Connotative Associations of Electoral Candidate Mentions in Swiss Newspapers

Draft prepared for submission to the SPSR Special Issue “The 2023 Swiss National Elections” Connotative Association s of Electoral Candidate Mentions in Swiss Newspapers Zurich, 12/18/2025 Abstract To what extend the mass media cover election campaigns fairly or, on the contrary, whether they cover elections in biased way , is a perennial debate among researchers, experts, political parties and citizens. We study the coverage of election candidates in the German -Swiss media for the federal elections of 2015, 2019 and 2023 . By fine -tuning a large language model and examining the generated data, we show that the coverage of political candidates varies systematically according to their gender, age and party affiliation. The variances relate to political issues that are typically associated with the corresponding characteristics of the candidates and are mainly found in more recent election years and with new candidates . In the 2015 elections and among incumbents, it is only party identification that explains some o f the differences in the media coverage of electoral candidates . 2 Introduction The assessment of the role of the mass media (print and online newspapers, television channels or radio stations) in democratic elections has changed considerably in recent decades. Until a few years ago, the debate about the mass media as the fourth estat e revolved mainly around the extent to which the media provide (or should provide) sufficient quality information about parties and candidates (Falck et al., 2020 , Schudson, 2008) . In line with many observations, Farnsworth and Lichter (2012) resumed that a functioning democracy “depends on a fair and critical news media to investigate public wrongdoing and to evaluate objectively the claims of the self -interested partisans who populate the fields of politics. ” (Farnsworth & Lichter, 2012, 548). Hence, t he most serious shortcoming was seen as the increasing tendency of journalistic reporting to focus on individual candidates and political competition (“horse race” and “clickbait” journalism) rather than on political content (Chiang & Knight, 2011, Banducci & Hanretty, 2014) . This relatively demanding expectation of the role of the mass media in election campaigns seems to have given way to a more sober understanding over the last years . Recent studies increasingly focus on the detection of biased reporting or even disinformation (Rodrigo -Ginés, Carrillo -de-Albornoz, & Plaza, 2024) . On the one hand, this is related to increasing political polarisation and the sharp rise of populism, both of which also lead to targeted campaigns against the media (Hameleers, 2021, Soontjens, Van Remoortere, & Walgrave, 2021) . On the other hand, it is linked to changes in the media system such as increasing competitive pressure and, above all, the emergence of online news and social media (Agirdas, 2015, Spinde, et al., 2023) . 1 A consistent argument is that if the media are indeed becoming more biased, they are providing less of the information that voters need to properly evaluate politicians and parties and hold them to account in elections (Wolton, 2019 , Falck et al., 2020 , Lichter, 2017). We are able to examine part of this conclusion by looking at newspaper reports on 1 News read online and shared on social media is now the main source of information for around half of Swiss citizens, compared to around a third ten years ago (Thommen, Eichenberger, & Sasso, 2024) . 3 candidates for the last three federal election campaigns in Switzerland (2015, 2019 and 2023). From these reports, we compile text paragraphs mentioning election candidates, which are, in combination with meta -data on the candidates and election campaigns, used to fine -tune a large language model (LLM). We subsequently let the LLM generate 2.6 million text paragraphs for different demographic and political characteristics of electoral candidates and explore these texts by looking at the connotative meanings of candidate mentions. In the follow ing, we first discuss the state of the literature in political communication on candidate -related differences in media coverage. We then discuss our data and methodological approach before presenting the results. Electoral candidates in the media Our point of departure is the reciprocal relationship that the journalists and the candidates are locked into with each other (Midtbø, 2011) . Politicians generally depend on journalists for publicity, while journalists, in turn, depend on politicians to provide interesting news that allows them to entertain their audiences and survive financ ially (Strömbäck & Dimitrova, 2006) . In recent years, this reciprocal relationship has evolved through developments such as increasing economic competition and media concentration, and the emergence of fast -paced news formats (Esser & Matthes, 2013, Hamborg, Donnay, & Gipp, 2019) . Particularly in light of the increasing pressure to prepare their news easily for publication on social media (Spinde et al., 2023), journalists are thought to be increasingly introducing unwelcome forms of reporting such as stereotyping, scandalisation or negativity (Kang & Yang, 2022) . However, as Nyhan (2012) rightly points out, there are still many journalistic norms and practices that limit the extent to which journalists allow such bias. We therefore suggest that while journalists may well introduce bias (unconsciously or consciously) into their reporting, m ost are still committed to providing voters with at least minimal information about a wide range of candidates in a way that enables voters to evaluate the political personnel they elect (Norris, 1999, Bennett , 1988). 4 Journalists also encounter politicians who are increasingly experienced and competent in dealing with the media. In this respect, Midtbø (2011) distinguishes betw een new and established candidates. New candidates depend on media attention in order to jumpstart their political career . They therefore increasingly seek the media spotlight with strategically planned campaigns. Established candidates , however, are “blame -avoiding politicians” (Mitbø, 2011, 226) . They weigh every word and are quick to blame the media because they are worried about their reputation. In line with this, Soontjens, Van Remoortere and Walgrave (2021) have shown that this 'hostile media effect', the idea that the news is biased against o ne's own party, is common among politicians. Despite these profound changes in the way journalists deal with politicians , precise empirical results on the strength , nature and evolvement of differences in the mass media coverage of elect oral campaigns are still rare . Therefore, we ask : How are differences in media content related to demographic and political characteristics of election candidates ? Possible links between differences in media content and varying news production processes in newsrooms (c.f. Agirdas, 2015) are beyond the scope of the study. The same applies to the possible effects of such differences on newspaper readership and thus on the course of the elect oral contest (Lengauer & Johann, 2013) . In addition, we cannot provide information on other types of differences in media coverage , especially in relation to gatekeeping , i.e. inequalities in the visibility of candidates in the media (Rodrigo - Ginés, Carrillo -de-Albornoz, & Plaza, 2024, Groeling, 2013) . Finally, we provide evidence only for the media system as a whole, and not for individua l media titles or different editorial formats in the newspapers (Eberl, 2019, Tresch, 2012) . Connotative asso ciations of candidate mentions For the purposes of our study, we assume that the content -related evaluation of the electoral candidates by journalists is defined in the context of their mentions. Linguistically, this contextual evaluation can be understood as the connotative associations of the candidate's mentions, which can be localised by the most appropriate words surrounding 5 this mention (Kroeger, 2022) . A different connotative meaning for different candidates may or may not be a bias. It can also signal a different evaluative statement that is informative for voters. Following Gross, Shalizi and Gelman (2012), we refer to bias, or misrepresentation, when the news describes the characteristics or behaviour of candidates on the basis of their personal characteristics rather than on the basis of their expertise or experie nce relevant to the office for which they are running. Greene and Lühiste (2017), for example, report gender bias in media coverage of European elections. They show that media reinforce specific stereotypes by linking “easy -to-observe descriptive traits, s uch as gender”, to issue statements that are “historically connected to these groups” (Greene & Lühiste, 2017, 717). Conversely, we can speak of journalistic evaluation when the information published refers to relevant skills and experience without being influenced by specific personal characteristics of politicians. In addition to gender, the age of politicians is a fundamental personality characteristic of candidates that, to our knowledge, has not yet been examined for differences in media coverage. And in terms of gender, when it comes to how the media report on female politicians compared to male politicians, the findings in the literature are still, at least to some extent, different (Van der Pas & Aaldering, 2020) . For example, while Midtbø (2011) measures a more positive tone in the coverage of female politician s, Lavery (2013) finds little of this. And while Gessler, Gilardi and Kubli (2024) as well as Greene and Lühiste (2017) measure a strong connection between news about female politicians and gender -specific topics such as gender equality policy, the results by Rohrbach, Fiechtner, Schönhagen and Puppis (2020) do not provide enough comparable evidence . The second major strand of literature on differences in media coverage about candidates in election campaigns looks at partisan imbalances (see Groeling, 2013), and again the results are inconclusive. Some studies have found that the media are generally linking certain parties and candidates to specific issues and that they report more positively on certain parties than others (Falck et al., 2020) . Other studies, on the other hand, have provided comparable evidence only for individual media titles, not for the media 6 system as a whole (Lengauer & Johann, 2013) . Despite this uncertainty, we hypothesise the following: H 1: Demographic (age, gender) and political (party affiliation) characteristics of electoral candidates are linked to connotative associations that evoke issues historically linked to the groups associated with these characteristics. Kim, Eunji, Lelkes and McCrain (2022) and Rohrbach, Fiechtner, Schönhagen and Puppis (2020) rightly point out that intervening variables moderate the relationship between political and demographic characteristics and media coverage. On the one hand, the differences may vary according to the professional experience of the politicians (Soontjens, Van Remoortere, & Walgrave, 2021) . For example, it seems plausible that younger politicians who already have parliamentary experience do not receive any different kind of media attention than incumbents of a different age. We therefore expect the following second hypothes is to hold : H2: The connection between candidate characteristics and differences in media coverage is moderated for incumbents. Data and Methods Previous approaches to measuring candidate -specific differences in political coverage have focused on manual annotations , topic models or classifications using statistical models, network algorithms, transfo rmer models or LLMs (c.f. Cruz, Rocha, & Cardoso, 2019, Alizadeh, et al., 2025) . With the exception of topic models, all of the approaches used so far rely on human annotation, either to generate training data or directly to generate results. For this study, w e decided to bypass the manual annotation ste p because it has led to persistent criticism of subjectivity in the definition and conceptualisation of differences in media coverage (Groeling, 2013 , Nyhan, 2012) . We therefore propose an inductive approach involving the fine -tuning of an LLM followed by data augmentation. Our study covers the election campaign reports of most German -Swiss media titles (c.f . Wüest, Krell, Wirz, & Meier, 2024 , and table A.1 i n the Appendix ). Unfortunately, 7 single titles that are important for the political discourse in Switzerland , such as the Repub lik, are missing , because they are not listed in available databases . On the whole, however, our sample covers the entire spectrum of the German -Swiss newspaper landscape, especially in terms of political orientation and type of newspaper. In a first step, we select all articles in which at least one candidate was mentioned in the 2015, 2019 and 2023 federal elections . The analysis was then restrict ed to the text passages surrounding the mention s of candidates, i.e. two sentences before and after the sentence with the mention of a candidate. This enables us to link the media's use of language directly to the personal characteristics of the candidates in order to establish their connotative associations . In total , we are able to use 411 271 text paragraphs . Table A. 1. in the appendix lists the number of these text passages for each newspaper title and in each election year . The subsequent step is to finetune the LLM LL äMmlein 120M 2 with the text paragraphs. Larger versions of th e LLäMmlein model rank as the best decoder model for German, consistently matching or surpassing models with similar parameter sizes. Furthermore, the 120 million parameter model is small enough so it can run on any laptop computer , which is why o ur analysis could be replicated without any specific technological infrastructure. We ran the fine -tuning for three epochs, each with 117 991 steps. The text passages described above as well as candidate -, article - and election - specific meta -data serve as input for the fine -tuning . More specifically, each row in our initial dataset starts with the meta -data (e.g. gender and party affiliation of the politician mentioned and the election year) and ends with the text paragraph cut from the corresponding media article . At the candidate level, we include gender (male, female and non -binary defined politicians in the category female), party affiliation (the big six parties and all other parties in a residual category), incumbency status for the job they are running for, the council the politicians are running for (National Council, Council of States or both) and 2 German Tinyllama 120 million , see https://www.informatik.uni -wuerzburg.de/datascience/projects/nlp/llammlein/ [accessed on March, 18, 2025] 8 the candidates’ age (below 45 years, 45 years to 55 years, above 55 years). At the article level, we distinguish between media articles related to Swiss politics (and therefore approximately also related to the election campaign) and articles unrelated to Swiss politics. At the election level , we simply distinguish the three different election years (2023, 2019 and 2015). Figure 1 shows the distributions of the initial text paragraphs by gender, age and party affiliation and t able A. 2 in the appendix lists randomly selected initial text paragraphs including their meta -data configuration. During the fine -tuning process, the model is trained to generate text based on the meta -data and news article paragraphs. It is hypothesised that the LLM, which is designed to replicate texts with a high degree of fidelity, will also replicate the differen ces in the media reports on electoral candidates. Consequently, the resulting text distributions from these replication processes can then be examined and compared to identify shifts in connotative associations as follows: 9 Figure 1: Distribution of text paragraphs by selected meta -data 16 Figure 4B: Most prominent words in the contexts of candidate mentions by party affiliation (II) Table 1 presents a more systematic comparison of the differences in the texts. The models show correlations between demographic and political characteristics of the candidates and the cosine similarity between the different tf -idf distributions. Positive estimates indicate greater similarity between the texts , while negative estimates indicate greater differences in the tf-idf distributions . The basic model 1 shows slightly greater differences in the coverage of female candidates and in the coverage of unde r 45 -year -olds compared to the reference category of 45 -55-year -olds . In terms of the parties, the candidates of the Greens (GPS) and also the SVP exhibit slightly greater differences in their respective tf-idf distributions , while the differences for the GLP appear to be very pronounced. Together with the differences found in the lists of the most prominent words across the two genders, the three age demographics, and the six political parties, these findings offer additional subtle evidence that lends sup port to the validity of hypothesis 1. 17 Table 1: Linear regressions of candidate characteristics on text similarities Model 1 Model 2 Model 3 Estimate (p) Estimate (p) Estimate (p) (Intercept) 0.708 (***) 0.740 (***) 0.689 (***) Female -0.025 (*) -0.034 -0.017 Age (ref=45 -55) <45 -0.029 (*) -0.032 -0.016 >55 -0.001 -0.015 -0.001 Parties (ref=Mitte) GPS -0.043 (*) -0.072 (*) -0.026 SP 0.013 -0.013 0.007 GLP -0.103 (***) -0.118 (***) -0.085 (***) FDP 0.012 -0.02 0 0.048 (*) SVP -0.039 (*) -0.069 (*) -0.034 Incumbent 0.011 0.012 0.051 Council (ref=n ational council) Council of states -0.14 0 (***) -0.144 (***) -0.144 (***) Both councils -0.143 (***) -0.143 (***) -0.143 (***) Politics 0.269 (***) 0.269 (***) 0.269 (***) Election (ref=2015) 2019 0.016 -0.003 0.016 2023 0.265 (***) 0.134 (**) 0.264 (***) 2019 x Female 0.014 2023 x Female 0.011 2019 x <45 -0.01 0 2023 x <45 0.04 0 2019 x >55 0.002 2023 x >55 0.067 (*) 2019 x GPS 0.037 2023 x GPS 0.072 2019 x SP 0.03 0 2023 x SP 0.075 2019 x GLP -0.026 2023 x GLP 0.148 (**) 2019 x FDP 0.031 2023 x FDP 0.098 (*) 2019 x SVP 0.009 2023 x SVP 0.139 (**) Incumbent x Female -0.015 Incumbent x <45 -0.025 Incumbent x >55 -0.001 Incumbent x GPS -0.039 Incumbent x SP 0.009 Incumbent x GLP -0.042 Incumbent x FDP -0.069 (*) Incumbent x SVP -0.013 Adjusted R 2 0.61 0.61 0.61 F-statistic 61.54 (14/536) 29.94 (30/520) 39.76 (22/528) Levels of significance: ***=0.001, **=0.01, *=0.05 18 Model 2 uses the interaction between the characteristics of the candidates and the election year to test whether the correlations decrease over time as postulated in hypothesis 2. With respect to age and gender differences, the relationship between the tf-idf differences actually decreases as soon as age groups and genders are considered over time. The differences between the parties, in contrast , largely remain the same across the various election years. Hypothesis 2 therefore has to be rejected . A similar result can be seen for the incumbency status of candidates (see Model 3 in Table 1) . The differences in reporting between genders and age groups are considerably narrowe r for incumbents. For new ly aspiring candidates , in contrast, the differences are significant . With regard to party affiliations, however, the media do not appear to distinguish between incumbents and non -incumbents. Hence, hypothesis 3 only holds, to some extent, for age and gender. Discussion We have presented results on the newspaper coverage of election candidates in the German -Swiss media for the federal elections of 2015, 1019 and 2023 . We have fine - tuned an LLM with text paragraphs containing candidate mentions and then had the trained model generate a large number of texts. Using these texts, we were able to show that the coverage of political candidates varies systematically according to their gender, age and party affiliation. The observed variances in the coverage are associated with political issues, that are typicall y linked to the corresponding groups of candidates . Since this applies mainly to new candidates, we assume that the media use personal characteristics as shortcut for their reporting when they know too little about candidates. This is problematic because voters are then provided with too little information for their voting decision. Our analysis also has some limits. The tf -idf analysis revealed some connotati ve associations that were difficult or impossible to place in the context of the federal election campaigns. To resolve these doubtful cases, we would have to analyze them in greater depth in the generated texts, for example by analyzing their word embeddings. 19 In a next step, w e could also focus on individual media titles or different journalistic formats and thus include media -specific factors such as the type of newspaper or the target audience. Literature Agirdas, C. (2015). What Drives Media Bias? New Evidence From Recent Newspaper Closures. Journal of Media Economics 28(3) , 123–141. Alizadeh, M., Kubli, M., Samei, Z., Dehghani, S., Zahedivafa, M., Bermeo, J. D., . . . Gilardi, F. (2025). Open - source LLMs for text annotation: a practical guide for model setting and fine -tuning. Journal of Computational Social Science 8(17) . Banducci, S., & Hanretty, C. (2014). Comparative Determinants of Horse -Race Coverage. European Political Science Review 6(4) , 621–640. Bennett, L. W. (1988). News: The Politics of Illusion. Pearson Longman. Branton, R. P., & Dunaway, J. (2009). Slanted Newspaper Coverage of Immigration: The Importance of Economics and Geography. Policy Studies Journal 37(2) , 190 -292. Chiang, C. -F., & Knight, B. (2011). Media bias and influence: Evidence from newspaper endorsements. The Review of economic studies 78(3) , 795 -820. Cruz, A. F., Rocha, G., & Cardoso, H. L. (2019). On sentence representations for propaganda detection: From handcrafted features to word embeddings. Proceedings of the 2nd Workshop on NLP for Internet Freedom: Censorship, Disinformation, and Propaganda (pp. 107 -112). Hong Kong, China: Association for Computational Linguistics. Eberl, J. -M. (2019). Lying press: Three levels of perceived media bias and their relationship with political preferences. Communications 44(1) , 5-32. Esser, F., & Matthes, J. (2013). Mediatization effects on political news, political actors, political decisions, and political audiences. In H. Kriesi, S. Lavanex, F. Esser, J. Matthes, M. Bühlmann, & D. Bochsler, Democracy in the age of globalization and mediatization (pp. 177 -201). Palgrave. Falck, F., Marstaller, J., Stoehr, N., Maucher, S., Ren, J., Thalhammer, A., . . . Studer, R. (2020). Measuring Proximity Between Newspapers and Political Parties: The Sentiment Political Compass. Policy & Internet 12(3) , 1944 -2866. Farnsworth, S. J., & Lichter, S. R. (2012). Authors’ Response: Improving News Coverage in the 2012 Presidential Campaign and Beyond. Politics & Policy 40(4) , 547 -556. Gessler, T., Gilardi, F., & Kubli, M. (2024). Advocacy Campaigns and Gender Bias in Media Coverage of Elections. Political Science Research and Methods , 1-15. Giger, N., Traber, D., Gilardi, F., & Bütikofer, S. (2022). The surge in women's representation in the 2019 Swiss federal elections. Swiss Political Science Review 28(2) , 361–376. Greene, Z., & Lühiste, M. (2017). Symbols of priority? How the media selectively report on parties' election campaigns. European Journal of Political Research , 717-739. Groeling, T. (2008). Who's the Fairest of them All? An Empirical Test for Partisan Bias on ABC, CBS, NBC, and Fox News. Presidential Studies Quarterly 38(4) , 631-657. Groeling, T. (2013). Media bias by the numbers: Challenges and opportunities in the empirical study of partisan news. Annual Review of Political Science 16(1) , 129-151. Gross, J. H., Shalizi, C. R., & Gelman, A. (2012). Does the US Media Have a Liberal Bias?: A Discussion of Tim Groseclose’s Left Turn: How Liberal Media Bias Distorts the American Mind. Perspectives on Politics 10(3) , 775 –779. Hamborg F ., Donna, K. & Gibbs, B. (2019). Automated identification of media bias in news articles: An interdisciplinary literature review. International Journal on Digital Libraries 20(4) , 391-415. 20 Hameleers, M. (2021). On the Ordinary People's Enemies: How Politicians in the United States, the United Kingdom, and the Netherlands Communicate Populist Boundaries via Twitter and the Effects on Party Preferences. Political Science Quarterly 136(3) , 487 –519. Kang, H., & Yang, J. (2022). Quantifying perceived political bias of newspapers through a document classification technique. Journal of Quantitative Linguistics 29(2) , 127-150. Kim, Eunji, Lelkes, Y., & McCrain, J. (2022). Measuring dynamic media bias. Proceedings National Academy of Sciences 119(32) . Kroeger, P. (2022). Analyzing meaning: An introduction to semantics and pragmatics. Language Science Press. Lavery, L. (2013). Gender Bias in the Media. Politics & Policy 41(6) , 877 -910. Lengauer, G., & Johann, D. (2013). Candidate and party bias in the news and its effects on party choice: Evidence from Austria. Studies in Communication Sciences 13(1) , 41-49. Li, Z., Si, L., Guo, C., Yang, Y., & Cao, Q. (2024). Data Augmentation for Text -based Person Retrieval Using Large Language Models. arXiv Computer Vision and Pattern Recognition . Lichter, S. R. (2017). Theories of Media Bias. In K. Kenski, & K. Hall Jamieson, The Oxford Handbook of Political Communication (pp. 403 –416). Oxford University Press. Midtbø, T. (2011). Explaining Media Attention for Norwegian MPs: A New Modelling Approach. Scandinavian Political Studies 34(3) , 226 -249. Müller, L. (2014). Comparing Mass Media in Established Democracies: Patterns of Media Performance. Palgrave. Norris, P. (1999). Critical Citizen: Global Support for Democratic Government. Oxford University Press. Nyhan, B. (2012). Does the US Media Have a Liberal Bias?: A Discussion of Tim Groseclose’s Left Turn: How Liberal Media Bias Distorts the American Mind. Perspectives on Politics 10(3) , 767 –771. Riccardo, P., & Snyder, J. M. (2015). Empirical Studies of Media Bias. In S. P. Anderson, J. Waldfogel, & D. Strömberg, Handbook of Media Economics (pp. 647 -667). Elsevier. Rodrigo -Ginés, F. -J., Carrillo -de-Albornoz, J., & Plaza, L. (2024). A systematic review on media bias detection: What is media bias, how it is expressed, and how to detect it. Expert Systems with Applications Volume 237, Part C . Rohrbach, T., Fiechtner, S., Schönhagen, P., & Puppis, M. (2020). More Than Just Gender: Exploring Contextual Influences on Media Bias of Political Candidates. The International Journal of Press/Politics 25(4) , 692 -711. Schudson, M. (2008). Why Democracies Need an Unlovable Press. Polity Press. Soontjens, K., Van Remoortere, A., & Walgrave, S. (2021). The hostile media: Politicians' perceptions of coverage bias. West European Politics 44(4) , 991 -1002. Spärck Jones, K. (1972). A statistical interpretation of term specificity and its application in retrieval. Journal of Documentation 23(1) , 11-21. Spinde, T., Richter, E., Wessel, M., Kulshrestha, J., Donnay, & Karsten. (2023). What do Twitter comments tell about news article bias? Assessing the impact of news article bias on its perception on Twitter. Online Social Networks and Media 37 –38 . Strömbäck, J., & Dimitrova, D. V. (2006). Mediatization and Media Interventionism: A Comparative Analysis of Sweden and the United States. The International Journal of Press/Politics 16(1) , 30 -49. Thommen, S., Eichenberger, R., & Sasso, S. (2024). Medienmonitor Schweiz 2023. Zürich: Publicom & Bundesamt für Kommunikation. Tresch, A. (2012). The (Partisan) Role of the Press in Direct Democratic Campaigns: Evidence from a Swiss Vote on European Integration. Swiss Political Science Review , 287 -304. Van der Pas, D. J., & Aaldering , L. (2020). Gender differences in political media coverage: A meta -analysis. Journal of Communication 70(1), 114-143. Wolton, S. (2019). Are Biased Media Bad for Democracy? American Journal of Political Science 63(3) , 548 - 562. Wüest, B., Krell, R., Wirz, R., & Meier, C. (2024). Der Wahlkampf 2023 in traditionellen Medien. Zürich: HWZ Hochschule für Wirtschaft Zürich. 21 Appendix Table A.1: List of media titles and number of text passages by election year Newspaper title 2015 2019 2023 Newspaper title 2015 2019 2023 20 Minuten 3298 2607 1754 Swissinfo.ch 357 501 858 Aargauer Zeitung 5953 6994 5810 Schweiz am Sonntag 2225 - - Anzeiger von Uster 119 132 - Schweizer Bauer 759 933 - Appenzeller Zeitung - 5671 2162 Schweizer Familie 10 4 38 Badener Tagblatt 3 - 2702 Schweizer Illustrierte 439 381 212 Basellandschaftliche Zeitung 1905 2395 2957 Schweizer Landliebe - - 2 Basler Zeitung 5530 8986 3864 Seetaler Bote 611 432 - Beobachter - - 355 Solothurner Zeitung 2048 4433 3462 Berner Oberländer - 3885 3659 Sonntagszeitung 739 733 373 Berner Zeitung 7096 9473 4036 Sonntagsblick 825 670 368 Bieler Tagblatt 665 2060 - St. Galler Tagblatt 4959 5690 2124 Bilanz 73 86 107 Tagblatt der Stadt Zürich 24 307 4971 Blick 2615 2017 5631 Tageswoche 1183 - - Bolero - - 4 Thalwiler Anzeiger / Sihltaler - - 264 Bote der Urschweiz 2542 3346 - Thuner Tagblatt - - 3620 Langenthaler Tagblatt - - 3587 Thurgauer Zeitung 2101 5706 2736 Bündner Tagblatt 2376 1810 - Toggenburger Tagblatt - 5508 1929 Cash 2941 2239 1452 Urner Zeitung - 4869 2243 Coopzeitung 6 11 - Volketswiler 39 7 - Das Magazin 54 73 47 Walliser Bote 2569 2510 - Der Bund 6057 9487 3746 Werdenberger & Obertoggenburger 2087 4243 1983 Der Landbote 3020 4345 1653 Wiler Zeitung - - 2050 Südostschweiz 2846 2142 - Willisauer Bote 1144 1216 - Weltwoche 1167 1036 1343 Zentralschweiz am Sonntag 471 287 - Wochenzeitung 475 431 194 Zofinger Tagblatt 885 4494 1371 Femina - - 10 Zuger Presse - - 158 Finanz und Wirtschaft 87 102 577 Zuger Zeitung - 5250 2453 Freiburger Nachrichten 1443 3261 - Zugerbieter - - 98 Furttaler 147 139 - Zürcher Oberländer 2107 4497 - Glattaler 312 158 - Zürcher Unterländer 1932 4034 3795 Glückspost - 15 19 Zürichsee -Zeitung 2731 4267 3893 Grenchner Tagblatt - - 2409 Encore - - 1 Handelszeitung 428 326 992 Landbote - - 2288 Infosperber 148 160 - SRF.ch 4731 6362 2139 Limmattaler Zeitung 1015 3308 2690 Watson 28 3245 - Luzerner Zeitung - 7151 5209 Zentralplus 888 1492 - Medienwoche 40 39 - Total 110 993 184 157 116 121 Migros -Magazin 68 44 - NZZ Folio - - 43 NZZ am Sonntag 887 872 449 Neue Luzerner Zeitung 5259 - - Neue Zürcher Zeitung 4177 3226 4264 Tages -Anzeiger 10 904 11 292 4004 Nidwaldner Zeitung - 5141 2290 Obersee Nachrichten 144 42 - Obwaldner Zeitung - 5242 2270 Oltner Tagblatt 781 1913 2403 Ostschweiz am Sonntag 457 309 - Rümlanger 63 120 - 22 Table A. 2: Random examples of initial texts for selected meta -data combination s party=Mitte, gender=Male, age=45-55, incumbency=yes, candidate=national council, issue=politics, election=2019 Martin Landolt sitzt im Frühstücksraum des Hotels Stadthaus in Burgdorf. Das Wetter draussen ist garstig, passend zur Lage der BDP. «Hadern Sie mit der Rolle als Kleinpartei, die nicht zwingend gebraucht wird?» party=SVP, gender=male, age=<45, incumbency=no, candidate=national council, issue=politics, election=2019 Über die Hälfte aller Gemeinden hätten Schwierigkeiten, ihre politischen Ämter überhaupt noch zu besetzen. Woran liegt das? Mögliche Lösungsansätze, um eine Trendwende herbeizuführen wurden in der Podiumsdiskussion mit Maja Riniker, Nationalratskandidatin FDP, Renate Gautschy, Präsidentin der Aargauer GemeindeammännerVereinigung, Clemens Hochreuter, Nationalratskandidat SVP, und Markus Freitag erörtert. In diversen Kantonen ist der Amtszwang eingeführt worden. Alle an der Diskussion Beteiligten waren sich einig, dass diese extreme Form nicht die gewünschte Qualität der Arbeit liefert. party=SP, gender=male, age=>55, incumbency=no, candidate=national council, issue=politics, election=2015 Weshalb sollen die Zuger ausgerechnet Sie wählen. Ich will den Zugerinnen und Zugern, welchen eine soziale Gesellschaft, eine intakte Natur und Heimat sowie der schonende Umgang mit Ressourcen wichtig sind, eine Stimme in Bern verleihen. Hubert Schuler ist 58 Jahre alt, ist verheiratet und Vater von zwei Kindern. Der Sozialarbeiter und Leiter des Sozialdienstes Baar wohnt mit seiner Familie in Hünenberg. Politisch ist Schuler Mitglied der SP und kantonaler Parteipräsident. party=GPS, gender=female, age=<45, incumbency=yes, council=national council, issue=not politics, election=2023 Dem ersten Profil idealtypisch zuzurechnen ist Parteipräsident Balthasar Glättli. Er hat sich aber im Interview mit der «Sonntags-Zeitung» bereits selbst aus dem Rennen genommen. Fraktionschefin Aline Trede sagt, sie sei für den ganzen Auswahlprozess im Lead, darum habe eine eigene Kandidatur für sie keine Priorität. «Aber ich überlege es mir sicher nochmals ganz gut.» Bastien Girod will in den nächsten Tagen in Familie und Partei sondieren, ob er sich zur Verfügung stellt. party=FDP, gender=female, age=>55, incumbency=no, cadidate=national council, issue=not politics, election=2019 Diese gilt es zu reduzieren und zu beseitigen und am Thema dranzubleiben. Es ist ein grosses Handlungsfeld offen, und jeder einzelne ist gefordert. Rosy Schmid (Kantonsrätin FDP, Hildisrieden LU): «Boden mit dem Richtigen füttern». Zur regenerativen Landwirtschaft. Das Wort «regenerative Landwirtschaft» ist längst zu einem Schlagwort geworden und steht für einen Aufbruch. 23 Table A. 3: Random examples of generated and tokenized texts for selected meta -data combination s party=SP, gender=female, age=>55, incumbency=no, council=national council, issue=politics, election=2019 streng neu ausserordentlich aufgeblockt erledigen sehen erhalten einreichen verlangen kautionspflicht auflage arbeit polizist bewaffnung vorbild arbeitgeber gesetz grosse polizist einstellung gewalttat grenze grenzkontrolle einsatzort motion erfordernis androhung party=Mitte, gender=female, age=45-55, incumbency=yes, council=national council, issue=politics, election=2015 vergangen erster führen bringen gewinnen merken abbilden sagen nationalratskandidat woche wahlkampf aufmerksamkeit bezirk kandidat region leute region werthenstein nähe dorf wahlumfrage thema party=SVP, gender=male, age=>55, incumbency=yes, council=national council, issue= not politics, election=2023 ehemalig kantonal besetzen ticken mitprägen abstimmen beteiligen urner mitte frage funktion präsident unterwalden erstellung liste nationalratswahl vorlage stimmrecht kantonalvorstand fall abstimmung ausgangslage party=GPS, gender=female, age=>55, incumbency=yes, council=national council and council of states, issue=politics, election=2019 neu erste st.galler treten gehören stehen präsentieren umzugsquartier energiegesetz kraft müllhaufen gelände firma module gangsaal abend stadtparlament ständeratskandidaten grüne gast anliegen bedürfnis wunsch party=FDP, gender=male, age=45-55, incumbency=yes, council=national council, issue=politics, election 2015 kantonal unschön stärken erklären passieren sagen einreichen knatsch finanzhaushalt anlass finanzdirektor rücken ungerechtigkeit abstimmungskämpfe problem vizepräsident woche postulat regierung