Introduction

Electoral volatility has long been the focus of scholarly attention. The dominant branch of analyses focuses on inter-election vote switches at the aggregate level and among individual voters (Dassonneville & Stiers, 2018; Hobolt et al., 2009), and the influence of electoral campaigns upon vote choice also has a long-established series of findings (Boomgaarden et al., 2016; Geers & Strömbäck, 2019). However, recent work has shown that changes in party preferences can occur in a much shorter term, e.g., within a single election campaign (Jennings & Wlezien, 2016; Johann et al., 2018).

Among the factors that influence short-term vote switches, the content of the media shaping voters’ political information about the parties is influential (Van Spanje & De Vreese, 2014). Recent studies about the effects of media on political behavior focus more closely on the influence of media content on attitudes and behavior (Schuck et al., 2016; Vreese & Semetko, 2004).Footnote 1 In this paper, we focus on the influence of the media on vote choice during short-term election campaigns.

Previous contributions to this line of research focused solely on established Western democracies and typically upon a single election campaign. However, the influence of media is likely to be sensitive to the changing electoral and media contexts. We employ original panel data from four general elections in Turkey between 2002 and 2015, during which time Turkey has moved from a highly fractionalized competitive party system to a hegemonic one dominated by the Justice and Development Party (Adalet ve Kalkınma Partisi, AKP) (Çınar, 2019; Laebens & Öztürk, 2021). The emerging party system is characterized by less competition and democratic backsliding reflected in the media system.

Turkey represents a case of an increasingly illiberal system with significant political polarization directly reflected in the media’s coverage of election campaigns. While internal pluralism across different newspaper readership communities is declining, external pluralism is rising. As a result, coverage of particular political parties in individual media outlets is increasingly biased (Çarkoğlu & Yavuz, 2010), and such coverage becomes increasingly polarized as election day approaches (Çarkoğlu et al., 2014). With rising press-party parallelism and increasingly polarized campaign coverage, different constituencies are likely to be isolated and closed to a constructive debate, making short-term volatility in vote choice unlikely due to media coverage. As all sides appear to retreat to their corners, respective media outlets only cover the campaign from a partisan perspective.

The recent democratic backsliding and decay in democratic freedoms may turn the media into a mere echo chamber without any tangible effect on political preferences by primarily offering partisan content and strategically manipulating the visibility of their favorite parties and candidates in their coverage. Therefore, we expect Turkey to represent a least-likely case where polarized partisan coverage of election campaigns is least expected to influence vote choice. In a hegemonic party system with a dominant political party, the media may also have more significant influence as it is the principal channel through which the party minimizes repression/authoritarian electoral manipulation and maximizes its vote. However, given increasingly high levels of political polarization in the country,Footnote 2 partisan groups are expected to gravitate towards co-partisan media groups. Thus, the mobilizing effect of media should be observed only among weak partisans convinced by their own media group to continue supporting their party. We expect a lack of media effect to be more relevant later in the AKP’s tenure as the Turkish media system polarized and voters retracted to their partisan groups. However, despite our expectations to the contrary, we provide evidence of significant media effects for four elections from 2002 until 2015. We show that the media still substantially influences intra-election vote choice. The media effects are heterogeneous across voters who change their choices in the short term and those who remain loyal.

By linking the media content to representative samples of voters in four consecutive elections, we provide evidence for the media influence and show that media still affects party choice even in relatively short periods during electoral campaigns. Since we focus on party choice during campaign periods, this effect is at the margins of electoral competition, and the bulk of party constituencies remain mostly irresponsive to short-term campaign effects over the period we analyze.Footnote 3 Despite the consolidation of political camps and cliques and the rising ideological bias in media content, those exposed to alternative media channels are still influenced to change their party choice. Moreover, this media effect is not uniform across parties and media groups. Hence, the competition to shape the media content and the government’s tutelage is not in vain since media remains a potent political factor even under the most unlikely circumstances.

Below, we first lay down the central tenets of our conceptual framework to formulate our main argument, followed by depicting the increasingly polarized scene of media and politics in Turkey.

Media Effect and Vote Choice

There is well-established literature on the relationship between media content and partisan choice, especially in advanced democratic contexts, with mixed findings. However, the overall picture suggests that the information environment substantially affects various issues, such as political evaluations and vote choice. Yanovitzky and Cappella (2001) link media content with a panel survey to show that political talk radio did not play a substantive role in voters' attitudes during the 1996 presidential elections in the United States. While radio may be ineffective in changing political attitudes, the partisan information environment has a substantive effect on vote choice through television news programs (Vreese & Semetko, 2004). Moreover, the attention paid to a political leader and how a leader is portrayed in the media can influence these leaders' effectiveness and legitimacy perceptions. However, voters who are sure about their political preferences are less likely to be affected by opposing information (Bos et al., 2011).

By linking the media environment to two different panel studies, Matthes (2012) shows that cross-cutting information can delay vote choice decisions among undecided voters. To our knowledge, Matthes (2012) is the only study to present more than one empirical study to support the argument by matching two-panel datasets from Switzerland about two referenda in 2008 on various issues. Other scholars have also employed methodologically innovative ways to understand the relationship between media and vote choice. For instance, Schuck and Vreese (2008) combine panel data with content analysis and support this study with experimental findings to show that a positive slant in media coverage can backfire among opposition supporters during a referendum campaign. Although an alternative causal direction is viable—i.e., voters become disillusioned and do not turn out to vote after receiving information because of the perception that they will be on the losing side of the referendum—the experimental evidence supports a mobilization effect for opposing information. Schuck et al. (2016) extend linkage study methods to a cross-national setting to present further evidence that the media can mobilize voters during an election campaign. By conducting a linkage study in a cross-country environment, they model cross-level interactions (individual and country-level effects) to show that more favorable EU polity evaluations in a country increase the magnitude of conflict framing effect for mobilization during the 2009 European Parliament elections.Footnote 4

We continue with the tradition of these studies, but unlike previous ones that exclusively focus on advanced Western democracies, we present evidence from a case that experiences democratic backsliding and a decline in media freedoms. Additionally, we present empirical findings from four consecutive campaign periods in Turkey from 2002 to 2015. Given these multiple studies during which we can trace the media system’s democratic backsliding trajectory, we aim to account for the contextual changes over time.

In line with the previous studies, our primary expectation is that media slant will positively affect vote choice. The more favorable coverage of a political party, the higher the likelihood is for the reader to support that party. In other words, favorable media content should push voters exposed to such content toward the given party during the campaign period when political attention and awareness among voters are elevated. The campaign period is when voters re-evaluate their political support, obtain new political information and cues, and update their political preferences. Therefore, we expect the content of the newspapers during this critical period to be influential in reshaping voters' political preferences during the short-term campaign period. However, in addition to testing the previous studies in an alternative setting, we expect the media environment and the institutional context of party competition to moderate the media effect.

The media effect should not be uniform across years and different partisan groups. High partisan polarization is associated with further parallelism between media and party choice (Stroud, 2010). In such settings, independent voters occupy a smaller percentage of the total electorate. Therefore, while higher favorability exposure increases the likelihood of voting among such independents, the media becomes an echo chamber for partisans who reinforce their current political affiliations. For voters who become less open to alternative sources of information and rely on a few sources of slanted partisan information provided by mouthpiece media of their party, media ceases to change the likelihood of party preference. Therefore, we expect the media’s effect on the likelihood of changing party support to be diluted by political polarization in a biased media environment. The media effect should be primarily relevant for undecided voters or independents who are likely to switch their party choice even if they previously supported another party.

In addition to polarization, the level of democracy in a country also affects whether the media serves its functions, i.e., holding politicians accountable by providing relevant information to voters. The relationship between backsliding and media suggests an interesting tension. Leaders who want to increase their powers try to silence opposition media; while the media becomes controlled by the government and polarized, the least it should affect voter behavior. For instance, Latin American presidents are likely to silence their critics in the media if the legislature and judiciary are weak against the president's executive powers (Kellam & Stein, 2016). A similar phenomenon can be possible in countries that experience democratic backsliding with the rise of authoritarian/populist leaders who consolidate their power through executive aggrandizement (Bermeo, 2016). Democratic backsliding is not a full-fledged shift into an authoritarian regime. Competitive elections may still occur, but liberal tenets of democracy increasingly become less viable in contexts where the opposition’s potential to oppose elected officials is limited, and individual rights and freedoms are increasingly more vulnerable to intrusions (Waldner & Lust, 2018). Thus, we expect mainstream media to lose its function of informing voters about politicians even if specific venues for opposition may still exist in such environments.

A democratic deficit can be a potential arbiter for the media effect, and the influence of democratic backsliding can be more nuanced. For example, there can be de facto prohibitions against certain opposition parties in the media, while mainstream opposition can still find space to be covered with a mixed slant (Yıldırım et al., 2021). However, a positive media effect is not viable when a political party is not allocated enough visibility or categorically covered negatively. Moreover, the media effect is only possible if a candidate or party is visible in the media. If the visibility of parties is impeded by effectively keeping them out of media coverage, then no media influence is possible. The worst-case scenario occurs when only negative coverage appears in the media for some political parties. With such strategically manipulated coverage under the control of a hegemonic party, democratic backsliding ensures that any media effect is only negative against the party in question. Therefore, backsliding and de facto control of media by the government can be selective in their impact.

Media influence under democratic backsliding and parties’ mobilization efforts provide alternative explanations for a short-term change in party support. Partisan information can increase the accessibility of the party platform or policy proposals and increase the party's popularity among the larger public, while mobilization efforts can frame relevant policy issues among potential voters (Chong & Druckman, 2007). Given these expectations, Turkey presents a crucial case where the media, parties, and voters are highly polarized. As a result, the media system exhibits a high degree of parallelism within democratic backsliding. Simultaneously, party organizations widely use mobilization efforts such as canvassing and clientelistic inducements (Aytaç & Kemahlıoğlu, 2021).

Democratic Backsliding and the Media in Turkey

In the aftermath of the Arab Spring, Turkish democracy was seen by many as a “role model” for the Muslim-majority countries (Tepe, 2005). AKP, as a new generation of pro-Islamists, came to power in 2002, reinvigorated democratic reforms, started membership negotiations with the European Union (EU), and took initiatives to resolve longstanding conflict with Kurdish ethnicity. However, this optimism was short-lived as AKP’s liberal democratization reforms ended (Özbudun, 2014). Restrictions on freedom of the press, television content, and internet use started to rise (Lavigne, 2019).

In June 2013, Turkey was hit by protests that came to be known as the Gezi Park protests (Yörük, 2014). A breakup within the conservative coalition followed the emerging authoritarian tendencies. In addition, prosecutors put forward graft allegations implicating prominent cabinet members. However, despite the graft allegations or the Gezi Park protests, the AKP’s founding leader Erdoğan assertively became the first popularly elected president in August 2014.

President Erdoğan remained active during the June 2015 Parliamentary election campaign to support the ruling AKP (Kemahlıoğlu, 2015). Nevertheless, due to a significant loss of support in the June elections, the AKP could not form a single-party government. The social-democratic Republican People’s Party (Cumhuriyet Halk Partisi, CHP), the Nationalist Action Party (Milliyetçi Hareket Partisi, MHP), and the pro-Kurdish left-wing Peoples’ Democratic Party (Halkların Demokratik Partisi, HDP) all exceeded the 10% threshold. As a result, parties failed to form a coalition, while the peace process to resolve the Kurdish conflict collapsed with this electoral setback for AKP, rising terror attacks, and a militarized response. President Erdoğan used this opportunity to call for a repeat (snap) election. As the country moved towards a second election, the political agenda changed, benefitting the AKP due to heightened security concerns (Aytaç & Çarkoğlu, 2019).Footnote 5

From surprising electoral victories and democratization reforms to establishing a hegemonic party with authoritarian tendencies, the history of the AKP since 2002 is also marked by rising polarization among voters. In his periodization of “pernicious polarization,” Somer (2019) argues that incremental democratic erosion started as early as 2008, accompanied by a rise in polarization. This polarization is perhaps most deeply reflected in how the AKP government treats media and journalists with increasingly repressive measures and censorship (İlkiz, 2019; Yeşil, 2018). According to the 2015 Freedom of the Press Index, Turkey ranked 142nd with “not free” status among 199 countries and 149th among 180 countries in the 2015 Press Freedom Index.Footnote 6

Recent research on Turkish media also shows that government influence over the media sector is increasingly intense on top of biased coverage across media outlets. As a result, ownership and control over the media sector in Turkey are becoming more concentrated (Akser, 2018). Since the 1990s, several structural changes have reshaped the media sector and contributed to declining press freedoms. One such factor was the liberalization of media ownership, which began in the 1980s and continued in the 1990s. These businesses then started pursuing clientelistic relationships with the state due to their investments in other sectors and public procurement deals (Christensen, 2007; Corke et al., 2014).

While press-party parallelism is a historical phenomenon in Turkey and the readership structure of the newspapers exhibits partisan alignments, parallelism is increasingly more concentrated around fewer partisan groups for each outlet (Yıldırım et al., 2021). Newspapers bring together partisans of a fewer number of different parties in their readership, and the coverage becomes increasingly polarized as election campaigns progress, pushing the newspaper coverage of parties farther away from one another as the election approaches (Çarkoğlu & Yavuz, 2010; Çarkoğlu et al., 2014).

In short, as democratic backsliding progressed, the Turkish media sector found itself under pressure. As a result, the media sector’s coverage of politics was increasingly polarized and biased. The readership of different media outlets was highly homogenous in partisan terms. Different media groups became increasingly dependent on state resources for their ad revenues, which rendered media conglomerates' financial viability and fortune more dependent on public procurement bids. Hence media outlets increasingly became biased in their coverage of politics. The readership of these outlets became increasingly homogenized based on their partisan affiliations as pro or anti-government camps. In these circumstances, we expect the media effect to be low due to the increasing slant that reinforces existing partisan lines among different reader groups. While specific political forces capture most media, independent venues either lose financial incentives or remain marginal. Combined with increasingly homogeneous party preferences of different reader groups, we expect media influence over vote decisions to be limited. However, as we show in our empirical models, the reshaping of media as an institution during the early period of democratic backsliding in Turkey was not in vain, as it still had a considerable impact on undecided voters and switchers. In other words, while democratic backsliding and decreasing levels of media freedom should attenuate the effect of partisan information on vote choice due to echo chambers and highly polarized partisan groups, there was still a substantive effect on vote choice.

Data and Method

Audience selectivity is a crucial aspect of political communication. Different partisan groups can self-select into specific media groups and receive information congruent with their political preconceptions creating a “reinforcing spiral” (Slater, 2007). This aspect of media selection poses methodological challenges, given that individual media choice and political characteristics are outcomes of the same long-term individual characteristics.

One way to deal with self-selection is to analyze individual preferences to strip the direct and indirect effects of the media influence in subsequent periods. In a recent review, De Vreese et al. (2017) mention several opportunities and limitations of linkage studies that match individual survey data with media content. For example, directly asking voters about self-evaluations of such a media effect may inflate or deflate the findings. Instead, one could tap into individual media use and then match this use with scored content over multiple periods. By doing so, we can untangle the reinforcing spiral.

To account for the reinforcing spirals model and its methodological implications, we employ multiple panel designs implemented in the Turkish Election Studies (TES) for 2002, 2007, 2011, and 2015. This design helps us strip the effect of newspapers from the pre-election campaign period on post-election reported vote choice. We collected three crucial pieces of information in all panel surveys: 1-pre-election reported vote intention, 2-post-election reported vote choice, and 3-respondent’s choice of newspaper in the pre-election. In addition to these questions, commonly used variables about ideology and demography to explain vote choice were available.

We randomly selected respondents from a clustered sample based on geographic regions. Up to three attempts were made to conduct surveys with non-respondents, and in the case of failure, we did not substitute these non-respondents. 2002, 2007, and 2011 TES surveys were two-wave panel studies conducted before and after the election. 2015 TES was a three-wave panel conducted before and after the 2015 June elections and after November 2015 snap elections. For 2015, we focus on the June elections and use the pre and post-June panels, thus, controlling the changing dynamics of snap elections.

In addition to these voter surveys, the fifteen most broadly read national newspapers were selected, and their content was analyzed. Table A.2 in the Supplementary Information provides the number of stories coded per newspaper. We coded 72,047 stories, and 36,167 of them (50.2%) had political content. We matched the TES survey respondents’ newspaper choice in the pre-election survey with favorability from this newspaper during the campaign period (90 days before the elections). For 2002 and 2007, our newspapers covered 65% to 78% of the nationwide circulation. In 2011, we added two newly established newspapers (Habertürk and Sözcü), and for 2011 and 2015, we captured 78% and 74% of nationwide circulation with these 15 newspapers, and they cover the whole ideological spectrum in the country (pro-government, mainstream, and opposition papers).

We have 12,848 respondents in our samples across the four surveys. However, our effective sample decreases to 2101 for several reasons. First, not all respondents participated in the panel study. In our post-election rounds, we reached about 50% of the respondents in the pre-election survey. Secondly, even if respondents participated in the panel, some had missing responses to pre-election vote intention, post-election vote choice, or newspaper readership questions.Footnote 7

Balance tests indicate that newspaper readers differ from the rest of the survey respondents. This difference is not surprising, as readers should be more politically involved. However, attrition due to non-response can produce selection bias in results. We find evidence for some limitations to the external validity of our findings in several respects, while in others, we find no effect due to attrition. Our balance tests on panel response (Table A.5) indicate no significant difference between the post-election non-respondents (i.e., attrition sample) and the panel sample for relevant variables such as gender, age, ideology, education, and income. However, the panel sample was less likely to be religious or from urban areas and more likely to be Kurdish. Therefore, we include these three and other relevant variables as controls in robustness checks replicating our main models. Additionally, even if self-selection is inherent to the observational nature of this study after considering the reinforcing spirals model, potential limitations are tested in the supplementary information, which shows that different partisan groups or undecided voters and switchers are not substantively more likely to select non-partisan, mainstream newspapers as opposed to partisan papers.

Our dependent variable is based on the pre (post)-election survey reported vote intention (choice). We can capture switchers who changed their preferences by asking the same question twice, before and after the election. We focus our study on politically relevant parties and construct a set of five choices for each respondent: AKP, CHP, MHP, Kurdish Parties (which changed names across the years), and the rest (including fringe parties and undecided voters). We also focused on respondents who switched their vote choice during the campaign period to analyze the change in vote choice. 720 out of the 2101 respondents (34.2%) were either undecided in the pre-election or switched to a different party post-election.

All four surveys asked respondents’ newspaper choices to link the media effect with individual voters.Footnote 8 In addition to this information, we analyzed the political news stories on the first page and a randomly selected second page with political news from the selected newspapers.

For each story, we first coded the story's tone towards the four politically relevant parties on an ordinal scale ranging from − 3 to + 3. Then, we calculated the average tone of all the political stories. We then weighted this average by the space occupied by the story, discounting smaller news stories that grab less attention.

For each newspaper in each campaign period, we calculated average favorability scores for the four politically relevant parties and another average for a combination of all fringe parties. This favorability could range from − 3 to + 3, the observed minimum is − 1.04, and the maximum is + 0.82.Footnote 9 Respondents who reported reading one of the fifteen newspapers were assigned these favorability scores. While we argue for increasing levels of polarization in the media, the gap between the observed range and potential measurement boundaries may imply the opposite. However, this gap is primarily due to numerous news stories that are either neutral to political parties or do not mention any political party explicitly. In such cases, we assigned the news story a neutral score of 0; therefore, after aggregating the average media tone for each newspaper, the observed range is deflated due to our aggregation and the number of neutral news stories that do not mention political parties. Nevertheless, even in this case, the observed range from lowest to highest value increases more than threefold.Footnote 10

In addition to the media tone, we also measured the visibility of political parties by coding three different components for each political story: 1-politician photos or party logos, 2-direct quotes from politicians, party leaders, and spokespersons, 3-whether the story refers to a party (even if there is no direct quote). We created an additive index of visibility based on these three components that measure visibility for each of the four relevant parties and the fifth category (i.e., fringe parties). Similar to favorability, we weigh visibility by the physical space given to the story. Based on this measurement, we calculated visibility as a percentage of all political stories in the newspaper, where 1 would indicate that every political story mentions the party, quotes the party, and has a relevant photograph. At the same time, 0 means that the newspaper did not allocate any visibility. Observed visibility scores range from 0.0016 for the combination of fringe parties to 0.55 for AKP. Similar to favorability, we matched this score with our respondents’ reported newspaper choice. Further details about visibility and favorability and their change across the years are in the supplementary information, and these details also attest to the AKP’s dominant position in the media.

We also collected similar data for TV news stories from 12 prime-time news programs for one month during the 2011 election campaign. However, data on TV is limited to making generalizable claims regarding TV consumption since we only have results for the 2011 election. Therefore we employed the TV effect as a robustness check in the supplementary information to show that even after controlling for the impact of the TV slant, we still see a stable newspaper effect.

In addition to the media effect, short-term campaign period change in party choice can occur due to parties’ mobilization efforts. Various communication strategies followed by party organizations play an influential role and have substantive mobilizing effects (Green & Gerber, 2019). Mobilization can influence the decision to turn out and the party choice by mainly persuading independent or undecided voters. Both impersonal and personal mobilization, such as rallies and party canvassing efforts, play an essential role in changing preferences during campaigns (Johann et al., 2018).

Some voters may change their vote choice because of the parties’ mobilization efforts. Therefore, we create an index of mobilization efforts based on three questions: 1-canvassing visits, 2-party rally/meeting attendance, and 3-reading a party brochure. Based on these three questions, we create four additive indices for each relevant party. 47% of our total effective sample (988 respondents) reports at least one or more mobilization efforts by a relevant political party.

For our analysis, we can approach the panel data in several ways. First, we can limit our sample to those who changed their party choice from pre to post-election or include the lagged dependent variable (party choice). Since the main independent variable (favorability) is measured only in the pre-election wave, we use a lagged dependent variable to model the party choice. We include pre-election vote intention as our primary control variable as it serves two purposes: first, it accounts for individuals whose vote choice remained the same across the pre and post-panel design, helping us delineate short-term effects better. Second, it reflects that individuals are not equally likely to switch across parties. Hence, party choice should follow a pattern based on individuals’ alternative propensities, and pre-election vote intention accounts for such differences. For instance, a person who reports the intention to vote for the nationalist MHP is not equally likely to switch to the conservative AKP and leftist parties. As an additional robustness check, we also conducted a sub-sample analysis only with respondents who reported switching. This additional model focuses only on respondents who reported switching to explain the differences between undecided voters and voters who switched from one party to another. As such, we focus on party support from pre to post-election rounds to address potential endogeneity problems. To evaluate these differences, we stack our data across alternative party choices and create voter-party dyads. Therefore, for each respondent, six observations determine their choice across four politically relevant parties plus other parties and another category for undecided voters. While using this stacked data, we cluster standard errors across individuals.

Findings

To observe a media effect on vote choice, we must first have some voters change their choice during the campaign periods. Table 1 shows the composition of switchers to and from the two largest parties and the average media favorability across these different switcher groups. Not surprisingly, the most populated group of switchers was in 2002, a watershed election that led to the AKP’s first electoral victory. In this critical election, two parties benefitted from a net positive switch during the campaign. The AKP and CHP gained 9.7% and 3.1% of all voters in the brief period between pre-and post-survey. In other words, slightly more than one-third of AKP voters in the 2002 election selected the party during the campaign period just before the election. Respondents that remained supportive in both rounds of the panel surveys also appear to be lowest in 2002 but steadily rose in consecutive elections. Thus, the vote decisions were more fluid in 2002 than in the later elections. We also note that those who declare voting for either AKP or CHP in the pre-election round are less likely to switch to another party in the post-election panel.

Table 1 Vote Switches and Average Newspaper Favorability across party supporters in four election campaigns
Full size table

The AKP benefitted from new supporters after 2002, whereas this is not the case for CHP. In 2007 and 2011, respondents reporting a switch from CHP, and to CHP in the post-election period were similar. In the AKP’s case, the party had a net positive benefit (row 3–row 2) of 9%, 6%, and 5.3% in 2007, 2011, and 2015. Across the four elections we studied, 40.8% of the undecided respondents switched to either AKP or CHP, whereas 16.2% switched from these two parties to another party in the post-election survey.

As a first stab at distinguishing the media effect on switchers, Table 1 also reports the average newspaper favorability towards the two parties. On average, those who do not support either of these two parties are exposed to the most negative media slant (row 1), while those who consistently support the party during and after the elections (row 4) read more positive newspapers.

Although the descriptive evidence from Table 1 supports our main hypothesis that more favorable media campaign coverage increases a voter’s likelihood to support a party even in a context such as Turkey, a multivariate model—where we account for alternative explanations in a more nuanced way—is necessary to pin down the media effect and the role of mobilization. We hence estimate logistic regression models.

Table 2 reports the results. Our baseline model (Model 1) supports our central expectation that more favorable coverage increases voters' likelihood of supporting a party. In addition, our primary control variable (pre-election vote intention) is in the expected direction with a substantive effect. This finding attests to preference stability in the short-term campaign period. Including additional controls in further models (Model 2) changes the magnitude of vote choice stability. More importantly, across our models, the effect of favorability remains robust, and a unit increase in media favorability increases a respondent’s likelihood to support a party in the election by more than 3.3 times (Model 2).

Table 2 Estimating vote choice (stacked dataset)
Full size table

We include several controls in our baseline model to elucidate the extent of this effect. First, we include campaign period, undecided voters, and party choice variables as additional controls in Model 2. As expected, undecided voters were more likely to switch to a party post-election, and this effect is robust across all specifications. In addition, undecided voters in the pre-election period are more likely to report a different party choice in the post-election period.

Moreover, since we are using a stacked dataset with five alternatives, we control for the stacked units (i.e., party choices), which show that compared to the reference category of AKP, voters are less likely to report other parties in the post-election periods. This attests to the electoral dominance of AKP. Finally, we introduce party mobilization efforts in Model 2. With each additional experience of the parties’ mobilizational efforts, we estimate that voters are becoming around 90% more likely to support the party. As expected, this effect is statistically significant and robust to alternative specifications.

Model 2 introduces media visibility, showing that higher levels of visibility increase a respondent’s likelihood to vote for a party, but this effect is not significant when we control for demographic differences. However, favorability and visibility can reinforce each other in media influence. For instance, a newspaper can provide a political party with exceptionally favorable coverage yet cover the party only in a couple of stories. In such a case, the media will not influence voters. Therefore, we expect a product of the two to be especially relevant. Model 4 shows that the interaction of favorability and visibility (Fav × Vis) is statistically and substantively a significant predictor of vote choice in the post-election wave, and this effect has more magnitude than the pre-election vote choice.

These two additional variables can also help us understand whether favorability or visibility drives the media's influence. We check for this alternative in Table 3 with Model 3, suggesting that higher levels of favorability (not visibility nor an interaction of the two terms) influence the media effect. Compared to previous models, Model 3 implies that considering the effect of visibility and the interaction of the two terms dampens the effect of favorability, but the media influence is still relevant through favorability. Moreover, visibility and its interaction with favorability are in the expected direction. Null hypothesis tests for the two coefficients are not statistically significant at conventional levels (p-value for visibility is 0.162, and the interaction term fav × vis is 0.11). However, because partisans and switchers are included in the model, this interaction term is not substantively important, as partisans are not expected to be influenced by media favorability or visibility.

Table 3 Full model specification (Model 3) with a comparative subsample (Model 4)
Full size table

To better illustrate the interaction effect of favorability and visibility, Fig. 1 plots the predicted likelihood to support a given political party as a function of the two terms based on Model 3. As expected, the likelihood of supporting the party increases with increasing favorability levels. For example, when voters’ preferred newspaper attacks a party and presents it in a negative slant, the predicted likelihood to support the party is estimated to be around 12%, regardless of visibility. With increasing levels of favorability, the likelihood to support increases, and at the highest positive level of favorability, we predict the likelihood to support to be around 31% when the party is very visible in the newspaper (90 percentile visibility). Our findings show that more favorable coverage of the political party in the preferred media environment increases voters' likelihood of voting for that party.

Fig. 1
figure 1

Predicted likelihood to support for a political party across levels of favorability and visibility

Full size image

More importantly, to test these effects and further elucidate the combined role of visibility and favorability, we focus on a subsample analysis of voters who reported switching from pre-election choices to a different party choice in post-election surveys. By definition, the pre-vote choice differs from the post-election choice for this sub-sample. Therefore, we replicate Model 3 with this sub-sample leaving out the pre-election vote choice control. This alternative specification (Model 4) correctly classifies 80.2% of switchers into the right political party. Furthermore, results show that favorability interacts with visibility and increases the likelihood of switching to a political party only when the party is visible and covered positively. To substantively interpret the results presented in Model 4, Fig. 2 plots the predicted likelihoods among the switchers sub-sample.Footnote 11

Fig. 2
figure 2

Predicted likelihood to support a political party across levels of favorability and visibility (only among switchers) and 2002 effects

Full size image

Figure 2 shows that favorable and visible messages are both necessary for media influence. When we considered the whole sample in Model 3, favorable parties were more likely to garner support regardless of visibility. However, there is a stark difference among switchers, and visibility plays an important role. When a party does not receive enough coverage (10 percentile visibility), the predicted likelihood to switch decreases from 22 to 18% across the range of favorability. However, for a highly visible party (90 percentile), the same change in favorability increases the predicted likelihood of switching from 12 to 28%. In other words, favorable coverage is necessary but not sufficient among switchers. A party must be covered repeatedly and be highly visible in the media to garner further support from voters who switch.

Furthermore, we also present a second panel where we estimate the marginal effects of the 2002 elections, which were more competitive and held before the media's domination by the AKP. The 2002 election was also the first election when AKP came to power to start its long tenure. Therefore, our results may have been explicitly driven by these elections. However, 2002 effects for switchers show that even for switchers exposed to high levels of visibility, 2002 results did not have higher media effects among switchers compared to succeeding elections as the second panel of Fig. 2 shows.

Moreover, Fig. 3 presents predicted probabilities from an additional model where we introduce a three-way interaction between favorability, visibility, and a dummy variable for switchers. Relevant independent variables from this three-way interaction are given in Table 4, Model 5.Footnote 12 This alternative specification also supports the finding that favorability and visibility have heterogeneous effects on party support among switchers vs. loyalists. Compared to party loyalists, switchers are influenced by a combination of favorability and visibility, and the three-way interactions are significant at conventional levels.Footnote 13 This interaction shows that the undecided voters and voters with weak party affiliation—who eventually switch their party choice—require favorable coverage that is also visible.

Fig. 3
figure 3

Predicted likelihood for switchers and supporters

Full size image
Table 4 Three-way interaction between favorability, visibility for switchers, and undecided voters
Full size table

Additionally, until now, the findings only control undecided voters and show that being undecided increases the likelihood of switching. However, by definition, undecided voters are switchers, and we assumed continuity for undecided voters if they reported not voting in the post-election survey. To open up the potential effects among these undecided voters and weakly partisan voters who switched from one party to another one, we also conducted an additional analysis in Model 6 where we estimated an additional three-way interaction between the media effect variables and being undecided (as opposed to a partisan switcher) within the subgroup of switchers. Results indicate that undecided voters are influenced by favorability (regardless of the visibility levels), whereas partisan switchers are influenced by the visibility of the party they decide to switch to in the newspaper they read.

Conclusion

This study aimed to contribute to the literature on media effects in three distinct ways. First, we offer a representative content analysis of newspapers for four general election campaign periods and present our results from multiple studies. Secondly, we adopt a panel design to eliminate potential endogeneity threats in media effects research. Lastly, we aimed to present media effects for a non-Western case where we expect the effect to be muted due to democratic backsliding and lack of pluralism in the media.

We focused on Turkey, which provides a robust test for our argument because of the country's highly polarized political and media system, where we do not expect to observe strong media influence. Turkish newspapers become increasingly polarized during campaign periods, consolidating positive and negative slants to selected parties. Nevertheless, even in such conditions, we show that the newspaper pitch makes a difference and increases the readers’ likelihood to support a party.

When we consider all voters together, favorability in their choice of newspaper increases their likelihood to support a political party. We show that favorability interacts with visibility to increase the likelihood of switching. Only when a party is visible and depicted positively do we observe an increase in the likelihood of switching to the party. However, favorability and visibility work differently for partisans and undecided voters who eventually support a party and partisan switcher. This does not challenge our main argument, as motivated reasoning may explain why partisan voters are influenced only by favorable coverage of their party. These voters may be selectively processing information regarding their party and its competitors, and regardless of its visibility in the media environment, they may be influenced by favorable coverage only for their party.

Moreover, given the lack of media pluralism in the Turkish context, these findings imply grave consequences for media polarization, its dominance by a single hegemonic party, and, eventually, the perpetuating electoral effects of media in the context of democratic backsliding. As the media slant consolidates towards a single dominant party, undecided voters are no longer exposed to alternative information. As a result, their likelihood of switching to the opposition decreases due to negative coverage, or the media does not influence it due to lack of visibility. Media polarization and (in)visibility may have dire consequences for a level playing field necessary for democratic party competition. Empirically, decreasing levels of variation in visibility among opposition parties imply that there are categorical barriers against an observable media effect for opposition parties.

As part of a global trend, newspaper readership figures have also decreased in Turkey. Over the nine years between 2011 and 2020, physical newspaper circulation decreased by more than 50%, and instead, online newspapers became a popular alternative. Additional results in the supplementary information show systemic differences between newspaper readers and the rest of the population. Certain demographic groups, such as women, rural voters, and less educated and poorer voters, read newspapers less. Therefore, changing dynamics of readership also present important conditions for the theorized effect in this paper.

Furthermore, according to the most recent figures, by 2020, only 43.4% of Turkey's voting-age population read newspapers regularly. 62.9% were following their newspapers from online sources.Footnote 14 Additionally, social media may play an essential role in moderating the effect of conventional media. This study focused on conventional media as a determinant of vote shifts. The supplementary information also provides additional robustness checks for the effect of TV news from 2011.Footnote 15 Results indicate that TV news does not change the substantive interpretation of our results. However, further research is necessary to elucidate how echo chambers and hostile political discussion in social media may attenuate or reinforce the effect we showed in this study.

Additionally, our paper contributes to the more extensive literature on the relationship between media and vote choice by showing that even in an increasingly polarized political environment where we expect voters to remain within different partisan camps, partisan information in the media can still have an effect. In this respect, our results corroborate previous findings showing the continuing role of media effect in a context where this effect should be minimal. Scholars have not extensively used media effects to explain political behavior due to self-selection problems. However, our study shows that media has a substantive effect, even in the short-term campaign periods. Therefore, models of voting behavior should include the media effect after controlling for endogeneity by proxy variables or panel design. Even in a context where the media effect should be minimal, it still plays a substantive role.