The Media and Democratization: A Long-Term Macro-Level Perspective on the Role of the Press During a Democratic Transition

ABSTRACT The media are assumed to play a key role in democratization. Much of the available evidence on the media’s role in democratic transitions is based on a comparative and global perspective, focusing on rather recent key political events. Although democratization is conceptualized as a process that occurs over a long time, there is limited longitudinal evidence. Focusing on Austria, we used a long-term macro-level perspective ranging from 1816 to 1932, including the transition from authoritarian rule (monarchy) to democracy (republic). Grounded in previous research on the demonstration effect, we investigated whether the press contributed to democratization processes. Content-analytic data on the Vienna-based press were used to assess the salience of the idea of democracy in the press for each year of the observation period: How much did the press report on democratic ideals, such as freedom or equality? The level of democratization was assessed using three available longitudinal measures of democratization. Using autoregressive integrated moving average (ARIMA) modeling to account for autocorrelation and the trend in the time series, we show that there was a covariation between increases in the salience of the idea of democracy in the press and increases in the level of democratization. Furthermore, we found that a higher salience of the idea of democracy in the press in a given year “Granger-caused” (i.e. prospectively predicted) future increases in the level of democratization. Although we acknowledge the limitations in terms of causal interpretations, these findings are consistent with the idea of a long-term macro-level media effect.

Tocqueville elaborated on the relevance of the press for democracy in the mid-nineteenth century (De Toqueville, 2000).
Despite such early writings, most available empirical evidence on democratization processes is related to specific recent political changes in regions such as post-communist Central and Eastern Europe, Latin America, the Middle East, North and Sub-Saharan Africa, and East Asia (Haerpfer et al., 2019;Highley et al., 1992).This claim holds true for most political communication research on the role of the media in this regard (R. Gunther et al., 2000;Voltmer & Rawnsley, 2019; but see Groshek, 2011, andWertz et al., 2021).In fact, much of the available evidence is based on a comparative and global perspective, focusing on rather recent key political events.The possible role of the media has been discussed for these individual cases.With regard to these rather recent key events, such as the fall of the Berlin Wall or the Arab Spring, democratization processes have often been defined in terms of days, weeks, months, or at best, years.Conversely, when thinking about many of today's established democracies, democratic transitions have unfolded over a much longer time period -defined not by months or years, but by decades or even centuries.Despite the evidence from available studies using a comparative and global perspective on rather recent cases, there is clearly limited longitudinal empirical assessment.This had also been emphasized by Jebril et al. (2013), who called for more research with longitudinal designs, such as time series analyses.
The primary contribution of the present paper is our study of the role of the media in democratic change over a very long time period, using a long-term macro-level perspective that relies on a time series approach.In fact, we aimed to investigate whether the press contributed to democratization in the long term.Focusing on Austria, our observation period includes the transition from authoritarian rule (Habsburg Monarchy) to democracy (First Republic), spanning the early nineteenth century to the first half of the twentieth century.Based on theorizing on the demonstration effect (Huntington, 1991;Voltmer & Rawnsley, 2019), which we outline in detail below, we investigate the interplay between the salience of the idea of democracy in the press and democratization.Given that much of the available evidence on the media's role in democratic transitions is based on rather recent key political events, we see our long-term perspective as a supplement to the currently dominant perspective, aiming to enrich our understanding of the role of the media in democratization.

Democratization
Democratization is conceptualized as a transitional process from authoritarian rule to democracy.Democracy -the ideal endpoint -has been defined in various ways (e.g., Dahl, 1971;Lipset, 1959;Popper, 1977).Following Vanhanen (2000), we can view democracy as "a political system in which ideologically and socially different groups are legally entitled to compete for political power, and in which institutional power-holders are elected by the people and are responsible to the people" (p.252).On a most basic level, the strength of a democracy is that bad rulers can be removed without bloodshed (Popper, 1977).
Of utmost relevance for our historical perspective, modernization processes, such as economic development, urbanization, technological innovation, higher education and literacy levels, rising living standards, or expanding informational and communicational connectivity, have been assumed to facilitate democratization (Lipset, 1959;Murtin & Wacziarg, 2014;Welzel, 2019).For example, Boix and Stokes (2003) presented evidence consistent with the idea that economic development caused non-democracies to democratize, using an observation period between the mid-nineteenth century and World War II.However, many recent studies have failed to find such a direct link (Munck, 2018).In addition, Boix (2018) argued that there is limited evidence that income -used as a (rough) proxy for modernization -relates to democracy in the short term.In fact, as Dahlum (2018) noted, there seems to be limited evidence supporting the idea that the modernization level in a given year elicits an "unconditional, instantaneous change" (p.4).She continued that although "the thin version of modernization theory may not hold," recent studies "point to substantial evidence in favor of a more refined version of modernization theory, suggesting that increasing income promotes democracy i) in the medium or long-term and ii) [is] conditional on certain triggers of authoritarian regime breakdown" (p. 4, italics added).Welzel (2019) argued that "multiple aspects of modernization all work together in producing a joint outcome: enhancing the material, intellectual, and connective resources available to the median person," a process that increases the "masses" capabilities to launch and sustain collective actions for a shared purpose" (p.29, italics added).However, mere capability is not enough for a democratic transition, as "only when rising emancipative values awake in people an intrinsic desire for the freedoms that define democracy, do they begin to consider dictatorial powers as illegitimate" (p.32, italics added).Emancipative values may thus act as a "trigger" sensu Dahlum (2018).Mass opposition, in turn, may then grow to such an extent that power-holders are forced to open the door to a democratic transition (Welzel, 2019).As Welzel (2019) put it, "to advance democracy, people have not only to be capable to struggle for its advancement; they also have to be willing to do so" (p.33).

Role of the Media
The media are assumed to play an important role in democracy.Although normative implications for the media depend on the specific model of democracy that one prefers (Strömbäck, 2005), Voltmer and Rawnsley (2019) emphasized two very basic functions.First, the media should provide a forum for public debate.The media should enable citizens to make informed decisions, including those related to elections (Aalberg & Curran, 2011).The "marketplace of ideas" metaphor has often been used in this regard.Furthermore, the notion of a bourgeois public sphere that has emerged since the eighteenth century (Habermas, 1989) and the role of the press that contributed to a mediated public sphere and provided the potential for deliberation (i.e., allowing citizens to actively listen and respond to different perspectives, including those of opponents) is also relevant in this context (Wessler, 2018).Note that the media's forum function also includes its capability to coordinate opposition (see Bueno de Mesquita & Downs, 2005;King et al., 2013).Second, the media should monitor the behaviors of political decision makers, especially phenomena such as misconduct or the abuse of power.This has often been termed the "watchdog" role or is referred to when calling the media the "fourth estate."This "checking function" should ensure that representatives uphold their oaths of office and carry out the wishes of the electorate (Aalberg & Curran, 2011).To fulfil these basic tasks, the media must be independent from external interference.Freedom of the press is thus one of the most important civil liberties that constitutes democracy (see Olechowski, 2004).Conversely, in societies under authoritarian rule, the media are primarily seen as instruments of the rulers, and access to the media's forum tends to be restricted to voices consistent with the position of the ruling elite.The aim of autocratic rulers is to demobilize and repress citizens in order to maintain their own power (see Linz, 1975).

Toward the Mechanism: The Demonstration Effect
One important line of research focuses on the demonstration effect (Huntington, 1991;Voltmer & Rawnsley, 2019), theorizing on the role of the media as a possible driving force in the struggle for democratic change.The media has been seen as an important source of alternative information for countries where authoritarian regimes aim to maintain a monopoly over the distribution of information.Due to the demonstration of democratic ideals, such as freedom or equality in the media, citizens of a given authoritarian regime may become more informed about and encouraged by changes elsewhere.Citizens may recognize that democratic ideals demonstrated in the media world are at odds with their experiences in actual authoritarian political reality.Eighteenth-century key events, such as the American and French Revolutions, have been used as prime examples of democratization processes occurring throughout the nineteenth century in Europe.Citizens may learn what is absent in their lives, and by inspiring hope and motivation, the media may possibly stimulate specific aspirations for similar outcomes as those that have occurred in other countries, possibly leading citizens to increasingly begin to press for democratic change.Of note, even if citizens have already learned about events elsewhere, repeated ongoing reporting about democratic ideals may constantly "remind" citizens of democratic ideals, such as freedom or equality.Furthermore, ruling political elites may become anxious about their downfall and thus adapt their behavior to become more conciliatory or reactionary (see O'Neil, 1998).Taken together, theorizing in previous research on the demonstration effect provides a mechanism through which the media may elicit an influence on democratization, that is, via giving salience to democratic ideals, such as freedom and equality.
Note that the idea of the demonstration effect is related to democratic domino theory that more broadly assumes that increases or decreases in democratization in one country can spread and "infect" neighboring countries, in turn increasing or decreasing their level of democratization (Leeson & Dean, 2009).Importantly, however, Huntington (1991) noted that snowball effects are unlikely to bring about democratization if a country lacks favorable internal conditions.In fact, he emphasized that the democratization of countries A and B will probably not be a causal reason for democratization in country C, unless the conditions that favored it in the former also exist in the latter.The demonstration effect explicitly focuses on the role of the media in this regard.Through the demonstration effect, the media may contribute to more favorable internal conditions, possibly increasing the likelihood that democratic dominoes will fall.
When theorizing on the demonstration effect, we broadly assume a transfer in the salience of the idea of democracy from the media to the public (see McCombs & Shaw, 1972), that is, the more salient the idea of democracy is in the media, the more salient the idea of democracy is in the citizenry (and the ruling elite).Citizens can not only learn about the idea of democracy, but also about how much importance to attach to the idea of democracy from the amount of information in the media.Using a phrase provided by McCombs and Shaw (1972), the media may put the idea of democracy on the "agenda."Stated differently, the idea of democracy may become increasingly part of the "marketplace of ideas." The demonstration effect can work via two possible routes of influence.Spreading the idea of democracy in the media may lead to actual effects on citizens, such as rising expectations, possibly leading to emancipative values, and even a change in behavior (direct influence).However, it may also lead to changes in the behavior of those in power based on their expectations of a media effect on the ruling elite's subjects, that is, the anticipated influence of the reporting on the "dangerous idea of democracy" on citizens (influence of presumed influence, see A. Gunther & Storey, 2003).Regarding the latter, Huntington (1991) pointed to an East European example that allegedly had its principal effect on the leaders of other authoritarian regimes, not on the people they ruled.He pointed to President Mobutu Sese Seko of Zaire (today, the Democratic Republic of the Congo), who allegedly reacted with shock and horror to televised pictures of the execution by firing squad of Romanian dictator Ceaușescu.Huntington (1991) argued that this media event (and thus its presumed influence) may have influenced Mobuto's subsequent political behavior.
Previous research relied almost exclusively on analyses of specific key political events using a comparative and global perspective (R. Gunther et al., 2000;Voltmer & Rawnsley, 2019) or focused on the role of media technologies (Groshek, 2011;Wertz et al., 2021).Without a doubt, this research substantially contributed to our knowledge.Unfortunately, based on the available evidence, it is currently "impossible to state with any certainty that media coverage (. ..) led to political change" (Voltmer & Rawnsley, 2019, p. 243).Similarly, a report from the Reuters Institute for the Study of Journalism on media and democratization (Jebril et al., 2013) noted that it "is unclear whether the media are an agent of democratic change" (pp.2-3).We aimed to contribute to answering this open question.

The Present Research
Grounded in previous research on the demonstration effect, the present study proffers longitudinal evidence on the role of the press in democratization.We focused on Austria due to the availability of (content-analytic) longitudinal data and our own expertise in this specific political context.We studied the transition from authoritarian rule (absolute monarchy) to democracy (republic), spanning the nineteenth century to the first half of the twentieth century.We investigated the interplay between (1) the salience of the idea of democracy in the press and (2) the level of democratization between 1816 (after the Congress of Vienna) and 1932 (the last year of the First Republic, before the era of authoritarian rule and the years of the National Socialist dictatorship).
We want to emphasize that we view a possible effect of the press on democratization during this long time period as a complex process that is highly conditional on other institutional, social-structural, and micro-level factors -consistent with previous research (R. Gunther et al., 2000): The causal processes linking the media and democratization processes are not conceptualized as one-dimensional in a stimulus -response fashion.Conversely, our understanding of the effect process is incompatible with oversimplistic media effects models.Consistent with R. Gunther et al. (2000), we view the possible impact of the press on democratization during this long time period as strongly shaped by the interplay among a myriad of macro-level and micro-level factors.In fact, macro-level factors, such as the political culture, the structure of society, the media and government institutions, the norms governing the relationship between journalists and politics, regulatory practices, and the level of technological development, may contribute to shaping media messages as well as their patterns of circulation.Importantly, for our historical perspective, macro-level factors also include the processes of modernization discussed above.In addition, micro-level factors, such as citizens' attitudes, the amount of exposure to newspapers, and citizens' levels of education, may shape citizens' receptivity to news.
Given that we investigated a very long time period using a macro-level perspective, we were unable to study the contributions of all factors.We had to focus.Acknowledging this fact, we grounded our study on the available theorizing on the demonstration effect (Huntington, 1991;Voltmer & Rawnsley, 2019).We aimed to test the hypothesis that the press contributed to democratization via a mechanism that centers on the salient "demonstration" of the idea of democracy in the press (General Hypothesis).We focused on two predictions in this regard.
First, if the press contributed to democratization via the mechanism predicted by the demonstration effect, there should be an observable covariation between increases in the salience of the idea of democracy in the press and increases in democratization.Stated differently, if the quantity of reporting about democratic ideals, such as freedom or equality, increases in a given year, there should also be a similar increase in the level of democratization in that given year.Therefore, we predicted a covariation between increases in the salience of the idea of democracy in the press and increases in the level of democratization (Hypothesis 1 [H1]).Second, we predicted that year-to-year changes in the media would not only be statistically related to beneficial political developments in the same years (as predicted in the first hypothesis), but would act as a causal agent of change.By assuming that the cause must precede the consequence, we predicted that increases in the salience of the idea of democracy in the press would prospectively predict future increases in the level of democratization (Hypothesis 2 [H2]).Please note that both hypotheses are not competing hypotheses, but focus on different aspects (i.e., mere statistical association vs. a causal agent).

Method
We used a longitudinal time series approach, which is considered one of the best designs for establishing causality when experiments are neither feasible nor ethical (Schaffer et al., 2021).We relied on two time series: (1) the salience of the idea of democracy in the press and (2) the strength of the democratization of the political system.

Salience of the Idea of Democracy in the Press
The first time series informs us about the salience of the idea of democracy in the press (SIDP).We used a content analysis of Vienna-based newspapers that were published between 1816 (i.e., after the Congress of Vienna) and 1932 (i.e., the end of the First Republic) by relying on a historical newspaper archive (Austrian Newspapers Online [ANNO], 2022).First, we selected all available daily newspapers published in Vienna, the then-capital of the Habsburg Monarchy, and, later, of the First Republic (i.e., the political center during the entire observation period).As indicators for the idea of democracy, we assessed the quantity of reporting of key terms associated with the concept of democracy for each year of the observation period: Freiheit (freedom), Gleichheit (equality), Demokratie (democracy), and Wahlen (election).We want to emphasize that we conducted a pretest to try to find adequate keywords that can be used in the computer-assisted identification of democracy-related news coverage.We tried several different terms, including verbs, before data collection, but we came to the conclusion that these four nouns represented good indicators.They have enormous face validity and worked very well in our "plausibility checks" conducted before data collection.
Each hit in the newspaper archive informs us that one newspaper reported on one day about the given search term in a given year.Given that we used all available newspapers published in Vienna, the annual scores can be above 365.We assessed the number of hits for each year, resulting in four time series corresponding to the four key terms.The general measurement strategy was to collect this quantity of reporting-based data from the newspaper archive and, afterward, to use principal component analysis (PCA) over these four time series.A resulting one-component solution from the PCA was used for all subsequent analyses.
Note that what was deemed of primary relevance for the study of long-term macro-level effects was how salient the idea of democracy was in a given year within the whole "marketplace of ideas."What counts for the present research is the press as a message system (see Gerbner & Gross, 1976), not the individual newspapers.Thus, we did not focus on individual newspapers but on the press as a whole.As we wanted to consider technological innovations, such as the rotary press, and the reduction of economic and legal restrictions that increasingly allowed for the production of high-circulation newspapers (Seethaler & Melischek, 2022) and possible problems related to the historical newspaper archive (e.g., accumulated missing issues or pages in some years), we used three variants of the SIDP measure.We now report on the three measures based on different scoring algorithms.

Absolute SIDP Score
This measure assesses the number of hits for each of the four key words for each year of the observation period, resulting in four time series: freedom (M = 1779.68,SD = 1209.60,range = 409 − 4201), equality (M = 310.47,SD = 223.68,range = 20 − 1076), democracy (M = 370.90,SD = 445.90,range = 0 − 2376), and election (M = 1499.99,SD = 1108.65,range = 49 − 3406).PCA was used to reduce dimensionality and to obtain a standardized time series, which is a linear combination of the four original time series provided by the indicators.One component explained 85.04% of the variance (Eigenvalue = 3.40).Loadings were high (i.e., freedom = .98,equality = .96,democracy = .80,and election = .94)and thus indicated unidimensionality.This measure assesses the quantity of reporting about democratic ideals in absolute numbers and thus does not consider how many newspapers were available in the archive in a given year, how many issues of a given newspaper were typically published within a week, or how many pages a specific issue typically had.All of these limitations are relevant for the present research, as the numbers increased on all three dimensions throughout the observation period.Although the absolute score is straightforward to interpret, it has limitations.For example, the score can also increase when democratic ideals do not appear more frequently within an issue, as the score can increase due to more available newspapers in a given year based on contextual factors, such as technological innovations or less economic and legal restrictions.More newspapers, more reporting.
The other two variants of the SIDP measure aim to account for these limitations.Nevertheless, absolute frequencies are helpful for an initial understanding of how reporting changes over time.

Percentage SIDP Score
This measure assesses the percentage of issues in which democratic ideals are reported.The basis for this measure is the absolute score.It also assesses the number of hits for each of the four key words for each year of the observation period in a first step.Next, however, we divided these numbers by the total number of newspaper issues that were available in a given year in the newspaper archive.Unfortunately, this total number was not directly available.We had to use a proxy.We operationalized the total number of available issues as the number of hits in the archive when searching for the letter "a."The idea behind this is that this letter simply appears in all issues.It is simply inconceivable that there would be a newspaper issue that did not include the letter "a."After multiplying by 100, this resulted in four percentage-related time series corresponding to the four words: freedom (M = 51.50%,SD = 21.36%,range = 10.68%−94.40%),equality (M = 9.21%, SD = 4.30%, range = 1.98%−24.62%),democracy (M = 9.16%, SD = 8.11%, range = 0.00%−37.12%),and election (M = 42.12%,SD = 20.82%,range = 4.81%−73.43%).PCA resulted in a onecomponent solution that explained 76.50% of the variance (Eigenvalue = 3.06) with high loadings (i.e., freedom = .94,equality = .90,democracy = .73,and election = .91).

Relative SIDP Score
Whereas the percentage score considers the number of newspaper issues available in a given year, an improved relative score also considers the fact that single issues of a newspaper have different lengths.In fact, the number of pages per issue varied substantially over the observation period -it increased, for example, due to technological innovations.In addition, some issues were apparently incomplete within the archive due to lost pages.Unfortunately, the exact number of (available) pages per issue was not available.Again, we used a proxy: We assessed the salience of four "control search terms" in the press reporting.The idea was that, if a given issue had a lower number of pages, there would be a reduced likelihood that this issue would report on democracy-related target terms and democracy-unrelated control terms.Less pages, less reporting.The relative score thus additionally assesses increases in the salience of the idea of democracy relative to increases in the salience of democracyunrelated control terms.We selected the following four control terms: Musik (music), Holz (wood), Sonne (sun), and Tisch (table).The selection was guided by the aim of using control terms that were unrelated to democracy or the political system.We assessed the absolute number of hits for each of these four control terms, resulting in four time series corresponding to the four words.Next, we calculated a percentage score as described above: music (M = 59.39%,SD = 22.15%, range = 18.70%−86.31%),wood (M = 46.78%,SD = 12.44%, range = 22.93%−81.66%),sun (M = 41.40%,SD = 16.73%,range = 10.41%−71.28%), and table (M = 39.97%,SD = 16.17%,range = 12.42%−65.98%).PCA resulted in a one-component solution that explained 85.92% of the variance (Eigenvalue = 3.44) with high loadings (i.e., music = .98,wood = .92,sun = .88,and table = .93).Afterwards, we subtracted the (democracy-unrelated) control search term score from the (democracy-related) target search term score for each year of the observation period, resulting in a "relative" time series, constituting the relative SIDP score.This scoring algorithm considers (1) the quantity of reporting about the four democratic concepts (see absolute score); (2) how many newspapers were available in the archive in a given year (see percentage score); and (3) additionally -broadly speaking -the issue length (available in the archive).This scoring algorithm is probably the most sophisticated of the three, as it handles the limitations of the other two scoring algorithms.Due to issues of transparency, we report the results for all three scores.

Democratization
The second time series informs us about the societal level of democratization.In a recent review on measures of democracy, Vaccaro (2021) provides information regarding the statistical properties, convergence, and interchangeability of four of the most frequently used measures.One of these measures (Freedom House, 2019) only provides data beginning in 1972 and thus cannot be used in the present study.The remaining measures provide annual data for the entire observation period.Given that all measures have limitations (Vaccaro, 2021), we decided to use all measures.All three measures conceptualize democracy as a graded concept, implying that political systems can be placed on a continuum (see Bernhagen, 2019).
First, Vanhanen (2000) combined two basic dimensions of democracy -competition and participation.Based on Dahl's (1971) two theoretical dimensions of democracy (polyarchy), Vanhanen argued that the two most important dimensions of democracy are the degree of competition and the degree of participation.He thus based his measure on the objective (i.e., documented) results of elections.Competition was measured by subtracting the percentage of votes won by the largest party from 100.Participation was assessed as the percentage of the population who actually voted in these elections.Vanhanen (2000) combined the two variables into his Index of Democratization by multiplying them and dividing the product by 100.This was done because both concepts are necessary for democracy, and a high level of competition cannot compensate for a lack of participation, and vice versa.
Second, the Polity 5 index (Marshall & Gurr, 2020) is based on historical sources, including constitutions and academic publications.Of note, this index is composed of the "Institutional Democracy" and "Institutional Autocracy" sub-indices, aiming to measure democratic and autocratic features.Both are then combined into the Polity 5 index by subtracting the Institutional Autocracy score from the Institutional Democracy score.We used the Revised Combined Polity Score, as recommended for time series analyses.
Third, we relied on the Civil Liberties Index (CLI) from the Historical Varieties of Democracy (Historical V-Dem) Project (V-Dem, 2022).Although elections (Vanhanen, 2000) or historical sources, such as constitutions (Polity 5), are important for assessing the level of democratization, narrowing down democracy merely to elections or a constitution "is too restrictive;" instead, "broader sets of freedom" (p.34) should also be included, as Welzel (2019) argued.Consistent with this notion, we used the CLI to broaden our perspective by additionally considering the extent to which civil liberty had been respected.This index understands civil liberty as liberal freedom, defined as a property of individuals, assumed to be constituted by the absence of physical violence committed by government agents and the absence of constraints on private and political liberties by the government.This index is based on expert coding performed by research assistants and historical country experts.
Each of the three measures views democratization from a (slightly) different perspective.Given that we wanted to use a measure that considers elections, historical documents, such as constitutions, and civil liberties, we decided to combine the three measures into a single democratization index.PCA resulted in a one-component solution that explained 85.90% of the variance (Eigenvalue = 2.58).High loadings (i.e., Vanhanen's index = .91,Polity 5's index = .98,and V-DEM's index = .89)indicated unidimensionality.The resulting time series considers elections, historical documents, such as constitutions, and civil liberties (rated by historical country experts) to assess the level of democratization in a given year.This measure was used for all subsequent analyses.

Statistical Analysis
To test for covariation (H1), we relied on autoregressive integrated moving average (ARIMA) modeling to account for autocorrelation and the trend in the time series (see R. J. Hyndman & Athanasopoulos, 2014;Schaffer et al., 2021).The latter is important because there can be increases (or decreases) related to other influences.ARIMA can handle general trends and assesses whether the salience of the idea of democracy in the press predicts the level of democratization when autocorrelation and the trend are accounted for.Note that we followed a recent guide for effects research using ARIMA modeling provided by Schaffer et al. (2021).More specifically, Schaffer et al. (2021) noted that automated algorithms in statistical packages now exist that simplify the process by identifying the bestfitting ARIMA model based on minimizing information criteria.This is substantial progress, as in the past, the manual model selection process has often been considered somewhat subjective.Following Schaffer et al. (2021), we used an algorithm that was implemented in the forecast package in R (R. Hyndman & Khandakar, 2008).This algorithm helps to reduce bias in research by making the model selection process more transparent.Importantly, we checked the appropriateness of the identified models via a test of residual autocorrelation.Note that ARIMA modeling with a regressor can also be seen as a "regression with ARIMA errors."Thus, the effect coefficients reported below can be interpreted similarly to the coefficients in a regression analysis, controlling for autocorrelation and the trend in the time series.
To test for a causal contribution (H2), we assessed whether the salience of the idea of democracy in the press in a given year would predict future increases in democratization.We relied on the concept of Granger causality (Granger, 1969) in this regard, helping to formally test whether there is evidence of a specific causal order between the idea of democracy in the press and democratization.Note that the concept of Granger causality assumes that the time series of the salience of the idea of democracy in the press "Grangercauses" (i.e., prospectively predicts) the time series of democratization if the lagged values of the salience of the idea of democracy in the press (t − 1, t − 2, etc.) provide statistically significant information about the future values of democratization (t) when simultaneously controlling for the lagged values of democratization (t − 1, t − 2, etc.).The idea is that the cause must precede the consequence.

Preliminary Analysis: Visual Inspection
As can be seen in Figure 1, democratization shows an increasing trend during the observation period.A similar increasing trend can be found in the absolute SIDP score.Note that the relative scores do not show a comparable strong trend due to their underlying scoring algorithms.However, the latter showed substantial spikes during the observation period.
(Note that the difference in the trend in the time series is not essential for ARIMA modeling and Granger causality tests, as the trend is controlled for in these analyses.What is of primary interest for our study of long-term effects is the year-to-year variation.)Changes in democratization and SIDP scores occurred against the backdrop of key political events (Austrian Parliament, 2021; see also Rathkolb, 2008): 1848 (Revolution: protests by citizens; the revolution was quashed by military force); 1861 (after an era of neoabsolutism and lost wars, the Emperor made concessions: the "Kurien" election system was introduced with indirect suffrage for the precious few); 1867 (Constitution; turned Austria into a constitutional monarchy; press freedom was constitutionally guaranteed); 1873 (census election system); 1882 (voting costs less); 1896 (Badeni's electoral reform: more voters but no equality); 1907 (male suffrage); and 1918 (universal suffrage in the First Republic).

Is There a Covariation Between the Press and Democratization?
H1 predicted a covariation between increases in the salience of the idea of democracy in the press and increases in the level of democratization.Figure 1 visualizes the four relevant time series: three press-related time series with different scoring algorithms and one time series related to the level of democratization.To test the hypothesis, we relied on ARIMA modeling to account for autocorrelation and the trend in the time series.As recommended by R. J. Hyndman and Athanasopoulos (2014), we first ran a KPSS test (Kwiatkowski et al., 1992), which indicated that first differencing was necessary to create a stationary time series.Note that a stationary time series is a prerequisite for ARIMA modeling.We used the automated auto.arima()function in R that uses a variation of the Hyndman -Khandakar algorithm (R. Hyndman & Khandakar, 2008), which combines unit root tests, minimization of the AICc, and maximum likelihood estimation criteria to obtain an ARIMA model (R. Hyndman & Khandakar, 2008).Based on the results of the KPSS test, we prespecified the d term in the ARIMA model (d = 1).The automated algorithm was allowed to automatically select the most appropriate values for the remaining terms, that is, p, q, P, D, and Q, which define the ARIMA model (see Schaffer et al., 2021).
For the relative SIDP score, we identified an ARIMA (0,1,1) model.This model accounted for the autocorrelation in the time series, as indicated by a Ljung-Box Q test on residual autocorrelation that was not significant, χ 2 = 0.81, df = 10, p = .999.Importantly for the test of the hypothesis, the analysis provided evidence of a significant covariation between increases in the salience of the idea of democracy in the press and increases in the level of democratization, B = 0.09, 95% CI [0.04, 0.13], SE = 0.02, z = 3.51, p < .001.Given that this score is probably the most sophisticated of the three because it handles the limitations of the other two, we prefer to test the hypotheses with this measure.
For the percentage SIDP score, we also identified an ARIMA (0,1,1) model, χ 2 = 1.20, df = 10, p = .999.The effect coefficient also indicated a significant covariation pattern, B = 0.12, 95% CI [0.07, 0.17], SE = 0.03, z = 4.54, p < .001.For the absolute SIDP score, we also identified an ARIMA (0,1,1) model, χ 2 = 0.54, df = 10, p = 1.000.Although the effect coefficient points in the predicted direction, our findings did not indicate a significant covariation, B = 0.05, 95% CI [−0.03, 0.13], SE = 0.04, z = 1.26, p = .207.Although the absolute SIDP score did not provide a significant finding, the weight of the evidence seems to support the idea of a covariation between the salience of the idea of democracy in the press and the democratization of the political system.We interpret the findings as support for H1.

Do the Press Predict Future Increases in Democratization?
H2 predicted that increases in the salience of the idea of democracy in the press would predict future increases in the level of democratization.We relied on a Granger causality test (Granger, 1969;see;Boix, 2018;Groshek, 2011, for an application of this analytical technique in the democratization literature).Thus, we aimed to formally test whether there was evidence of a specific causal order between the idea of democracy in the press and democratization.First, we used the VARselect function of R's vars package to identify the optimal lag structure (i.e., how many lagged values we should consider in our analysis).This function returns the AIC for sequentially increasing the lag order.We defined the lag structure based on this criterion.Second, we ran the Granger causality test using the grangertest function of R's lmtest package.This analysis relies on a Wald test that compares the unrestricted model (i.e., democratization is predicted by the lags of democratization and the lags of the salience of the idea of democracy in the press) and the restricted model (i.e., democratization is predicted only by the lags of democratization).Note that we also used first-order differencing in this analysis, as in the ARIMA modeling reported above.
For the relative SIDP score, we identified an optimal lag structure of 10 based on the AIC.The Granger causality test provided evidence for a directional influence: The salience of the idea of democracy in the press Granger-caused (i.e., predicted) future values of democratization, Wald's F = 2.20, p = .025.This indicated that changes in democracy-related reporting in a given year predicted changes in future levels of democratization.Although the individual coefficients of the lagged variables are difficult to interpret, we additionally ran a vector-autoregressive model using the VAR function of R's vars package.Among the ten lagged SIDP variables, one lagged variable (t − 2) showed a significant effect coefficient, B = 0.13, SE = 0.03, t = 3.89, p < .001.Conversely, democratization did not Granger-cause the salience of the idea of democracy in the press, Wald's F = 0.39, p = .946.As indicated by a vector-autoregressive model, none of the individual lagged democratization variables were related to subsequent increases in the relative SIDP score.
For the percentage SIDP score, we identified an optimal lag structure of 5. Again, we found evidence for a directional influence: The salience of the idea of democracy in the press Granger-caused future values of democratization, Wald's F = 4.55, p < .001.As indicated by a vector-autoregressive model, one lagged variable (t − 2) showed a significant effect coefficient in the predicted direction, B = 0.13, SE = 0.03, t = 3.91, p < .001.Unexpectedly, a second lagged variable (t − 1) also showed a significant effect, albeit that the coefficient pointed in the reverse direction, B = −0.08,SE = 0.03, t = −2.37,p = .020.Although this negative sign was not predicted, the net effect is still in the predicted direction, as the former lagged variable (t − 2) showed a stronger effect coefficient.Conversely, democratization did not Granger-cause the salience of the idea of democracy in the press, Wald's F = 0.84, p = .525.None of the individual lagged democratization variables was related to subsequent increases in the idea of democracy.
For the absolute SIDP score, we identified an optimal lag structure of 4 based on the AIC.The Granger causality test provided evidence for a directional influence: The salience of the idea of democracy in the press Granger-caused future values of democratization, Wald's F = 9.56, p < .001.Among the four lagged idea-of-democracy variables, one lagged variable (t − 2) showed a significant effect coefficient, B = 0.24, SE = 0.04, t = 5.62, p < .001.Conversely, democratization did not Granger-cause the salience of the idea of democracy in the press, Wald's F = 2.36, p = .058,and none of the individual lagged democratization variables were related to subsequent increases in the absolute SIDP score.Taken together, the weight of the evidence supports the idea of the press as a predictor of future changes in democratization, supporting H2.
Taken together, findings seem to indicate that (1) there was a covariation between the press and democratization and (2) the press predicted future increases in democratization.Table 1 provides a summary of these two key findings.

Discussion
Despite the available evidence on the role of the media in democratization, which is largely based on a comparative and global perspective on recent key political events or on studies on the role of emerging media technologies, there is limited longitudinal empirical evidence.As our primary contribution to the literature, we provide a long-term macro-level perspective ranging over a century, including the transition from authoritarian rule to democracy.The findings indicated that the salience of the idea of democracy in the press was statistically related to the level of democratization: The strength of the year-to-year increase in the salience of the idea of democracy in the press was positively related to the year-to-year increase in the level of democratization.Furthermore, and possibly even more importantly, we found that increases in the salience of the idea of democracy in the press Granger-caused (i.e., prospectively predicted) future increases in the level of democratization.By assuming that the cause must precede the consequence, the latter finding is consistent with the concept of a long-term macro-level media effect.
To the best of our knowledge, such long-term evidence on the role of the salience of the idea of democracy in the press is unprecedented in the study of the role of the media in democratic transitions.Our findings thus contribute to the political communication literature on the role of the media in democratization, especially to research on the demonstration effect.Our findings also contribute to more general research on

There Was a Covariation Between the Press and Democratization
Using autoregressive integrated moving average (ARIMA) modeling to account for autocorrelation and the trend in the time series, analysis indicated that there was a covariation between increases in the salience of the idea of democracy in the press and increases in the level of democratization.In fact, the strength of the year-to-year increase in the salience of the idea of democracy in the press was positively related to the year-to-year increase in the level of democratization.
The weight of the evidence seems to support H1.

The Press Predicted Future Increases in Democratization
Furthermore, and possibly even more importantly, we found that a higher salience of the idea of democracy in the press in a given year "Granger-caused" (i.e., prospectively predicted) future increases in the level of democratization.
(Conversely, democratization did not prospectively predict future increases in the salience of the idea of democracy in the press.)Therefore, the weight of the evidence seems to support the idea of the press as a predictor of future changes in democratization, supporting H2.Acknowledging the fact that the cause must precede the consequence, this finding is consistent with the idea of the press as a causal agent of change (i.e., a long-term macro-level media effect).However, there are limitations in terms of causal interpretations.
democratization in the political science literature, especially to research on the role of modernization processes, as revealing the effects of the press, which was a part of modernization processes, supports the claim that modernization processes have contributed to democratization.Given that many previous studies used economic growth as a (rough) indicator for modernization, evidence related to the press presented in the present paper is a valuable supplement, enriching our knowledge base.We are very careful when interpreting our findings.We acknowledge the limitations in terms of causal interpretations.For example, we provide evidence that an increase in the salience of the idea of democracy in the press preceded democratic change.Acknowledging the fact that the cause must precede the consequence, this finding is consistent with the idea of the press as a causal agent of change.However, this finding cannot prove causal effects.It is possible that increases in the salience of the idea of democracy in the press preceded democratic change without the press eliciting a causal influence.One finding is important in this regard: The main cause of universal suffrage in 1918 was not the press or other internal factors, but an external factor (World War I; see Rathkolb, 2008).Although the press increasingly reported on the idea of democracy in 1917, and thus an increase in salience preceded this key event (see Figure 1), the First Republic was based on "imposed democratization," using Welzel's typology of democratization (Welzel, 2019).However, who knows whether the foundation of the First Republic, universal suffrage (1918), and the Constitution (1920/1929) would have been realized in that way and so fast had the media not provided the breeding ground by making salient the idea of democracy before regime change.
Despite the complexity of this phenomenon, we generally view the nineteenth century and the Belle Époque as an emancipatory path toward democratization (Welzel, 2019), and we carefully argue -based on the empirical evidence provided in the present paper -that the press may have contributed to this process.Consider the Constitution (1867).The salience of the idea of democracy in the press swelled during the preceding years (see Figure 1).Combined with other political events that weakened the monarch's power (e.g., the Battle of Königgrätz in 1866, which was a decisive battle in the Austro-Prussian War in which the Kingdom of Prussia defeated the Austrian Empire), we think that it is plausible to assume that the press causally contributed to democratization -as one of many factors.The demonstration effect was put forward as a possible mechanism.Broadly speaking, increasingly spreading the idea of democracy in the press may have led to actual effects on citizens (i.e., direct influence).However, increasingly spreading the idea of democracy in the press may also have led to changes in the monarch's behavior based on his presumed effect of salient reporting on the "dangerous and toxic idea of democracy" on citizens (i.e., influence of presumed influence, see A. Gunther & Storey, 2003).The latter may be especially pronounced in a time period in which the ruling elite was weak (e.g., after a lost decisive battle).Although the monarch never gave up the concept of the monarchy, he made numerous concessions, leading, over time, to a higher level of democratization.We acknowledge that answering questions about a long-term macro-level media effect and its underlying mechanism(s) with increased confidence definitely requires more research.We see our study as a first step.
By "using the past to study the present" (Lawrence, 1984), the evidence of the present study may also contribute to current understandings of the role of the media in democratic transitions by highlighting the media's earlier phases and subsequent evolution.For example, our historical findings support the claim made by R. Gunther et al. (2000, p. 5) in current times that it is a much too simplistic view when arguing that the media in nondemocratic regimes is only associated with the suppression of popular, accountable governments, while crucially contributing to their healthy functioning in democracies.This claim is echoed by Voltmer and Rawnsley (2019), who argued that "only few authoritarian regimes manage to exert full control over the media messages" (p.241).Thus, even if there are severe restrictions, it is likely that some "inconvenient" messages will find their way to citizens -especially in the context of the process of liberalization that has evolved throughout the century, that is, the relaxation of state control over the press.In fact, restrictions were highly severe during the first half of the nineteenth century during the Vormärz ("pre-March") period (i.e., before the March Revolution in 1848), which included a repressive absolutist police state built upon censorship and mass surveillance, but which became increasingly relaxed over time.

Limitations
This study has several limitations.First, we want to emphasize again that we are very careful when interpreting causal effects, and we especially acknowledge the limitations of our historical data in this regard.Second, we did not investigate the question of why newly formed democracies, following the breakup of the big European empires in 1918, could not withstand the emergence of fascist and totalitarian movements in the first half of the twentieth century (see Seethaler & Melischek, 2022).In fact, we studied the period between authoritarian rule (Habsburg Monarchy) and democracy (First Republic).Future research could also study the role of the media in the reverse direction: the transition from a democracy to authoritarian rule.A thorough understanding of this question is of relevance, as authoritarian tendencies are resilient in some parts of today's political world (Haerpfer et al., 2019).It seems possible that media effects are somewhat different during a transition from a democracy into authoritarian rule compared to countries that transition from authoritarian rule to democracy.A higher salience of the idea of democracy in the media in established democracies with an emerging resilience to authoritarian tendencies may sometimes even be indicative of a weaker level of democratization, as journalists may bring specific threats to democracy, such as corruption, into public awareness.Third, the press was part of modernization processes, and changes in the press occurred against the backdrop of other societal change processes, such as industrialization, urbanization, secularization, and phenomena relevant for the readers' market, such as higher living standards and literacy levels.Technological innovations, such as the rotary press, increasingly allowed for the production of high-circulation newspapers, and more liberal press laws during the second half of the nineteenth century had expanded citizens' opportunities for participation in the public sphere (Seethaler & Melischek, 2022).We focused only on the press.Future research may wish to investigate the relationship between the press and other macrolevel processes.It seems to be worth testing whether the strength of a long-term macro-level media effect depends on other macro-level factors, such as literacy rates.For example, do literacy rates moderate the strength of the media effect?Fourth, we used four search terms for the content analysis (i.e., freedom, equality, democracy, and election).Although PCA provided evidence for unidimensionality, we do not know whether this keyword approach provided all of the relevant news content.For example, journalists may have written "between the lines," making it difficult for our keyword approach to identify such content.This may raise validity concerns.Although we conducted several successful "plausibility checks" before data collection (i.e., we read numerous articles to assess whether the identified texts actually reported on democratic ideals), a future study could rely on human coders and conduct a large-scale content analysis of democracy-related news coverage throughout the entire observation period.It seems conceivable that an SIDP score that differentiates between the mere quantity of reporting about democratic ideals (as in the present study) and how the press covered democratic ideals (e.g., with a positive or negative valence) may allow for an even better prediction of future changes in democratization.Thus, the present study may eventually even underestimate actual effect sizes.It is important to note that even if an article provides the meta-narrative of "democratic ideals, such as freedom and equality, not being good," this text may still increase the salience of the idea of democracy in readers' minds, as this article would still actually mention democratic ideals (see Gawronski & Bodenhausen, 2006, p. 710).

Conclusion
Acknowledging the limitations in terms of causal interpretations, we provide supporting tentative evidence for the idea of a long-term macro-level media effect over a century.Even if the process between the media and democratization is highly complex and a myriad of intertwined factors contribute to it, the longitudinal evidence we present is consistent with the claim that the press may have been one contributing factor -among many others.

Figure 1 .
Figure 1.Salience of the idea of democracy in the press (three scores) and democratization.

Table 1 .
Summary of key findings.