Electoral news sharing: a study of changes in news coverage and Facebook sharing behaviour during the 2018 Mexican elections

ABSTRACT Patterns of news consumption are changing drastically. Citizens increasingly rely on social media such as Facebook to read and share political news. With the power of these platforms to expose citizens to political information, the implications for democracy are profound, making understanding what is shared during elections a priority on the research agenda. Nevertheless, to the best of our knowledge, no study has yet explicitly explored how elections transform news sharing behaviour on Facebook. This study begins to remedy this by (a) investigating changes in news coverage and news sharing behaviour on Facebook by comparing election and routine periods, and by (b) addressing the ‘news gap’ between preferences of journalists and news consumers on social media. Employing a novel data set of news articles (N = 83,054) in Mexico, findings show that during periods of heightened political activity, both the publication and dissemination of political news increases, the gap between the news choices of journalists and consumers narrows, and that news sharing resembles a zero-sum game, with increased political news sharing leading to a decrease in the sharing of other news.


Introduction
News sharing on social network sites (SNS) has risen to a position of prominence in our understanding of digital news. With platforms such as Facebook allowing users to instantly share news articles to large, personal audiences, citizens increasingly rely on SNS to find and engage with news, while media organizations simultaneously make use of news sharing to reach larger publics. Moreover, as a key source of political news and information (Nelson & Webster, 2017), news sharing does not only raise questions for the study of journalism, but also poses implications for democracy. A growing field has therefore explored what drives news sharing, analyzing features of the content shared, and characteristics of users that share news (Orellana-Rodriguez & Keane, 2018). Much less has been said about the effect particular contexts have on news sharing.
We know from a rich field of agenda-setting that changes in context can have an impact on journalists' coverage of, and citizens' attention to, news topics (Kepplinger & Habermeier, 1995;Kepplinger & Lemke, 2016). This is also the case with elections (Tewksbury, 2006): Scholars have studied elections, finding drastic changes in both the coverage of political news, and in the amount of political information consumed by citizens. There is therefore reason to believe that such moments of heightened political competition could impact news sharing. So far, research has used news sharing to understand the development of specific political periods, from the Arab Spring (de Fatima Oliveira, 2012) to Danish elections (Ørmen, 2019); nevertheless, there have been little-to-no efforts to explicitly explore how these events change news sharing behaviour away from routine moments.
Elections have not only been shown to have an impact on journalists' production of news and consumers' engagement, but also on the divergence in these respective groups' preferences. By building on the notion of the 'news gap'suggesting that news consumers seem to be less interested in reading political news than editors are in publishing it )we not only address how news sharing changes during electoral periods, but also whether significant variance exists between journalists' preferences and news consumers' choices. We take a comparative approach to examine the effects that the presence of an election campaign has on the publication and consumption of political and non-political news. We focus on the understudied case of Mexico to examine: To what extent is political and non-political news content published and shared on Facebook during campaign periods versus routine periods?.
We combined Facebook sharing data with an original data set of news items from an election (March-July 2018) and a routine (March-July 2019) period (N = 83,054) and used automated content analysis to analyze the data.
This study provides various contributions. First, we research how an election can change the supply side of the news ecosystem. Second, we address the demand side of news ecosystems, evaluating how audience engagement on social media drastically shifts between electoral and non-electoral periods. Third, we analyze how elections shape the rift between journalists' and news consumers' preference for political news in the 'news gap'. Overall, by comparing election and routine periods, the findings contribute to the conversation on the generalizability of communication research taking place in isolated settings (Kepplinger & Habermeier, 1995;Kepplinger & Lemke, 2016), providing evidence for strong, contextual effects on news engagement on social media.

Theoretical background and related research
2.1. Political news production: election vs. routine periods There are numerous reasons why political news is different during election periods (Druckman, 2005;Strömbäck, 2005). Zaller (2003) proposed that increased coverage of political news during election periods is an example of media adhering to what he calls 'the Burglar Alarm standard', according to which journalists 'call attention to matters requiring urgent attention, and … do so in excited and noisy tones ' (p. 122). This is reflected in more recent empirical work. Van Aelst and De Swert (2009) show that media coverage of political news differs substantially between election and routine periods, with an upcoming election generating election news, boosting political coverage, and reducing soft and sensational news. Vliegenthart et al. (2011), analyzing news in the United Kingdom and the Netherlands from 1990/1991 to 2007, found a stronger primacy for political parties during election periods compared to routine times. This is supported by agendasetting work, which suggests that key events, such as disasters or elections, drive coverage away from other topics and towards these events (Kepplinger & Habermeier, 1995).
Other scholars have argued that today's media environment has led to a permanent campaign, suggesting blurred lines between election and routine periods (see e.g., Larsson, 2016;Ornstein & Mann, 2000;Vergeer et al., 2013). Yet, recent empirical findings have predominantly indicated profound differences between election and routine periods in politicians' usage of Twitter (Vasko & Trilling, 2019) and Facebook (Ceccobelli, 2018).
Few studies specifically address both election and routine periods to examine online political news coverage. Here, we explore political news on Facebook by focusing on the context of Mexico. We pose the following first research question:

RQ1:
To what extent do election periods increase the amount of political news being published online?

Political news sharing: election vs. routine periods
Voters increasingly rely on SNS for news and information about politics (Nelson & Webster, 2017), accessing news by following links on SNS, and sharing it to their own networks. News sharing on social media can be explained by three groups of features (Orellana-Rodriguez & Keane, 2018): user (e.g., demographics), content (e.g., article topic), and context (temporal and spatial aspects). Regarding user features, the number of 'friends' or 'followers' (Bakshy et al., 2011), activity on social media platforms (Choudhury et al., 2010), and news consumption preferences (Hermida et al., 2012), among others, have been shown to affect news sharing. When studying content features, previous research has taken a news-value-based approach, contending for structural characteristics making some stories more 'shareworthy' (Karnowski et al., 2021;Trilling et al., 2017). Literature following this tradition has focused on characteristics such as article topic and article frames. Regarding topics, research has shown that audiences are more likely to share non-political articles, such as on lifestyle (Trilling et al., 2017). Despite this lack of interest in politics, research has shown that readers seem to be less interested in sharing political news than editors are in publishing it (Bright, 2016). Studies have journalists and social media editors are aware of these considerations, with journalists recognizing the higher engagement entertainment news receives (Lischka, 2018). Research on the role of frames is less conclusive. For instance, Trilling et al. (2017) and García-Perdomo et al. (2018) find that a human-interest frame and conflict frames increase sharing. In contrast, Valenzuela et al. (2017) find no effect for human-interest frames and even a negative effect for conflict frames.
The literature on context effects on news sharing is sparse, especially as pertaining to elections. While previous research has looked at individuals' news consumption habits during elections (Ørmen, 2018) and how news content itself changes within an electoral context, these studies fall short of addressing how these sharing patterns diverge from non-electoral routine periods. Recent work has shown how unexpected events lead to spikes in news sharing (Salgado & Bobba, 2019), suggesting that changes in context are indeed tied to changes in news sharing. Moreover, recently, Vasko and Trilling (2019) analyzed 285,456 tweets by Members of Congress during and after the 2016 U.S. elections. The results indicate that, during a routine period, politicians tweet more about hard news, compared to the campaign period or the lame duck period. There is however a need to address whether and how these identified relationships on news sharing are product of actual consumer and publisher behaviour, or can be attributed to the particular context being investigated.
Citizens' involvement with politics fluctuates in conjunction with election cycles. Numerous models of democracy (Ferree et al., 2002;Strömbäck, 2005) concede that the involvement of ordinary citizens is not continuous but often limited to participation in elections (or, maybe, in specific protests). The term 'monitorial citizen' was first coined in the 1990s by Schudson (1998). He advocates for a model in which, rather than trying to follow everything, citizens monitor politics for events that require responses. Citizens only become active once the media ring the 'Burglar Alarm' (Zaller, 2003). Continuous involvement on a high level, he argues, would be an unrealistic expectation.
It is not our aim to make any normative claims about what role we believe citizens should or should not have, but there is some evidence that citizens indeed behave differently during election periods. Neudert et al. (2019) found that European elections (i.e., in France, Germany, and the United Kingdom, in 2017), generate large amounts of political news coverage on Twitter. Nevertheless, this comparative hypothesis has yet to be addressed when it comes to online engagement with news. Thus, if there is more political content available (a reasonable assumption, given the heightened political activity) and citizens are more motivated to interact with it, then we would expect the sharing of political news on social media to spike in an election context. Because of the assumptions laid out by the 'monitorial citizen' approach to the understanding fluctuation of citizen engagement, we pose the following research question: RQ2: To what extent do election periods increase the amount of political news being shared on Facebook?

Closing the news gap?
We also aim to understand the connection between the production and the dissemination of political news. Ten years ago,  introduced the 'news gap'the idea that readers seem to be less interested in reading political news than editors are in publishing it (Boczkowski & Mitchelstein, 2013). They examined journalists' and citizens' news choices in eleven online newspapers from six countries in Western Europe, Latin America, and North America, including Mexico. The results indicate a major gap: journalists selected considerably more hard news as the most newsworthy stories than their audiences. While their measure of audience interest is based on clicking behaviour, others have argued that clicking on news is not equivalent to newsworthiness or interest (Costera Meijer & Groot Kormelink, 2015). We believe that while this limitation does exist, click activity and article consumption at the aggregate level do contain some signal of audience interest. Today, consumers are not only exposed to news, but also share them (Kümpel et al., 2015), pushing research to highlight the existence of a 'social news gap' between news reading and news sharing behaviour (Bright, 2016). Here, news shared on SNS is different from news consumed directly through online platforms. These changes have reframed the 'gap' for media organizations and society at large.
As Trilling et al. (2017) have argued, the concept of 'newsworthiness' can be extended to a concept of 'shareworthiness'. Based on a large-scale analysis of the sharing of Dutch news articles on Facebook, they find evidence that traditional criteria of newsworthiness play a role in predicting the number of shares. But they also argue that newsworthiness and shareworthiness are not identical -'one needs to extend and modify' (p. 45) the former and the 'relative importance' (p. 45) of news values such as distance, negativity, positivity, conflict, human interest, and exclusiveness may differ between newsworthiness as perceived by journalists and shareworthiness as perceived by the audience. Moreover, their analysis suggests that news items covering social affairs, as well as culture, and entertainment, are shared more often than others, including political news (Trilling et al., 2017). Here, again, we can see signs of a potential gap. One of the possible explanations for this behaviour lies in people's desire to avoid sharing news that may potentially be controversial (Valenzuela et al., 2017). Moreover, work interviewing social media news editors suggests that even though these actors are aware of the audience's preference for entertainment news, they still strive to provide a balanced diet of soft and hard news (Lischka, 2018). Boczkowski et al. (2012) also indicate that the 'news gap' changes during periods of heightened political activity, such as elections. The results indicate that the gap between journalists' and consumers' preferences is smaller during an election period. Work on divergence between news publication and sharing has been contradictory, with some finding minimal evidence of a gap (Martin & Dwyer, 2019), and others suggesting it does manifests itself on social media (Bright, 2016). While a comparative approach has been taken with regard to the 'news gap' at large, the same cannot be said for the 'social news gap'. For example, Bright (2016) did find evidence of a divergence in the reading and sharing of political news, however, the study was not comparative in nature. With previous work arguing that the 'news gap' shrinks during a period of heightened political activity, we pose the following hypothesis: H1: The 'news gap' between the production and the dissemination of political news on Facebook diminishes during election periods.

More of politics, less of everything else?
Our final research question concerns whether the sharing of political news comes at the expense of sharing non-political news content, or whether it has no meaningful impact. It is likely that news sharing is essentially a zero-sum game: The amount of articles that citizens share may be fixed; and if they share more articles on politics, they inevitable cut back on their sharing of other topics. Two strands of literature support this argument. First, research on media use suggests that people have a fixed time budget allocated for media use (Ha & Fang, 2012). Using more time to browse the internet, hence, would lead to less time available to, for instance, watch television or read a book. Dimmick et al. (2004) argue that the internet indeed has such displacement effects and that these can be explained by the theory of the niche: if there is a large overlap between the gratification opportunities that two media offer, the one that is perceived as superior will displace the other, unless it finds a different niche.
Applied to our topic of investigation, this could indicate that the time citizens allocate to sharing links will remain constant, and if they already got their gratifications (e.g., social recognition, relaxation) from sharing political news, this will be at the expense of sharing other content.
RQ3: Does news sharing resemble a zero-sum game (where an increase in sharing of political news will decrease that of non-political news) or a cumulative practice (where an increase in shared political news does not have a meaningful impact on other news sharing)?

Case
We analyze political news coverage and sharing behaviour on Facebook during the 2018 Mexican elections and compare it to a routine period a year later. Reports on Mexican Facebook use to place it as the country with the fifth most Facebook users in the world, and that over 63% of users report to sharing news regularly (Newman et al., 2019). The 2018 Mexican elections have also been widely recognized as witnessing unprecedented levels of online engagement (Glowacki et al., 2018), something that is especially true on Facebook (de León & Trilling, in press). Mexican electoral laws also establish a clear-cut four-month campaign period for the presidential elections that allows us to easily identify the start and end point of the official electoral period.
On election day (July 1st, 2018) Mexicans not only voted for the presidency, but for hundreds of other representatives, making it the largest democratic exercise in the country's history (Greene & Sánchez-Talanquer, 2018). The election featured a third-time bid by left-wing Andŕes Manuel Ĺopez Obrador (AMLO), who brought the issues of widespread poverty, rampant corruption, and devastating violence to the forefront of the campaign. While the campaign did feature high degrees of polarization among the electorate, AMLO led by a comfortable margin throughout, with the anti-AMLO vote being split by two competing establishment candidates (Garrido & Freidenberg, 2020). This resulted in the least contested election in decades, with a clear and overwhelming victory for AMLO and his MORENA party. While the focus of this study is not on the particularities of this political process, we provide these details in order to contextualize our work and to provide boundary conditions for our work.

Sample
The sample consists of a novel data set of news articles published throughout the entirety of the electoral campaign period (March-July 2018), and a four-month reference period a year later (March-July 2019) from five leading outlets in Mexico (N = 83, 054), El Universal, Milenio, Excelsior, Proceso, and El Financiero. None of these outlets are 'digital born', all being online versions of established, quality print papers, with decades of existence and a country-wide reach. With the exception of Proceso, which is known for its left-wing slant, these outlets are not characterised by ideologically-driven coverage. These outlets are all active on Facebook, but have diverging number of followers on their pages: while Milenio, Excelsior, and El Financiero had around 1.5 million Likes each in 2019, El Universal and Proceso had above 4 million. Following de León and Trilling (in press) articles were collected using a web-scraper that makes use of 'Archive.org'. To be included in the sample, the news outlet had to be featured among the leading online newspapers in the country (Newman et al., 2019) with at least one daily snapshot on Archive.org for both the election campaign and routine period.

News topic classification
We classified articles into six distinct topic categories: news on (1) Politics, (2) Crime and Disasters, (3) Culture and Entertainment, (4) Economic and Business, (5) Sports, and (6) Other (news on technology, religion, the environment, and all other articles that did not fit into previous categories). We used supervised machine learning to classify news articles into these categories (Trilling et al., 2017). For this purpose, a random stratified sample of 2,000 articles were manually coded. Two coders were trained and tasked with manually annotating these articles. 140 randomly selected articles were coded by both, allowing for the calculation of an intercoder reliability (Krippendorff's α) of 0.79. This sample was then split into training material (80%) and testing material (20%). To train a supervised machine learning classifier to distinguish between these six topics, a pipeline was established to test three distinct text-preprocesing steps, six different classification algorithms, and varying hyperparameter combinations, resulting in 18 different classifiers (Appendix 1). We tested each of these algorithms on the unseen testing data, reporting the precision, recall, and f1-score for each (Appendix 2). Based on these results, we identified the Support Vector Machine using full texts (no pre-processing) as the bestperforming classifier, meeting standard performance benchmarks, with precision and recall scores for each topic >0.75. The algorithm was then used to classify the full sample of news articles.

News publication and sharing
We operationalized editorial preference for specific topics to be a simple count of the number of articles published by topicsomething only possible with large inclusive samples of the news outlet. If journalists do prioritize political news, this should be reflected in the quantity of political news published. We gathered Facebook data from CrowdTangle, a SNS monitoring service, linking each article to the number of Facebook shares received. Specifically, the querying returned information for each 'public' post that included the respective article link, and information on how many times the post itself was interacted with by private accounts. Similar to comparable studies, we cannot distinguish the reason behind a share: some shares may be generated by clicking on a button on the news site itself, others by re-sharing content from someone's timeline or from a group. We will re-visit this aspect in the discussion.

Analytical strategy
We take two approaches to analyze our data. We take a descriptive approach to answer RQ1 and H1. News articles are aggregated by period and topic, followed by a calculation of the relative share by topic. Evaluating fluctuations in the share of news articles published by topic allows us to establish changes in patterns of publication, quantifying the heightened attention to political news during the campaign period.
Second, we evaluate changes in news sharing. Here, we use classified characteristics to empirically estimate the number of shares each article will receive. This mirrors the approach that Trilling et al. (2017) took in their aforementioned study on 'shareworthiness,' where they predicted the number of shares using negative binomial regression models. These models allow us to account for the count distribution of the sharing data, where a vast majority of articles receive 0 shares (Figure 1), and control for the influence of other variables, such as the presence of these news sites on Facebook, as the outlet has been shown to be one of the major drivers of sharing (Karnowski et al., 2021;Trilling et al., 2017). We then compare both periods. Table 1 displays summary statistics of the final data set by showing both the share of articles by topic for each period, as well as the aggregate sharing behaviour that each topic received, allowing for some immediate observations. First, there was a shift in the overall number of articles published (+8,000 in the election period). News about crime, culture, economy, and other topics only changed by a couple of hundred articles-the vast majority of the increase thus lies in news about politics (+4789) and sports (+1636). By looking at the relative share of articles by topic, political news increased from 32% to 37% of the articles being published. These results indicate that during election times, journalists are more interested in covering the political grapevine (RQ1).  Table 1 also allows us to address RQ2, examining sharing behaviour during election periods. A first observation here is the shift in total shares received: Articles in the Routine (2019) period received over four million more shares than those in the Election (2018) period, even though less articles were published in the former compared to the later. There also is a drastic shift in the extent to which political news is shared. While for both periods, political news received the greatest percent of shares, there is a drastic change in the proportions received: it comprised 55% of all shares during elections, and decreased to 34% in the routine period. Furthermore, while the total shares for all topics almost doubled from the electoral to routine period, the total shares of political news dropped from 4.7 million to 4.3 million. Evidence for disengagement with political news can also be found in the mean number of shares received by each topic: in the electoral period, political news held the highest mean of sharesin the comparison period, however, the mean of shared political news was lower than news about crime, other topics, and almost equal to culture news. We, therefore, see a drastic increase in the amount of political news being shared on Facebook during elections.

Changes in online publication behaviour
We now address H1 on whether electoral periods lead to a closing of the 'news gap' between the publishing and sharing of political news online. To do so, we must first establish the existence of a 'news gap'. During the routine period, political news received the largest amount of engagement both in terms of journalistic coverage as regards total articles written on the subject, as well as the largest number of total shares. While this might be interpreted as evidence against the existence of a news gap, the number of total shares could be the product of the very fact that there are more political articles produced about politics. We, therefore, turn to metrics that account for this: the mean and median number of shares received by each topic. For politics, both the mean (355.74) and median (11) number of shares are below the equivalent matrices for crimes (419.36 and  Figure 2 in Trilling et al., 2017).

23)
, and other (555.57 and 13). Therefore, we find evidence of a gap during the routine periodin total, journalists preferred to write about politics, while on average, Facebook users preferred other topics. Does this gap close during elections? As discussed previously, the electoral period witnessed large shifts in political sharing behaviour, with political news sharing increasing by 21.5 points: from making up 33.9% of news shared during the routine period to 55.4% during elections. Turning to the mean and median of shares by topic, we observe an increase in the sharing of political news during elections relative to other topics in the same period. The median sharing of political news (11) was below other (13) and crime (23) news during the routine period. During the elections, the median sharing of these topics are still ahead of politics, but the gap has significantly reduced: politics, with a median of 0 shares, is shared on average almost as much as other (1) and crimes (1). During the elections, political articles had the highest mean sharing of all topics, while in the routine period this mean was below other and crimes. Therefore, looking at both average measures of sharing shows a closing in the 'news gap'. Table 2 displays the results of three negative binomial models predicting the relationship between news topics (politics as reference category) and sharing statistics, for both the election and routine periods. The results are in the form of Incidence Rate Ratios (IRRs): for every one unit increase in the independent variable, the expected value of the dependent variable, shares, is obtained by multiplying by the IRR. Specifically, the IRRs 'represent the change in the dependent variable in terms of a percentage increase or decrease, with the precise percentage determined by the amount the IRR is either above or below 1' (Piza, 2012, p. 3). Therefore, positive effects are those above 1, while negative effects are below 1. Since our variables of interest are binaryan IRR of 1.28 leads to a 28% increase in the number of expected shares (thus, 128% of the original value), while an IRR of 0.90 leads to 90% of expected shares (a 10% decrease).

Changes in electoral sharing behaviour
Model 2 on the routine period shows that news on crime, culture, and other are shared significantly more than news on politics, receiving 31%, 17% and 138% (p < .001) more shares than a news article on politics, respectively. In Model 1, on the electoral period, news on crime and culture are not different from politics at a statistically significant level. News on other topics receive 53% more shares than political news (p < .001). The results show that going from the routine to electoral period, all news topics are shared less often relative to political newsonly sports increased in shares.
Model 3 evaluates whether the shifts between models are statistically significant by using a pooled model where each topic is interacted with the electoral period. When switching from routine to electoral period, the performance of each news topic relative to political news worsens significantly: the IRR for crime news drops by 0.514 (p < .001), for culture by 0.605 (p < .001), for economy by 0.635 (p < .001), and for other news by 0.415 (p < .001). These results provide evidence that sharing political news increases significantly in the run-up to elections. These results confirm H1, indicating that during electoral periods, there is a large and significant spike in how Facebook users share political news, arguably closing the gap between journalists and citizens' interest in political news.

Zero sum or cumulative
Our evidence suggests that news sharing resembles a zero-sum game. Table 2 shows that in the electoral period, political news sharing increased relative to all other news topics (except sports). This is a relationship that is yielded statistically significant by the interaction effects in Model 3. If the relationship was cumulative, we would not expect such strong decreases in the sharing of non-political topics relative to political ones, but rather for them to remain stable since increased political news sharing would not come at a cost to the sharing of other news. This is a relationship that can be more intuitively observed in Figure 2.

Conclusion and discussion
This study seeks to understand how elections shifted the Mexican national news environment by evaluating changes in the publication and sharing of political news on Facebook. The results reveal four main findings. First, journalists' interest in political news spikes during elections, while the coverage of other news remains stable. Second, there is a dramatic growth in political news sharing during elections. Third, we find support for the notion that the 'news gap' between the public's and the media's interest in political news is significantly reduced during elections. Fourth, the increase in sharing political news during elections has a detrimental effect on the sharing of other news types, suggesting that news sharing resembles a zero-sum game.
In this study, we asked how elections change the publication of political news (RQ1). Our results provide evidence for the notion that journalists are more prone to cover politics during periods of heightened political activity (Van Aelst & De Swert, 2009), because there is more going on (Druckman, 2005), and because political stakes are higher (Strömbäck, 2005;Zaller, 2003). During the 2018 Mexican electoral period, there was a notable increase in the amount of news articles produced about politics in comparison to the same period a year later. Nevertheless, this increase did not come at the cost of other news production, as suggested by Van Aelst and De Swert (2009)-the amount of news produced for all other topics barely deviated from one year to the next.
We also ask how elections change political news sharing habits (RQ2). We provide evidence that the presence of elections significantly increases the number of political news being shared, which is in line with theories on political engagement fluctuation during elections (Zaller, 2003). We theorize that this dramatic increase is fueled by two complementary processes. First, it's a result of citizens engaging with the crucial political conversation dominating the news. The second process behind this drastic change in news sharing is simply how un-popular political news is during routine periods. People's general hesitation to share political news has been discussed in detail within numerous other studies (e.g., Trilling et al., 2017), with findings pointing to the fact that, because of the risk of inciting controversy, political news is not as shared as other less controversial topics, especially on 'strong-tie' social networks such as Facebook (Valenzuela et al., 2018). Our results suggest that elections mitigate at least part of this controversy-avoidance, making political news a topic shared at least as much as Entertainment and Culture.
We also build on the notion of the 'news gap' , hypothesizing that the news gap would diminish during election periods (H1). We find a large difference between the publication and the sharing of political news during routine periods, in contrast to Martin and Dwyer (2019), but in line with Bright (2016) and . This supports the idea that a divergence exists between what news publishers choose to publish, and, in this case, what audiences share across social media. While a trend that might be worrying for democratic processes, fears of audience disengagement are somewhat quelled when looking at the electoral period. Similar to , we find that this 'news gap' is significantly reduced during elections: while journalists' interest in political news increases, it is far outpaced by the increase in political news sharing behaviour.
Lastly, in this study we ask whether increases in the sharing of political news during elections comes at the cost of the sharing of other types of news (RQ3). We find preliminary evidence that it did. This interpretation has its theoretical underpinnings in studies positing that individuals have a limited time budget they are willing to allocate to media (Ha & Fang, 2012). Moreover, agenda setting theories and case studies on disaster response lay out similar expectations at the aggregate level, where attention to specific agendas and topics come at the direct expense of others (Bright & Bagley, 2017;Jonkman et al., 2018;Zhu, 1992). Our results suggest that similar mechanisms are at play for the sharing of political news during electionsheightened dissemination of political news by individuals will make them more likely to not share other types of news, simply because politics is what is occupying their limited attention, leaving less space for other news items.
Despite the contributions, the paper has certain limitations that should be addressed in future research. The first regards our Facebook sharing data: our study does not account for concerted efforts to boost specific content, either through automation or paid workers. As the study focuses on an electoral period, where political actors might benefit from the proliferation of particular news stories, it is not inconceivable that such behaviour took place. On the other hand, Theocharis and Jungherr (2021) highlight that the fear of so-called bots is generally over-stated. Our study cannot disentangle where the shares originate. Most likely, some shares originate from publisher's websites, others are re-shared links from somebody's Facebook friends, Facebook groups, or otherwise suggested content on someone's timeline. This means that a holistic interpretation of what drives news sharing needs to take into account not only the user's agency, but also the role of network ties and Facebook's recommendation algorithms. In particular, as Lischka and Garz (2021) pointed out, these affordances are not stable over time. The major change in the Facebook algorithm that they describe is outside of the time period we study, but also dos Santos et al. (2019) point out that (unknown) algorithm tweaks make estimates of the influence of specific features on news sharing unstable over time. Hence, such changes in the platform's affordances can offer an unobserved alternative explanations that can atleast partly explain our findings. One may speculate, for instance, that in 2019, a reconfiguration of the affordances may have contributed to the higher amount of shares compared to 2018.
Our findings have to be interpreted in this light, and are contingent on the current setup of the Facebook ecosystem. This limitation is hard-to-impossible to overcome, but complimentary research, such as qualitative studies in which users are observed in a natural setting over an extended period of time, may help contextualize our findings. Second, even if sharing, in general, is done by ordinary humans Facebook's algorithms have some agency here: they influence which news items are distributed and prioritize showing certain content.
Finally, while we do believe that a simple aggregated count of all articles published by topic is an indicator of the attention and importance attributed to a specific news topic, this method diverges from Bright (2016), who use a combination of article placement on the front page of news site, and the amount of time the article spent on the front page to calculate journalistic priority.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Notes on contributors
Ernesto de León is a Doctoral Student at the Institute for Communication and Media Science, University of Bern (Switzerland). He is interested in questions of political news consumption and its effect on political identities and behaviour, as well as the role that social network sites play in an electorate's interaction with news and its implications for democracies. He is also interested in exploring the benefits computational methodologies can offer the study of political communication online. [e-mail: ernesto.deleon@ikmb.unibe.ch].
Susan Vermeer is a postdoctoral researcher at the Amsterdam School of Communication Research (ASCoR) at the University of Amsterdam. Her research particularly focuses on news consumption in a world of news sites, algorithms, and social media. [e-mail: s.a.m.vermeer@uva.nl].
Damian Trilling is Associate Professor for Communication in the Digital Society at the Department for Communication Science, University of Amsterdam. He is especially interested in how the changing media environment changes how people engage with news and current affairs. [email: d.c.trilling@uva.nl].