worked on the image of Spain in German media and this earlier work suggested that Spain plays an important role for understanding the PIGS heuristic (Garrido 2018). Secondly, preliminary analysis of Zeit articles (before the estimation of the STM) suggested that it was particularly in connection with Spain that other PIGS countries were mentioned. Thirdly, searching for ‘spanien’ (Spain) in DiaCollo resulted in a high number of references that promised a large enough sample size. As of 6 April 2020, DiaCollo allows to access Zeit articles up until 2018 and in this larger timeframe, Spain is mentioned 52,046 times (Portugal 17,616 times; Italy 70,438 times; and Greece 61,749 times).

3. This opens up the question of how significant this exclusion is. As a small, relatively ‘underdeveloped’ peripheral economy for most of the 20^th century, is it possible to imagine that that there are Zeit articles that might exhibit similar negative perceptions of Ireland but do not show up in articles explicitly mentioning Spain. However, our cursory reading of the Zeit articles dealing with Ireland suggests that such a heuristic was not present: The articles in our database mentioning Ireland mostly refer to economic aspects in a neutral, descriptive tone (agricultural production, or participation in German fairs); there are also some political articles (about Ireland’s relationship to England, and Great Britain’s potential membership in the European community). The only constant reporting in which Ireland was associated with a developing country that needed to carry out structural reforms concerned the discussions on the expansion of the common European market. In this case, however, Ireland was listed not only together with Spain and Greece but also with such heterogeneous countries as Iceland, Turkey, and Finland, which implies that this reporting cannot be seen as a direct predecessor of the PIGS heuristic. A negative tone, e.g., in connection with budget deficits or perceptions of the Irish economic culture, foreshadowing the later association with the PIGS, cannot be detected. This finding makes the later inclusion of Ireland in the PIIGS heuristic all the more interesting. Further research should, therefore, aim to shed light on the mechanisms behind Ireland’s association with the Southern peripheral economies.

4. In our database, each article is a row in a.csv file, with the text contained in a variable called text.

5. As can be also seen from the figure, the frequency with which Spain was mentioned in articles over Spain declined over time.

6. Here, and in the remainder of this paper, we use single-inverted commas to refer to explicit terms in the corpus (so-called token) in order to distinguish them from the estimated topics (which will be shown in italics and capitalized).

7. For technical details, see: https://cran.r-project.org/web/packages/stm/index.html (accessed: 06.04.2020). For more examples, see: https://www.structuraltopicmodel.com (accessed: 06.04.2020).

8. Note that STM allows using topical prevalence covariates, a topical content covariate, both, or neither.

9. See, e.g., the R package ldatuning developed by Murzintcev Nikita: https://cran.r-project.org/web/packages/ldatuning/vignettes/topics.html (accessed: 06.04.2020).

10. A smaller or larger number of topics merely increases the number of vague topics with conflicting words that are difficult to interpret.

11. In particular, the inputs to the stm function (of the R stm package) were specified as follows: documents = out$documents; vocab = out$vocab; K = 15; prevalence = ~ country + s(year); content = ~ country; max.em.its = 75; data = out$meta; init.type = ‘Spectral’; gamma.prior = ‘L1.’ For an algebraic explanation of the model, see Roberts, Stewart, and Tingley (2019).

12. As a starting point, we read all articles that contained the most significant words of the respective topics as estimated by the STM and after analyzing these articles, we had gained a sufficient understanding of the respective topics allowing us to carry on the corpus search with additional, manually created, keywords.

13. All quotations from Die Zeit are own translations. As primary sources, all Zeit articles quoted from our corpus are not included in the bibliography, since we accessed them only through the DiaCollo search function when creating the corpus and not through individual copies of the newspaper. Since the references given by DiaCollo do not allow for a precise localization of the respective article in the newspaper and many of the articles do not feature an author name, they cannot be cited in the usual manner. Moreover, even in the case where an author name is given, the resulting citation would be not distinguishable from a secondary source. We thus opt for citing the Zeit articles, our primary sources, by referring to the publication as such (Die Zeit) and the year of publication. Additional information can be obtained from the separate reference list at the end of this article.