Policy signals in party communication: explaining positional concreteness in parties’ Facebook posts

Abstract Policy signals are often conceived of as positions on an ideological scale. However, apart from the position – considered here as the policy objective – the policy instrument and the concreteness of the instrument must also be taken into consideration. In the article, a new conceptualisation of policy signals is developed, which integrates policy objectives, policy instruments and how concrete these are. Drawing on issue competition research, a set of expectations is advanced about the importance of actors’ control over outcomes for positional concreteness. Then, policy signals are looked at in the unmediated context of Danish parties’ Facebook posts ahead of the 2019 national election. Based on all textual and audio-visual posts in the year before the election, it is found that the levels of positional concreteness are generally high. Yet – in line with expectations – positional concreteness depends on parties’ incumbency status and the policy field.

protection, more-less European integration, and so on. However, a policy signal is more than a position on a scale. Apart from the position -what we might consider the policy objective of the party -policy signals vary depending on how specific they are. The specificity of a policy signal is important because a vague, imprecise statement carries a lot less informational weight than an exact statement does. Compare the two statements: 'I want to reduce immigration from Mexico to an absolute minimum' and 'I want to reduce immigration from Mexico to an absolute minimum -and to do so, I will build a wall on our southern border and have Mexico pay for it' . Both statements contain the exact same objective, but the latter appears much more committed because it details how the party or candidate will implement the position. In short, the key issue is whether a policy objective is combined with a policy instrument and how concrete this instrument is.
Another problem with analysing policy signals is the data sources themselves. Election manifestos, as used by the Manifesto Project (Volkens et al. 2019), which has coded election manifestos from up to 56 countries back in time using the same coding scheme, are important for the parties that make them but rarely for the voters, who cannot be expected to be familiar with them. In the literature, it is assumed instead that the information in the manifestos filters down via mass or social media to the voters, but whether this actually happens is uncertain (Adams et al. 2011). Expert surveys, such as the ones employed by the Chapel Hill project, make it even harder to relate to any specific stage in the policy signalling process because experts are likely to base their assessment on a mix of what parties say they want to do and what they actually have done in the past (Bakker et al. 2015;Netjes and Binnema 2007).
Our contribution is three-fold. First, we develop a conceptualisation of policy signals that integrates policy objectives, policy instruments, and how concrete these are. Second, drawing on a literature on issue competition (Green-Pedersen and Mortensen 2010; Klingemann et al. 1994;Thesen 2013), we advance a set of expectations about the importance of actors' control over outcomes for positional concreteness. Third, we code all relevant Facebook posts of Danish parties the year before the 2019 general election (N = 3,992). Facebook is excellent for this purpose because parties use it to communicate with the wider public, which sets it apart from parties' election manifestos. At the same time, Facebook is unmediated, meaning that we obtain a measure of what the parties wanted to signal, not what the news media found interesting to report on.
Our findings indicate that parties engage in a substantial amount of policy signalling and do so in a way that fits our argument. Far from limiting themselves to emotional appeals or the mere promotion of abstract goals, parties outline which instruments they want to use to achieve the goals in half of the cases and clearly specify these instruments in around a fifth of the cases. As expected, an important factor to understand the variation in policy signals is the potential control over outcomes that actors might have. We find much higher rates of positional concreteness for (a) government parties and (b) policy fields for which the national government can implement policy instruments to pursue their policy objectives (e.g. health politics or education in comparison to international relations or civil rights).
The results suggest the validity of the theories of democratic politics assuming that policy signalling takes place. Parties signal their policy positions to their voters, and the public takes notice in a meaningful way. More broadly, our findings speak to the often heard concern in public discourse that politics in modern-day democracies is all about form rather than substance. Despite increased focus on individual candidates (Bøggild and Pedersen 2018;Fridkin and Kenney 2011), negative campaigning (Lau and Rovner 2009;Walter et al. 2014), and plain incivility (Skytte 2021), parties nevertheless devote a considerable amount of their communication to policy signals with substantive content of value to the electorate.

Conceptualising policy signals
Policy signals, as conceived of in the literature from Downs (1957) onwards, entails that parties as a minimum communicate a policy objective. Important policy objectives found among modern-day parties include less poverty, a worthier life for the old, protecting the national culture, sustaining the environment, ensuring economic growth, fewer immigrants in the country, more cancer survivors, a better balance between the cities and the countryside -just to mention a few. More abstractly, party communication includes a policy objective if the communication specifies an aim or ambition of changing or protecting the social or natural world.
The presence of a policy objective is arguably what makes a piece of communication political in the first place. Yet, a policy objective is often not enough for citizens to base their vote choice on. Many policy objectives are valence issues; that is, outcomes few would disagree with. The core question is often not what a party wants to achieve but how they are going to do it. Will climate change be fought via gentle nudging of companies or heavy consumer taxes on long distance holidays? Will low unemployment be achieved by expanding the public sector or lowering unemployment benefits to make low-paid 'junk' jobs more attractive? This applies even for policy objectives that are not valence issues. The number of new immigrants is a positional issue in many countries, but the way to handle this presumably still matters. If a voter prefers fewer immigrants, (s)he often still needs to choose between those parties wanting to achieve this by spending more on foreign aid and those wanting more border control and other high-powered deterrents. On the other hand, if a voter favours a more open approach to immigration, the question remains what this implies more precisely. Some parties favour a green card arrangement where the qualifications of immigrants weigh heavily, others that the social needs of immigrants and their families are pivotal.
The consequence is that we need to broaden our understanding of policy signals to include the policy instrument by which the party wants to achieve the policy objective. Drawing on Vedung (1998), a policy instrument is defined as an action by a public authority at the supranational, national, regional or local level. A policy instrument may be either regulatory (e.g. a ban on smoking or higher speed limits), economic (e.g. more government spending on health care or larger fines for littering), or informational (e.g. an awareness campaign against hate crimes).
In a second step, we also need to consider how concrete the policy instrument is. Often (as we will see), parties claim to want 'more' or 'less' of a policy instrument, leaving the volume unspecified. Yet, for voters, it obviously matters if speed limits are to be reduced with 10 km/h or 40 km/h or if income taxes are to increase with one percent or 10 percent. A voter may agree with a policy objective of reducing plastic waste as well as the policy instrument of introducing a new tax on plastic bags -but, at the end of the day, find the proposal unappealing when learning that the new tax will be two euros per bag in the shopping mall.
In sum, the value of policy signals depends on three elements. First, is there a policy objective at all? Second, is there a policy instrument specified of how the policy objective will be achieved? Third, is the proposed policy instrument more concrete than simply a direction (more/ less, lower/higher, etc.)? To voters -and elite actors such as competing parties and interest organisations -the informational value of policy signals increases with each element. Importantly, parties may not always want policy signals to be very specific. Some recent work suggests that parties use vague policy signals to attract voter segments that may not actually agree with each other (Bräuninger and Giger 2018;Somer-Topcu 2015;Tromborg 2021), and sometimes, being specific on issues that are plain difficult to solve may reveal that there are no easy solutions available. At other times, specificity may be exactly what a party wants because it helps create an edge to other parties. As noted, it is one thing to want fewer immigrants but quite another to 'build a wall and have Mexico pay for it' . The latter type of statement has a greater potential for generating conflict and attracts attention to a party's favourite issue, which can be electorally beneficial (Green-Pedersen 2019; Schattschneider 1960). 1

Control over outcomes as a predictor of policy signals' concreteness
As of today, we have limited knowledge about the landscape of policy signals as conceptualised here. One important dimension in this regard, we hypothesise, is the degree of control over outcomes. Government parties, for one thing, tend to have a greater ability to affect outcomes than opposition parties do. In parliamentary democracies, the government is formed with the support of a majority in parliament, meaning that either the government on its own or together with its support parties can pass legislation. At the same time, governments have access to the ministerial bureaucracies, which leaves them with substantial powers over administrative regulation and legislative agenda setting (Laver and Shepsle 1996;Lijphart 1999;Strøm et al. 2003).
Because governments have more power than the parliamentary opposition, or at least are perceived as having more power, they are also expected to provide concrete solutions to social problems, whereas the opposition has more freedom to engage in blue-sky thinking; proposing more lofty ambitions than hard paths to see them through (Klingemann et al. 1994;Green-Pedersen and Mortensen 2010). The result of this -real or perceived -asymmetry in control and, hence, responsibility should be reflected in the type of policy signals that parties use. Governments will employ more specific policy signals than the opposition (cf. also Eichorst and Lin 2019), who, on the other hand, might be expected to use policy signals that particularly emphasise policy objectives; that is, the goals of the parties rather than how to achieve these. This leads to the first hypothesis.
Hypothesis 1: Government parties' policy signals contain fewer policy objectives but more concrete policy instruments than opposition parties' policy signals.
Many governments consist of more than one party. Indeed, multiparty governments are quite common in Europe and we might speculate that the logic of control outlined above even applies within these governments. A party with the ministerial responsibility for a certain area may be perceived as particularly responsible and very often actually have a better opportunity to develop specific policy proposals (Laver and Shepsle 1996). This would lead to an expectation of intra-governmental variation in concreteness: Hypothesis 2: Within multi-party governments, parties' policy signals contain fewer policy objectives but more concrete policy instruments on issues covered by their portfolios.
Control may also matter across policy areas because some areas are seen as easier to provide concrete policy instruments for. Some areas are embedded firmly within the national political and administrative hierarchy and have been so for decades. This means that governments can be expected to provide concrete solutions to problems. Health care and the educational system are two examples of long-established services under the control of national or regional governments, but also areas such as law and order and labour regulation have a strong national component to them even in an era of European integration. Other areas are outside the direct control of governments, especially for EU member states. Civil rights are nested in a European and international legal framework, just as immigration is to a large extent (control over policy areas presumably varies between countries, so we return to this point in more detail when we discuss our case selection). Importantly, whether or not an area is outside the control of governments should not affect parties' propensity to signal policy objectives, only the likelihood that such objectives are coupled with policy instruments or that these policy instruments are more concrete than simply suggesting a direction, such as more or less, higher or lower and so on. This is the third hypothesis.
Hypothesis 3: Policy signals about policy areas where governments have greater control over outcomes contain more concrete policy instruments than policy areas where governments have less control over outcomes.
Given the logic outlined above, we expect that government status and policy area might interact. For areas where government cannot exert much control, becoming too concrete can be electorally dangerous even for government parties as expectations are raised in a situation where the government cannot deliver. On areas where governments tend to have -and are expected to have -control, the difference between government and opposition parties should be all the greater. Here, the notion that governments both could and should solve society's problems ought to be prevalent. This, in sum, is the fourth and final hypothesis.
Hypothesis 4: The difference between government and opposition parties regarding the concreteness of policy instruments in their policy signals is greatest on policy areas where government parties have most control over outcomes.

Research strategy
As noted in the introduction, large-N empirical studies of parties' policy signals frequently rely on the Manifesto Project (Volkens et al. 2019), which has coded election manifestos from up to 56 countries for several decades, or the Chapel Hill expert surveys (Bakker et al. 2015), which gauge party positions based on experts' statements. With these datasets, it is possible to connect a party to a single point on a predefined ideological scale (e.g. more-less redistribution, more-less environmental protection, more-less European integration) -what we call parties' policy objectives. These extant datasets clearly have many advantages and have been put to much valuable use over the years. Nevertheless, from the point of view of studying parties' policy signals in detail, they are wanting.
First, they cannot inform us whether a policy signal contains only a policy objective or if it also includes a policy instrument, and whether that instrument is concrete or not. To the best of our knowledge, no dataset exists that codes the presence of these -we argue, vital -elements of policy signals (although this is not an inherent disadvantage of these data sources, and certainly not party manifestos, which could be coded for policy instruments and concreteness). Second, there is a real concern that neither manifestos nor expert surveys capture policy signals in the sense we conceptualise them here; that is, as parties' deliberate attempt to convey the policy objectives and associated policy instruments directly to their potential voters. Of course, manifestos may be argued to be parties' signals to voters during election campaigns. Yet, few imagine that voters actually read manifestos, and research suggests that the policy objectives of the manifestos do not filter down to voters (Adams et al. 2011). This may be because journalists do not read manifestos either or because the aggregated way in which various issue dimensions are constructed does not correspond to how voters are confronted with policy signals in real life. Expert surveys such as the Chapel Hill project are tougher still to relate to any specific stage in the policy-signalling process because experts are likely to base their assessment on a mix of what parties say they want to do and what they actually have done in the past (Bakker et al. 2015;Netjes and Binnema 2007).
In the current day and age, much party communication takes place online via social media. Social media such as Facebook or Twitter has the great advantage for parties that they are unmediated by journalists or other gatekeepers, except in the extreme event where the social media itself censors a post. Party communication via social media is, of course, likely to be most effective if it is picked up by the mass media, but from the point of view of our study, parties' social media communication is an ideal data source. First, data on parties' social media posts captures what parties would like to convey to their potential voters, not what journalists chose to report. Second, parties typically have a large number of followers on social media channels, who are exposed to the social media posts from the party -and many of these will share or retweet the posts to their own followers. The Conservatives in the UK, for instance, have around 750,000 and Labour around 1.1 million followers on Facebook. In Denmark, with a population around 10 times smaller than the British, the Social Democrats have 115,000 followers, while the Danish People's Party have 105,000. While this is only a fraction of the total electorate, it is still enough to make policy signals communicated via a social media platform much more likely to reach a large audience than most other forms of communication. 2 Third, a large segment of the public receives its news mostly via social media (e.g. Pew Research Center 2021), meaning that even if parties' Facebook posts were not to affect the content of the mass media, they might still affect the public's news consumption. Finally, however, studies actually find that the social media agenda affects the mass media agenda as much as the mass media affects social media, so there clearly is an independent and well-documented effect of communication online (e.g. Harder, Sevenans, and Van Aelst 2017;Valenzuela, Puente, and Flores 2017;Su and Borah 2019;Gilardi et al. 2021).
In our study, we collect the Facebook posts of the nine parties in the Danish parliament for the last year until the June 2019 National Election. With nine parties -each with its own Facebook site -our dataset comprises 3,963 posts, equivalent to 237,720 words or around 600 pages of unabridged text (not taking into account the audiovisual material that we also code), yielding a large enough sample to draw robust conclusions about parties' policy signals, and how they vary across government and opposition as well as across issues. We focus on Facebook rather than Twitter or other social media because Facebook remains by far the largest social media platform. In 2020, Facebook had 2.7 billion monthly users, while Twitter had 330 million monthly users. We study parties' Facebook posts rather than candidates' because our theoretical concern -in the context of parliamentary democracies -is about parties' policy signals more than individual candidates' policy signals. Of course, this does not foreclose investigating candidates' Facebook posts, or that these posts are not as important as parties' , only that this is not the natural place to start given our research question (we concede that personal profiles may be more personal/less political, but we do not know).
Denmark has a multiparty system and intense electoral competition and is, in this sense, representative of most European democracies. There is no particular reason to expect that Danish parties' usage of Facebook should be distinctive from how parties elsewhere in Europe use the platform, although an obvious path for future research is to study how the institutional structure affects the distinction between government and opposition, which we hypothesise to be important. One might imagine, for instance, that the type of government -majority or minority, single party or coalitional could matter for perceptions of control. Indeed, this is at least partly captured in our second hypothesis, which only relates to a multiparty government setting. Denmark may also be distinct from other countries in terms of governments' control over specific issues, which may vary cross-nationally depending on the degree of centralisation at the national level as well as the power a country can project internationally. As we discuss further below, Denmark is a small, but rather centralised country and this affects what issues we imagine will exhibit high/low levels of concreteness.
The three core variables we coded for the current study are the presence of a policy objective, the presence of a policy instrument, and whether the policy instrument is concrete or not. The variables were coded as dummies (no = 0; yes = 1) and defined this way in our codebook: Policy objective: Is there any aim specified that relates to changing or protecting the social or natural world (e.g. less poverty, a worthier life for the old, protecting the national culture, improving the environment, ensuring economic growth, fewer immigrants in the country, more cancer survivors, a better balance between the cities and the countryside)?
Policy instrument: Is there a policy instrument specified to reach the aim? A policy instrument is defined as an action by a public authority at the supranational, national, regional or local level. A policy instrument may be either regulatory (e.g. a ban on smoking or higher speed limits), economic (e.g. more government spending on health care or larger fines for littering), or informational (e.g. an awareness campaign against hate crimes). Note that a non-decision is not an instrument, so if a post mentions specific instruments that it finds inappropriate but does not mention an alternative instrument, it should be coded as 0.
Concreteness: Is the instrument more specific than just directional (e.g. numbers such as quotas, minima, maxima are provided; bans; abolishment of laws and regulation)?
In order to give the reader a better idea of the sentences we capture and how these distinctions play out when applied to Danish parties, we list examples for posts that contain 1) no objective, 2) an objective (but no instrument), 3) a policy instrument (but not concrete), and 4) a concrete policy instrument. These posts are translated from Danish and exclude the picture, audio-and video content we coded. In Online Appendix 2, we provide the full codebook and a draw of 114 examples of posts that have been categorised as containing no policy objective, a policy objective, a policy instrument or a concrete instrument (we drew on a 10% sample and subtracted picture, audio-and video posts). As can be seen, some of these posts are longer than the examples we are able to list in the following:

No policy objective:
Winners of the 4th Advent Quiz -10 cool DF-mugs. Now DF's Christmas Elf found the lucky winners of some cool DF-mugs. The winners […] will get a personal email. Again and again the blue government overstates the numbers. However, if you look a bit closer, you will see that only a few people benefit from the government's senior pension. We will introduce an actual right to an earlier retirement for the people that are worn out. A right where you don't have to show up at the council office with your hat in your hand [i.e. as a beggar] to ask for early retirement. Everyone deserves a worthy pension. Share if you agree! [Socialdemokratiet/Social Democrats]

Policy objective, policy instrument, concreteness:
It's simply not OK that it's that easy to cheat regarding taxes. We need way better control. We want to establish 4 new tax-centres around the country and hire 1000 new employees. That's how we can increase control and make life harder for those people cheating and swindling.
[Socialdemokratiet/Social Democrats] It's now only the government that cannot see it: We have way too little adults to take care of our children. Parents, researchers, pedagogues and leaders agree that we NEED more adults if we want to make sure our children receive care, learning and play in day care. SF says that we can never have more than 3 children for every adult in baby nursery and never more than 6 children for every adult in nursery. Minimum staffing now! [SF/Socialist People's Party] Apart from these three variables, we relied on the Comparative Agendas Project's coding scheme to assign each post to one of 21 issue categories. As outlined at the beginning of the codebook and in the description of variables 20 (policy field 1) and 21 (policy field 2), we code the 'order of appearance' in the post and focus on the 'most relevant' CAP code (100 re-codings yield a Krippendorff 's Alpha .82, see Table A3.1). Since less than 20% of the posts relate to a second policy field, we disregard that aspect. It turned out, perhaps unsurprisingly, that a large majority of posts fell into just 10 issue categories (health, law and crime, labour, education, macroeconomics, social welfare, environment, immigration, civil rights, and international affairs), which are the ones we focus on in the issue-specific analyses below. In Denmark, health care and education are delivered by subnational units but are largely under the legislative and regulatory control of the national government. Law and crime, macroeconomics, and labour regulation are also under government control, although not least the last two are influenced by European priorities and corporatist negotiations on some issues. As a small EU member, Denmark, on the other hand, depends highly on other countries and institutions when it comes to international affairs, civil rights, and to some degree also immigration (on the impact of EU rules in Danish legislation across different policy areas, see Christensen 2010; Jakobsen and Mortensen 2012; on the influence in international affairs, see Petersen 2006). The reason that, in case a post contained two or more issues, we coded the issue that appeared first in the post, is that we want be as systematic and transparent as possible, but also because we assume that the most important information will tend to come first in fast-paced media like Facebook (as in news media generally).
Two student coders coded the variables after being thoroughly trained in the codebook and after intercoder reliability tests had been performed for all the variables. After two rounds of extensive training and feedback, the Krippendorff 's alpha for 100 coded posts was 0.85 for the policy objectives, 0.75 for the policy instrument, 0.78 for the concreteness of the policy instrument, and 0.82 for the policy area, when we used the codes of the Comparative Agendas Project (we only coded the 21 top categories, not the 256 subcategories). In addition, we held regular team meetings to provide guidance on difficult cases. Table A3.1 summarises all intercoder reliability scores.
A special feature of our dataset is that we also code videos. These audiovisual elements -speeches, interviews, political advertisements, and so on -are often an integral part of parties' Facebook posts, with much or all of the relevant communication embedded here rather than in the plain text. Indeed, 35% of all posts in our dataset contained a video, and frequently, these posts included very little text. Because we employed manual coding, rather than automated classification based on machine learning or a dictionary approach, we were able to code these audio-visual elements with the same coding scheme used to code the text of the posts. 3 Figure 1 summarises the proportion of all 3,963 posts that includes a policy objective, policy instrument, or a concrete policy instrument. More than 87% of all posts include a policy objective. The implication is that less than 13% of parties' posts are purely 'soft news'; that is, posts with no value as a policy signal to voters about what parties would like to achieve. It is worth underscoring that many posts with a policy objective contain non-political elements (personal stories, various forms of entertainment, etc.), but that is part of the attraction of Facebook and a reason it is so popular with the mass public. However, the key question is whether Facebook users are presented with a relevant policy objective, and they very frequently are. Around 51% of the time, they are also presented with a policy instrument. That is, they are not only told what a party would like to achieve, but also how they will achieve it. 18% of the posts contain a concrete policy instrument -that is, more specific than just higher or lower, more or less. In other words, there is a steep drop-off as we move from policy objective over policy instruments to concrete policy instruments. Parties are overwhelmingly talking about what their goals are but less talkative when it comes to the details of reaching the goals. Figure 2 breaks the data reported in Figure 1 down by the parties' status as belonging to either the government (the Liberal Party, the Conservative Party, and the Liberal Alliance) or the opposition (the Social Democrats, the Unity List, the Socialist People's Party, the Social Liberals, the Alternative, and the Danish People's Party). The overall pattern is the same across the two groups of parties with policy objectives much more prevalent than policy instruments and especially concrete policy instruments. However, it is evident that there are substantial differences in the expected direction. Facebook posts of government parties include a policy objective in 80% of the time, while opposition parties' posts include one in 92% of the time. In contrast, government posts include a concrete policy instrument 21% of the time, whereas this is the case with 15% of opposition parties' posts. A six percentage point difference might appear trivial but still reflects that government parties include a concrete policy instrument near to 50% more frequently than the opposition. In sum, Figure 2 suggests that government parties tend to be less focussed on grand goals and more on concrete policy instruments than opposition parties are.

Policy signals across government-opposition status and policy issue
In Table 1, we report three logistic regression models (M1-M3) to see if the association holds up if we control for the number of words in each post (since longer posts have more space for more content; the exclusion of this control does not affect the reported results) and with standard errors clustered around parties (because parties may differ in their policy signals disregarding their government status). The results corroborate the conclusions drawn from Figure 2. The odds ratio for the government party dummy in M1 is .49 (p = .07), indicating that posts by government parties are less likely to include a policy objective than  posts from opposition parties. While there is no statistically significant difference for policy instruments (in M2), there is a large difference for concrete policy instruments (in M3). Indeed, by this estimate, government parties are 72% more likely to include a concrete policy instrument compared to the opposition. Overall, this confirms the first hypothesis. The second hypothesis stated that within the government -which at the time we collected our data consisted of three parties -parties should be more concrete on issues that are under their own ministerial jurisdiction. To explore this, we matched the 18 CAP categories to the ministerial portfolios of the Danish government (leaving out three categories that have no natural association with any portfolio, cf. the list in the online appendices). Figure 3 displays the descriptive results. Posts from parties with a relevant portfolio are almost twice as likely to include a concrete policy instrument than parties without a portfolio. They are also more likely to include a policy objective; indeed, they do so in all of their posts. This suggests that there is no necessary trade-off between being concrete and advocating a position. Table 1 (M4 and M5) reports the results from two regressions that confirms the descriptive pattern for policy instruments and concrete policy instrument (because all posts contain a policy objective from the portfolio holder we cannot estimate a model for this outcome).
There are a number of potential party-based arguments, which one could imagine might also be at play, but it turns out that the government-opposition distinction is the main party-based predictor (the results for these analyses are reported in Online Appendix 6, Table  6.1-6.5). For instance, there is no statistically significant difference between parties to the left and right of the ideological spectrum in terms of the content of their Facebook posts. Neither is it the case that parties situated on the fringes vary from more centre-oriented parties. Vote share is, in fact, positively associated with more mentioning on policy instruments, but not more concreteness. This finding is hard to interpret, but might reflect that size goes together with government experience and ambition (but again: size is not associated with concreteness). It might also be that parties are more concrete on issues that they 'own' or are closely associated with (e.g. New Left parties with the environment and New Right parties with law and order). To see if this is the case, we grouped the parties according to their position on the (old) economic dimension and the (new) socio-cultural dimension, respectively, and included these in regressions that only included issues that characterise these dimensions. Yet, again, there were no systematic relationships; it is consistently the case that the centre-right parties are more concrete, and this, we argue, is because they happened to be in government at the time. It is important to be precise here: our results do not indicate that issue ownership or ideological reputation is not associated with issue emphasis, as a long line of work otherwise suggests; our results strictly relate to whether parties mention a (concrete) policy instrument or not. 4 Since we collected data for the runner-up year to the 2019 National Election, it is possible to test if proximity to the election affects party communication -either making parties more or less concrete. The election campaign of 2019 in Denmark lasted for four weeks, but there is no indication that this period is different from the previous 11 months, and neither is it the case that the type of communication altered as the election drew nearer (employing a linear count variable instead of an election campaign dummy). Overall, the conclusion is straightforward: the major and comparably invariant factor appears to be government-opposition status.
Moving to the test of the third hypothesis, Figure 4 reports the content across the 10 policy areas that have more than 100 posts. We deselect the smaller areas to be able to have more confidence in the empirical patterns we locate. The areas are ranked according to their proportion of concrete policy instruments with the area with the largest proportion at the top. Three findings stand out. First, virtually all posts include a policy objective. When parties communicate about one of these policy areas, they always communicate about something they want to achieve. Only a few percent of all posts that are related to one of these 10 issues do not have a policy objective (this is most easily visible with regard to the civil right issue). Second, there is a strong gradient in the volume of concrete policy instruments. The ordering of the policy areas fits the notion of governmental control, with those areas embedded in the national political and bureaucratic hierarchy seeing much more specific policy signals than areas that are not. Posts relating to health are by far more likely to include a concrete instrument than all the other areas. 42% of all health-related posts contain a concrete policy instrument compared to 28% of post relating to law and crime, the area with the second-highest proportion (the difference is statistically significant with p < 0.05). The difference between the next areas becomes smaller, but the ends remain significantly different from each other. For instance, the 28% concrete policy instruments in posts relating to law and order deviate in a statistically significant way from the 16% of posts relating to immigration, 13% to civil rights, and 12% to international affairs (p < 0.05). This hierarchy of issues mirrors existing research documenting the highly varied impact of EU legislation on Danish law-making and the generally limited impact Denmark has on international affairs (Christensen 2010;Jakobsen and Mortensen 2012;Petersen 2006).
Third, it is noteworthy that the proportion of policy instruments is substantially higher than the volume of concrete policy instruments and that the gradient is less pronounced, although the rank order is roughly the same. For instance, health-related posts contain a policy instrument 76% of the time, while this is only true for 48% of the posts on the environment. Yet, this is the extreme ends, and for most of the areas in-between, the observed differences are not statistically significant. One interpretation of this is that, disregarding the policy area, it is easier to indicate a direction in a policy instrument than having to commit to a specific setting or amount. Parties can claim to want tougher sanctions on foreign states supporting terrorists (a policy instrument) but refrain from explaining how this increase should come about more precisely because the issue cannot meaningfully be settled by a Danish government on its own. In short, being more precise risks exposing the impotence of the party communicating the policy signal.
Returning to Table 1, we see that M6-M8 report logistic regression models supporting these conclusions (the number of observations differ as this part of the analysis only includes the issues where there is more than 100 posts). The explanatory variable is a new variable where the policy areas are ranked according to the volume of concrete policy instruments; that is, in the order the policy areas appear in Figure 4. The regressions again include the posts' word count as a control and cluster the standard errors around parties. While M6 shows an insignificant relationship between the policy issue variable and policy objectives, there are highly significant associations for instruments in M7 and concrete instruments in M8. Since the ranked policy area variable is coded Figure 5. the predicted probability of government-opposition status across policy issues. note: the figure reports the marginal effects from two logistic regressions predicting the probability a Facebook post will contain a policy instrument (on the left) or a concrete policy instrument (on the right). Whiskers represent 95% confidence intervals.
based on the observed variation in Figure 4, the main takeaway from these regressions is that the pattern is not confounded by variation in word count or what party is making the post. As an alternative to the continuous ranked policy area variable, Table A8.1 in the Online Appendix report a set of regression where we include issue-fixed effects instead. Using this approach yields a similar conclusion.
So far, we have shown that government-opposition status and certain policy areas are associated with a greater chance that a post will contain not least concrete policy instruments; that is, policy instruments that are more specific than simply stipulating a direction. A natural next step is to explore if the two factors interact, as formulated in the fourth hypothesis. Of course, virtually all posts related to policy areas contain a policy objective (cf. Figure 4), so by mathematical necessity, there can be no significant, let alone substantially interesting, interaction between government-opposition status and policy field. When it comes to the presence of policy instruments and concrete policy instruments, whether the two explanatory factors will interact is much more open. To study if they do, we rely on the model specification employed in Table 1 with word count as control and standard errors clustered around parties -but this time, we include an interaction term between the Government party and Ranked policy area variables. The estimated marginal effects are reported in Figure 5.
The basic pattern is the same both for the logistic regression predicting the probability that a post will contain a policy instrument (on the left-hand side) and the one predicting the probability that it will contain a concrete policy instrument (on the right-hand side). Both the government and the opposition are rather unspecific when they address issues over which the national government in a small EU member state such as Denmark has little control. This changes as we move towards issues that are easier to affect, but the effect is largest for the government. The probability that a Facebook post from a government party contains a policy instrument doubles from 42% on international affairs issues to 82% on health issues. For the opposition, the equivalent increase in probability is from 40% to 69%. For macroeconomics, education, labour, law and crime, and health, the differences between government and opposition are statistically significant. Turning to the probability that a post will contain a concrete policy instrument, we see that the differences between the government and opposition are amplified, albeit at a lower absolute level (because there are fewer posts with concrete policy instruments overall). The opposition is equally unlikely to include a concrete policy instrument in a Facebook post no matter what policy issue they focus on. On the other hand, posts from the government are strongly affected by the policy issue in question. For issues such as law and crime and health, there is a 40% probability that a government Facebook post will contain a concrete policy instrument, up from less than 20% for international affairs and civil rights.

Conclusion
According to a long line of research, political parties signal their policy position to potential voters, who then select the party to vote on (e.g. Adams et al. 2005;Downs 1957;Klingemann et al. 1994;Manow et al. 2018;Somer-Topcu 2015;Tromborg 2021). What such policy signals look like more precisely has received less attention. We have added to the literature by conceptualising policy signals as consisting of three elements: the policy objective, the policy instrument, and the concreteness of the policy instrument. Doing so allows us to differentiate between weak and strong policy signals as well as to conduct a more systematic analysis of parties' policy signalling.
Overall, we find that parties engage in extensive policy signalling. This is good news for extant research, which assumes that such signalling occurs. However, looking deeper into the data, we also found considerable variation between government and opposition parties as well as between policy issues. We hypothesised that the variation is driven by differences in control. Governments are expected to solve problems and have the power to do so, whereas the opposition is free to speculate and propose more or less realistic ideas. This is reflected in the fact that Facebook posts by government parties much more frequently contain concrete policy instruments rather than simply stating a goal -and this is particularly true for parties that have the ministerial responsibility for an issue area. 5 Party communication relating to policy areas under national legislative control also tended to be more specific than communication about policy issues outside national control. Interestingly, the two factors interact so that the government-opposition distinction is especially pronounced for policy areas that governments can exert the most control over.
There is a booming literature on broad-based -or vague -policy appeals. Generally speaking, the argument is that parties sometimes gain from not being too specific about what they want to achieve (e.g. Bräuninger and Giger 2018;Somer-Topcu 2015;Tromborg 2021). Our results add to this literature by showing that vagueness varies according to control, but also that how we understand what policy signals are to begin with can matter significantly for the conclusions we draw. Whether we focus on policy objectives or (concrete) policy instruments makes a difference for whether we want to consider parties' communication vague or not. This point is also of value to new research on the rhetorical vagueness (Eichorst and Lin 2019) and future orientation (Müller 2021) of election manifestos: party communication is (also) about how parties pursue their policy objectives.
The findings reported here are good news for those concerned with the state of modern democracy. It is well established that there is a strong focus on individual candidates, so-called personalisation (Bøggild and Pedersen 2018;Fridkin and Kenney 2011), more negative campaigning (Lau and Rovner 2009;Walter et al. 2014), and a generally uncivil tone between elite actors (Skytte 2021). However, these observations should not distract from the fact that parties also -and in fact to a fairly large extent -talk about politics; what they want to achieve and how they are going to do it. This is the precondition for voters to make an informed choice and, hence, ground for optimism for those caring about the state of democracy.

Notes
1. Other elements could be added to policy objective, policy instrument and concreteness, e.g. 'reasonableness' or 'feasibility' . A policy objective may be linked to a highly concrete instrument, but the proposal may nevertheless be extremely unlikely to be implemented or simply nonsense. Currently, we avoid this thorny issue because there can be a high degree of subjective evaluation -and hence social construction -in what is regarded as reasonable or feasible. 2. It is not possible for us to assess how many individuals read each of the posts. However, we do have information on user reactions and the number of followers relative to the number of members. First, regarding the latter, it is an indication for the reach of social media that the follower count exceeds the number of party members by far for all parties (Alternativet 82000 to 2998, Dansk Folkeparti 106502 to 10596, Enhedslisten 92887 to 9831, Konservative Folkeparti 48750 to 12257, Liberal Alliance 97582 to 4200, Radikale 49241 to 6872, SF 55623 to 9206, Socialdemokratiet 119120 to 35470, Venstre 37692 to 29360). Second, we know that posts resonate with people. Parties' posts were on average 'liked' 510 times and shared 134 times. Yet these averages cover high variation. 10% of the posts got more than 11,000 likes and were shared 17,000 times. Preliminary analysis suggests that comparing opposition and government yields no robust patterns independent of the size of the follower count. Finally, the potential alternative -manifestos -have a lower reach in Denmark then elsewhere and are often quite short (in 2019 a few hundred sentences). 3. Details on how all the items are worded and coded and how we dealt with the specific challenges of audio-visual posts can be found in the codebook in the online appendices. We should also add that the parties did not issue 'promoted' posts. As outlined in the codebook instructions, we exclude 'Facebook events' or mere updates of profile pictures. 4. As a sensitivity test, we also ran the main models excluding one party at the time. The results were generally very stable.
5. Regarding the question whether ministerial portfolio or issue ownership increase the probability of a post including a policy instrument or a concrete policy signal, we should add that it could be that both play a role. A party that owns a policy issue may be more concrete about this issue and hold the corresponding portfolio since the issue is more important for the party. While it is not possible to disentangle these differences in this article, future research should address this potential mechanism.