Outside lobbying

Outside lobbying describes “attempts by interest group leaders to mobilize citizens outside the policymaking community to contact or pressure” policymakersFootnote ²⁰ . Outside lobbying has two components.

The first component of outside lobbying is conflict expansion. Interest groups need to raise issue-salience, lower the costs of mobilization, and change evaluations of an issue (framing)Footnote ²¹ . Increasing issue-salience is important to gain attention and to mobilize existing supportersFootnote ²² . More importantly, however, most interest groups need to recruit new supporters through framing.

Framing describes the selection of “some aspects of a perceived reality and mak[ing] them more salient in a communicating text […] to promote a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation”Footnote ²³ . A frame therefore provides a rationale for a position toward a policy. Since most policies and issues are multidimensional, interest groups can use framing to highlight those issue dimensions that align with their preferred positions in order to influence public or elite opinionFootnote ²⁴ .

Pushing a preferred frame, however, does not in itself influence policy decisions. Instead, interest groups need to change the collective definition of an issue in order to be influential. Successful framing, thus, depends on how well a frame sticks with the target audience. A frame’s stickiness is determined by the perceptibility of the target audience (the bonding surface) and the characteristics of competing interest groups (the glue).

First, frames need a perceptible bonding surface to stick to. Chong and Druckman argue that successful frames are connected to “a positive idea or value that is widely available in the population”Footnote ²⁵ . Similarly, successful frames relate to dimensions previously emphasized in related debates. For instance, frames that “give[…] a coherent overall structure to a number of arguments that have […] been around for decades”Footnote ²⁶ are stickier. Finally, interest groups can enjoy first-mover advantages when framing a new issue or face obstacles when pushing a frame in a longstanding and cluttered issue areaFootnote ²⁷ .

Second, a frame sticks to the right surface only if the right glue is used. The credibility of the source from which a frame originates and the perception of an interest group as manipulative influence the stickiness of frames.Footnote ²⁸ . At the coalition level, the unity and frequency with which an interest group coalition pushes a frame affect its success.Footnote ²⁹ As framing is a competitive processFootnote ³⁰ , competing sides can challenge each other’s credibility and push counterframes that undermine the original framingFootnote ³¹ . Therefore, not only the stickiness of one’s own frames but also that of the frames pushed by opposing interest groups determines lobbying successFootnote ³² .

In addition to conflict expansion, outside lobbying involves signaling public opinion and electoral consequences to policymakers. Signal strength is crucial for outside lobbying successFootnote ³³ . KollmanFootnote ³⁴ argues that interest groups already supported by public opinion can send more credible signals to policymakers because they face smaller mobilization costs. While in the absence of outside lobbying campaigns the relationship between public support and lobbying success is negativeFootnote ³⁵ , outside lobbying is indeed more successfulFootnote ³⁶ when public support is high. Additionally, big and diverse coalitions signal broad societal support to policymakers and are more successful (outside) lobbyistsFootnote ³⁷ , especially under conditions of high salienceFootnote ³⁸ . Further, signal strength is also determined by how successful conflict expansion was, i.e., issue salience, mobilization, and whether the framing is unambiguousFootnote ³⁹ .

To sum up, outside lobbying success depends on initial public opinion, an interest groups’ ability to expand the conflict, and the strength of the signal to policymakers. Given that business tends to lobby against public opinion and is more successful when salience is lowFootnote ⁴⁰ , it is unsurprising that outside lobbying is more frequently used by labor unions and citizen groupsFootnote ⁴¹ . Dür and MateoFootnote ⁴² argue that citizen groups need to signal their engagement to a diffuse group of supporters to ensure organizational survival, whereas business organizations represent “a clearly defined constituency for which political decisions have concentrated costs and benefits”. Further, business can typically afford resource intensive inside lobbying and possess specialized information that they can trade for accessFootnote ⁴³ . However, this does neither mean that business does not use outside lobbying strategiesFootnote ⁴⁴ nor that it cannot do so successfullyFootnote ⁴⁵ . In fact, I argue below that platforms, as a special type of business, are uniquely positioned to mobilize the public.

Platform outside lobbying and its limits

I argue that because of the techlash platforms face an uphill battle against public opinion. However, platforms have three advantages for outside lobbying built-in their business model. Through their intermediary position they can reach network users easily by posting messages on their apps or websites. Because of the winner-takes-all logic of platforms their reach exceeds that of even the biggest conventional interest groups both in numbers and in diversity. Platforms also have free and frequent access to users because they are embedded in users’ daily routines. Users approach the platform, while conventional interest groups have to approach supporters to expand conflict. Third, because platforms surveil all network interactionsFootnote ⁴⁶ , messages can easily be targeted to users.

This combination of reach, access, and targeting provides platforms with a built-in solution to successfully make people aware of an issue (salience), frame it, and provide users with a course of action (mobilization). Figure 1 displays the mechanism of platform outside lobbying. The central independent variable is platform opposition to legislation. To prevent legislation, platforms engage in conflict expansion (part I), i.e., influence the debate in traditional and social media by (1) increasing issue-salience, (2) pushing favorable frames, and (3) mobilizing the public. As a result, media reporting increases and issue evaluations change. Additionally, the public is mobilized and joins protest actions such as demonstrations, online petitions, or letters/emails to their representatives (part II). While actions by the traditional media, social media, and the public reinforce each other (grey dashed arrow), the public debate signals electoral consequences to legislators. The signal is stronger the more unified the debate. Legislators respond by retracting their support (outcome).

Figure 1. Summary of the platform outside lobbying mechanism.

Following the structural perspective of platform power, successful platform outside lobbying should not be surprising in copyright legislation because consumers enjoy free access as a direct benefit from weak copyright. Indeed, in 2012 platforms successfully used outside lobbying against copyright legislation in the USFootnote ⁴⁷ . In the case of the EUCD, however, the techlash provided a perceptible bonding surface for anti-platform and pro-copyright frames. I therefore expect (1) that in countries where the public was less (more) accustomed to critical reporting on Big Tech, platform outside lobbying has better (worse) success chances. I expect further (2) that the characteristics of competing interest groups influence the success of platform outside lobbying. This involves the relative frequency with that a frame is pushed, the unity of the coalition pushing it, and the credibility of an interest group pushing a frame. For instance, SMEs are more credible than big businesses when highlighting the exploitative nature of platforms. Conversely, frames that emphasize user rights are more credible when criticizing major rightsholders rather than artists. In the following section, I present my methodological approach to test these expectations.

Combining process-tracing and time-series methods

Process-tracing aims to uncover the mechanism, i.e., the entities and activities that connect a cause and an outcomeFootnote ⁴⁸ . Inferences in process-tracing are based on context-specific within-case observations that are compared to the theoretical expectations. While the compatibility between quantitative evidence and process-tracing is disputedFootnote ⁴⁹ , I argue that time-series analysis is ontologically compatible with process-tracing. Time-series analysis analyses within-case observations and has the central assumption of temporal precedence. Time-series analysis can therefore produce evidence for/against the sequence proposed in a mechanismFootnote ⁵⁰ .

Combining process-tracing with time-series analysis is beneficial in two ways. First, by explicitly theorizing the mechanism and collecting qualitative evidence, process-tracing adds depth to time-series analysis and contextualizes findings of temporal precedence. Second, quantitative analysis is beneficial in situations where too many entities are engaged in a process for an encompassing qualitative analysis (e.g., interest group lobbying) and steps of the mechanism are easier to measure quantitatively (e.g., framing and salience).

My goal is to uncover the scope conditions of successful platform outside lobbying amid the techlash. Thus, I compare a deviant case with a typical caseFootnote ⁵¹ . After analyzing the debate on the EUCD in eight languages, I decided to dive deeper into the German and French cases. In Germany platform outside lobbying succeeded despite the techlash in signaling public opposition to MEPs. In France, in contrast, platform outside lobbying failed to create this signal.

Data collection and dictionary construction

To quantitatively trace the process, I collected data from Twitter, national newspapers of reference, the EU transparency register, and interest group websites. I collected 3,000,000 tweets and 1010 newspaper articles from 11 EU countries on the EUCD via the Twitter Academic API, Nexis Uni and Factiva.Footnote ⁵² For a list of newspapers and search queries see appendix A1-A2.

Second, I collected data on the interest groups active on copyright by scraping the EU transparency register through lobbyfacts.eu. I also included interest groups that submitted a non-anonymous answer to the public consultation on the review of the EU copyright rules of 2014. I scraped data on their inside lobbying activities, their website entries referring to the reforms, and their Twitter handles (see appendix A3 for a detailed description of data gathering)Footnote ⁵³ .

Finally, I manually coded the interest groups’ type as either rightsholder (companies and business associations from media, publishing, telecom, manufacturing, and the chemical sector); creatives; platforms and their business associations; the civil society coalition; lobbying agencies; other; and a category including firms that are platforms and substantial rightsholders (Platform & ICT & Media, e.g., Microsoft, Netflix).

Using interest groups testimonies in their consultation responses, public communications on blogs and websites, and platforms’ online protests I constructed an original dictionary. The dictionary aims to identify the frames used in the debate. Based on a list of words associated with each frame, the dictionary identifies frames in each tweet, each paragraph of the newspaper articles, and each paragraph of interest group website entries. I analyze paragraphs rather than full articles or websites because of their greater topical and evaluative coherency. Each text is classified into either including the frame (1) or not (0) depending on whether it includes at least one of the key words. Within the same text multiple frames can be present.

A dictionary method is preferable to unsupervised methods of quantitative text analysis as the goal is to trace the prevalence of known frames rather than detecting unknown framesFootnote ⁵⁴ . The full dictionary is available in the reproduction data, and Table 1 shows all frames, how they depict the legislation, and example keywords.

Table 1: Frames used in the debate on the EUCD

Note: See replication data for the full dictionary.

To analyze debates on the EUCD, the English dictionary was translated to Dutch, French, German, Italian, Polish, Portuguese, and Spanish using Google translate. While purpose-specific dictionaries perform wellFootnote ⁵⁵ and the automatic translation of dictionaries generally yields good resultsFootnote ⁵⁶ , validation is paramountFootnote ⁵⁷ . To validate the English dictionary, a student assistant hand-coded a random sample of 500 tweets and 500 newspaper paragraphs for both cases. The German dictionary was validated in the same way to assess the performance of Google translate. Intercoder agreement (Fleiss’ Kappa) ranges between 0.67 (German: 0.5) and 1.0 (German: 1.0) with less frequent frames having lower values. Figure A1-A4 display additional validity measures.

Before applying the dictionary, I converted the text to lower case and removed punctuation, stopwords, and double spaces. To detect multi-word phrases, I included tetragrams.

Time-series analysis

To further analyze how outside lobbying affected framing, I run time-series regressions. I run analyses for each language and for the full sample. The time-series starts with the first proposal of the EUCD in September 2016.

I use vectorautoregressions (VAR) to estimate the effect of platform outside lobbying on the net framing of the debate in newspapers and tweets. The dependent variable is computed as the summed relative frequency of the pro-copyright frames (Fair Pay & Creator Rights + Criminality & Public Safety + Bad Platforms & Market Power + Astroturfing) minus the summed relative frequency of anti-copyright frames (Censorship & Uploadfilter + Break the Internet + User Rights & Cybersecurity). The more negative the dependent variable, the more anti-copyright is the net framing of the debate. My key independent variables are platforms’ protest actions such as website blackouts, display of censorship banners, or a coordinated ad campaign on YouTube (see below). I also control for parliamentary events, i.e., the introduction of the directive, parliamentary hearings and debates, and the proceeding of the EUCD to the next stage. The variables are included as exogenous variables and modeled as an intervention covariateFootnote ⁵⁸ . I coded the time-period when the parliamentary event/outside lobbying occurred as one and let the effect depreciate by a factor of 0.5 for the following time-periods to model medium-term effects.Footnote ⁵⁹

I further include five variables to capture the relative framing in (1) all tweets, (2) newspapers, (3) tweets of legislators, (4) website entries and tweets of pro-copyright interest groups (i.e., rightsholders and creatives), and (5) website entries and tweets of anti-copyright interest groups (i.e., platforms and the civil society coalition). The latter encompass traditional forms of outside lobbying such as press releases, reports on (upcoming) events, or media campaigns. These variables are included as endogenous variables to reflect that the framing in one medium reacts to changes in other media and vice versa. All variables are computed separately by language except for interest group communications. For these I took the full sample, because of insufficient data in most languages and to model the pan-European character of most interest groups websites. The VAR regressions also include a dummy variable controlling for weekends. All time-series were tested negatively for a unit root.

In all models, I specified the lag structure to eliminate serial correlation and used heteroscedasticity-robust standard-errors where augmented Dickey-Fuller tests indicated heteroscedasticity. Unless otherwise noted p < 0.05 is considered significant. Full regression results can be found in tables A2-A9.

Part 1: Platforms engage in conflict expansion against the EUCD

After the Judiciary Committee of the European Parliament moved the EUCD to the plenary in June 2018, platforms and the civil society coalition tried to mobilize the public against it. Google urged press publishers to contact MEPs and express opposition, YouTube creators began criticizing the directive in their videos, tech pioneers wrote an open letter to the EU, and an online petition to “SaveTheInternet” eventually reached over 5 million signaturesFootnote ⁶⁸ . On July 4 and 5 the English, Portuguese, Spanish, Italian, Polish, Latvian, and Estonian Wikipedia pages blacked out and/or displayed protests banners against the directiveFootnote ⁶⁹ . Amid these online protests, the parliament sent the bill back for amendments. However, on September 12, 2018, it adopted the draft and initiated trilogue negotiations to mitigate disagreements between the Council, the Commission, and the Parliament. In October, YouTube CEO Susan Wojcicki urged YouTubers to cover the reform and use #saveyourinternet leading to a spike in videosFootnote ⁷⁰ . Twitch made a similar move in DecemberFootnote ⁷¹ . In November, YouTube displayed ads before videos informing users about the EUCDFootnote ⁷² . Finally, on March 21, 2019, five days before the final vote in the EP, many medium-sized platforms, among them Twitch, Reddit, and PornHub, ran protest banners on their websites (March blackouts)Footnote ⁷³ . During these protest events, platforms urged users to contact MEPs and express opposition to the EUCD. In addition, demonstrations took place in 20 countries after the March blackoutsFootnote ⁷⁴ . In Germany up to 100,000 people participated and demonstrations were substantial in Austria, Switzerland, Poland, and SwedenFootnote ⁷⁵ . Platforms and the civil society groups, thus, engaged in different forms of outside lobbying to expand the conflict around the EUCD.

Part 2.1: The effect of conflict expansion on salience and framing

Were platforms successful in conflict expansion? Conflict expansion involves increasing issue-salience, pushing frames into the debate, and mobilizing the public. Figures 2 and 3 show descriptive results on how successful conflict expansion was in newspapers and tweets across the full, multi-language sample, and across different languages in the EU.

Figure 3. Net relative frame frequency in newspaper articles and tweets on EUCD.

Figure 2A depicts the number of newspaper articles and tweets by week and highlights specific platform outside lobbying events through vertical lines. It shows that issue-salience increased substantially in European newspapers and on Twitter in anticipation and during the Wikipedia blackouts. The EP’s adoption of the EUCD’s proposal in September 2018 also led salience to spike. During YouTube’s ad campaign salience was consistently higher than before the Wikipedia blackouts. Finally, the March blackouts were followed by a huge spike in salience leading to the highest number of newspaper articles and tweets.

Turning to the by-language samples, figure 2B depicts the average number of newspaper articles or tweets per day before and between the different platform outside lobbying events (phases). It reveals substantial differences in salience between languages. Reporting on the EUCD was quite salient in German, Polish, French, Italian, and English but not in Spanish, Portuguese, and Dutch newspapers. On Twitter, the vast majority of tweets on the EUCD is German or English. Spanish, French, Polish, and Italian tweets were also somewhat frequent, but Dutch and Portuguese were not. While the overall level of salience differs across languages, salience in all languages substantially increased after platform outside lobbying events (figure 2B). Overall, platform outside lobbying successfully raised the salience of the topic across all languages but to different degrees.

The second component of conflict expansion is to influence the framing of the debate. Figure 3 depicts the net relative frequency of pro-copyright frames by language and phase. Negative values indicate that anti-copyright frames where more frequent and positive values indicate that pro-copyright frames where more frequent. Figure 4 depicts the significant results of the VAR of platform outside lobbying events on the net framing of the debate (negative effects indicate the debate became more anti-copyright).

Figure 4. Results of VAR intervention analysis on relative framing of EUCD. Note: Only languages where at least one platform intervention was significant are displayed. Full results in appendix.

In Dutch, German, and Portuguese newspapers, the debate was slightly anti-copyright even before platforms started their campaign. In English it was slightly pro-copyright, whereas in French, Italian, Polish, and Spanish it was distinctively pro-copyright. After platforms started their campaigns, the debate turned distinctively anti-copyright in Dutch and German. For German, this is also supported by the VAR results, where the March blackouts made the debate significantly more anti-copyright (figure 4). Although the debate became more negative after platform’s actions, it remained balanced in Polish and slightly pro-copyright in Italian (figure 3). In English, the Wikipedia blackout surprisingly turned the debate more pro-copyright but the YouTube ad campaign reversed this trend and after the March blackouts the debate became slightly anti-copyright. In Spanish, platform outside lobbying turned the debate distinctively anti-copyright but the March blackouts somewhat reversed this trend (figure 3).

The most interesting findings are in French and Portuguese. In French, the Wikipedia blackouts significantly increased the net frequency of pro-copyright frames (figure 4), but descriptive results indicate that the YouTube ad campaign turned the debate more anti-copyright (figure 3). However, at least in the descriptive results, after the March blackouts the debate turned substantially more pro-copyright again (figure 3). The platform blackouts in French, thus, had the opposite effect than intended. This also holds in the descriptive results for Portuguese (figure 3).

Figure 3 also reveals that the framing on Twitter was consistently more negative than in newspapers even before platforms started outside lobbying. During no phase and in no language the net framing on Twitter was pro-copyright. However, in French the debate was balanced throughout most phases (except during YouTube’s ad campaign). Surprisingly, in all languages the platform outside lobbying campaign was followed by a more positive framing of the EUCD. This is likely due to an increase in salience which led to a lower proportion of tweets by internet activists and content creators in all tweets. While in most languages the debate remained anti-copyright nevertheless, it became quite balanced in French, Polish, and Dutch. Figure 4 reveals that the YouTube ads significantly made the relative framing more anti-copyright in Spanish, Portuguese, and Italian. Contradicting the descriptive results, the March blackouts also significantly promoted anti-copyright frames relative to pro-copyright frames in Italian (figure 4). Finally, Figure 4 also reveals that the relative framing of the debate among MEPs on Twitter became significantly more anti-copyright in Portuguese after the YouTube ads started. Most importantly the March blackouts had significant but opposite effects among German and French MEPs. In German, their tweets became more anti-copyright, whereas in French MEPs’ tweets became substantially more pro-copyright.

The combined evidence from Figure 3 and 4 reveal quite some variation across languages in the effects of platform outside lobbying on the framing of the debate. French and German are at different ends of the spectrum of this variation. I therefore analyze the French and German cases in more detail in the next section.

Part 2.2: Frames and their stickiness in German and French debates

Figure 5 provides a detailed overview of the relative frequency of 12 frames in all newspaper articles and tweets in French (asterisk) and German (dot). Three differences between the languages become evident in traditional and social media: First, the censorship & uploadfilter frame that highlights the negative impact of the EUCD on freedom of expression is far more frequent in German than in French. Second, the fair pay & creators frame that highlights the positive impact of the EUCD on artists’ remuneration for their creative works is far more frequent in French than in German. Third, the lobbying, misinformation, and bad platforms & market power frames are (far) more frequent in French than in German. Why did these different frames stick in French and German? As described in the theory section, frame stickiness depends on the frame-specific perceptibility of the target audience (the bonding surface) and the characteristics of competing interest groups (the glue). In what follows, I will go through these two factors for French and German.

Figure 5. Framing of EUCD in French and German newspaper articles and tweets.

Bonding surface: perceptibility of the public to frames

I find that the differences in framing in French and German can be traced back to differences in the perceptibility of the target audiences toward specific frames. Importantly, existing debates and public opinion toward Big Tech and content regulation and the size of firms in the domestic creative sector created different bonding surfaces.

Figure 6 maps the salience of terms referring to tech scandals such as tax evasion, data breaches, or surveillance in European-, French-, and German-language newspapersFootnote ⁷⁶ . Figure 6 starts in 2013, the year of the Snowden leaks and one of Big Tech’s first major scandals. It shows that negative reporting on Big Tech remained relatively salient in most of Europe after 2013, challenging an automatic consumer-platform alliance. However, in French and German, the media narrative toward Big Tech differed substantially. In German, negative reporting steadily decreased until 2017. After a slight increase in 2018, the year platforms started outside lobbying on the EUCD, negative reporting decreased again to 33 percent of 2013 levels in 2019. In contrast, reporting in French newspapers’ stabilized at around 55 percent and began to increase in 2017. In 2018 and 2019, the salience of the techlash in French newspapers reached the same level as in 2013. I argue that since French-speaking journalists and readers already were more accustomed to negative reporting on Big Tech, frames highlighting the outsized market power and exploitative practices of platforms found a more perceptible bonding surface than in German-speaking countries.

Figure 6. Salience of techlash in newspapers.

Note: Data on Portuguese, Maltese, and Belgian French-language newspapers is only available from 2015 and not included. “Other languages” refers to English, Spanish, Polish, Italian, and Dutch.

Public opinion on the effectiveness of existing content regulation also differed critically between France and Germany. A Flash Eurobarometer from June 2018, right before the platforms’ outside lobbying campaign took off, reveals virtually no difference between French and Germans regarding their overwhelming support for the protection of freedom of expression online (Q1.3) and arrangements “to limit the spread of illegal content” (Q1.2). Thus, citizens, in principle, perceived the goals of anti-copyright and pro-copyright groups as equally legitimate. However, 56% of French respondents thought that the existing provisions of platforms to tackle illegal content were ineffective (vs. 38% of Germans; Q1.4). That a majority of the French perceived the pre-EUCD situation as insufficient suggests that they might have been more susceptible to fair pay and creator rights frames than Germans.

Finally, differences in the economic structure of the cultural economy of France and Germany provided a bonding surface for frames delegitimizing rightsholders and creatives as big business lobbyists in German, whereas they made the fair remuneration and creator rights frame more credible in France. Eurostat data on the number of persons employed per enterprise in the cultural economy in 2019 reveal that the sector is most concentrated in Germany (6.4), while France is below average (2.6)Footnote ⁷⁷ . This difference is particularly pronounced in the publishing sectors, in particular newspaper publishing (236 vs. 16.7)Footnote ⁷⁸ . Frames highlighting a market asymmetry and exploitative relation between Big Tech and small artists were thus more credible in France than in Germany. In addition, the high concentration of the creative sector in Germany created a bonding surface for frames delegitimizing rightsholders and artists as agents of big business.

In France, the existing media narrative, public opinion on content regulation, and the structure of the creative economy created a good bonding surface for the fair pay and creator rights frame and the delegitimization of the anti-copyright coalition. In Germany, in contrast, the media narrative against Big Tech was less negative, the need to reform content regulation was perceived less pressing, and the creative economy was highly concentrated. Frames, however, do not stick even to the best bonding surface if the groups pushing them apply weak glue.

Characterists of competing interest groups (the glue)

In addition to the bonding surface, the stickiness of a frame is determined by the characteristics of competing interest groups. This involves (1) the (relative) frequency with that a frame is pushed, (2) the unity of the coalition pushing it, and (3) the credibility and legitimacy of the frame. While platform outside lobbying events follow a similar pattern across French- and German-speaking countries, there are considerable differences in the composition of and the frequency and unity of the frames pushed by the anti-copyright coalition.

The outside lobbying data from interest groups websites and tweets reveal substantial differences in composition. While the number of rightsholder (FR: 27 vs. DE: 24) and anti-copyright civil society groups (12 vs. 16) that posted on the EUCD is rather similar, more interest groups representing creatives (21 vs. 9) posted in French and more platforms in German (3 vs. 6). Additionally, the composition of the pro-copyright groups in the inside lobbying data differs substantially. Reflecting the concentration of the creative economy in Germany, most German pro-copyright groups were big businesses such as Springer, Bertelsmann, Wiley, or DeGruyter (all publishers), sports leagues, television and movie firms, or the chemical industry (table A1). Representatives from German publishers were also key actors in the European-level Publishing AssociationsFootnote ⁷⁹ . In contrast, French rightsholder groups mostly represented small enterprises and individual creatives, although there were some big movie and TV associations/businesses (table A1).

The composition of the German pro-copyright coalition was used by anti-copyright groups to delegitimize their activities as big business lobbyism. Especially Springer became a focal point with demonstrations in front of their headquarters and tweets highlighting the close connection between Springer and the CDU/CSUFootnote ⁸⁰ . Pro-copyright politicians including rapporteur Axel Voss, Commissioner Günther Oettinger (both CDU/CSU), and Helga Trüpel (Greens) were criticized for their close connections to the publishers. Especially the CDU/CSU was attacked by anti-copyright activists with #niewiederCDU (#CDUneveragain)). In contrast to the French debate, were the conflict was depicted as between small creatives and Big Tech, in German the perceived conflict was between competing business lobbies which delegitimized claims of the pro-copyright coalition.

Differences between the countries are also reflected in the frequency and unity with which groups pushed specific frames. In German, the civil society and platform groups unified around the censorship & uploadfilter and the break the internet frames and pushed it very frequently (figure 7). As Figure 7 reveals, the censorship & uploadfilter frame stuck with newspapers and on Twitter. This is also consistent with public opinion of Germans regarding freedom of expression on the internet (see above).

Figure 7. Relative frequency of frames pushed by interest group type.

In French, surprisingly rightsholders are the most active group in distributing anti-copyright frames. Looking at the content of these posts, the frames are largely used to delegitimize claims of the anti-copyright coalition. For instance, rightsholders retweeted an allegation that many of the tweets containing saveyourinternet were sent from Washington, DC “the lobbying base” of Big TechFootnote ⁸¹ , urged users to not “believe anyone who says the #CopyrightDirective will censor the internet,”Footnote ⁸² or widely shared a tweet of the Palais de l’Élysée stating that “[c]opyright is not censorship, it is freedom of creation and diversity of information!”Footnote ⁸³ . This delegitimization strategy is also reflected in the attempts of rightsholders and creatives in French to push astroturfing, lobbying, and misinformation frames (figure 7). While the latter group of frames was pushed rather infrequently, they nonetheless stuck with the media, especially newspapers (figure 5).

In French, rightsholders and creatives were not only successful in delegitimizing the anti-copyright groups but also in pushing their own frames. Rightsholders and creatives unified around the fair pay & creator rights frame and pushed it frequently in French (figure 7). Coupled with a favorable bonding surface for the frame, it is unsurprising that it was successful in newspapers and on Twitter (figure 5). Although rightsholders and creatives unified around the fair pay & creator rights frame also in German (figure 7), it was not as frequently pushed, the bonding surface was less favorable than in France, and the pro-copyright coalition was delegitimized. Thus, the frame stuck considerably worse than in French media (figure 5).

To sum up this section, I find that in France the pro-copyright frames fell on fertile soil because of a skeptical position toward tech in the public and media and a creative sector characterized by individuals and SMEs made for a good bonding surface. Further, it enabled the pro-copyright coalition to delegitimize the anti-copyright coalition. In Germany, a less tech critical media debate, the belief that pre-EUCD regulation is sufficient, and the concentration of firms in the creative sector, notably of publishers, created a bonding surface for censorship frames and the delegitimization of the pro-copyright coalition. In the next part, I will trace how MEPs responded to the signals sent by these different frames.

Outcome: policymakers respond to signaling

Cross-language differences in the framing of the debates are largely reflected in the voting behavior of MEPs on the EUCD on March 26, 2019. Three groups of countries can be distinguished. First, countries where strong support for the directive transcends party lines. More than 80 percent of French MEPs voted in favor and significantly tweeted more pro-copyright after the March blackouts (figures 4&8). In the parliamentary debate, they mirrored the delegitimization frames toward the anti-copyright coalition as did the CDU/CSU MEPs and the commission. They disregarded demonstrations and concerns of the public as the result of a misinformation campaignFootnote ⁸⁴ , alleged platforms to buy demonstratorsFootnote ⁸⁵ , or to “have bombarded [MEPs] like vulgar cultural intellectual political prisoners”Footnote ⁸⁶ . In addition to French MEPs, also Belgian and Portuguese MEPs voted largely in favor of the EUCD (figure 8) reflecting the framing of the debate in their languages. In Spanish, where the framing was mixed and salience was low, MEPs also largely voted in favor of the EUCD. This suggests that MEPs did not respond to the public amid low politicization of the issue. Second, in countries where framing in newspapers was mixed and salience was high (Italian, English/UK), MEPs votes differed along partisan lines. Third, MEPs from Poland, the Netherlands, Austria, and Germany largely voted against the reform (figure 8). In all these countries, except for Poland, the framing of the debate was strongly anti-copyright after platforms’ outside lobbying. Among German-tweeting MEPs, the March 2019 blackouts significantly increased the relative frequency of anti-copyright frames (figure 4), and EUCD opponents referred to public protests to justify their positionFootnote ⁸⁷ . In Germany, legislators from the CDU/CSU are the exception as they largely voted in favor of the reform justifying their votes by delegitimizing the anti-copyright coalition (see above and figure 8)Footnote ⁸⁸ .

Figure 8. Percentage of MEP votes in favor of the EUCD by Country and Party Group.

Source: Own compilation based on data provided by pledge2019.eu.

Notes: Black vertical line indicates average by party group.

Still, the alignment of MEPs voting behavior with the framing of the EUCD in their home countries becomes evident, showing that MEPs are generally responsive to the signal of media reporting.