Mobilizing or chasing voters on Facebook? Analysing echo­ chamber effects at the UK parliamentary general election 2019

Alongside the growth of online campaigning has been an increased anxiety around its effects on democratic institutions and processes. Many have suggested that in a (new) media environment that privileges choice, citizens will increasingly segment themselves into echo chambers, tuning out dissenting voices. But the debate on the existence of echo chambers is mixed, and the extent to which political parties campaign to easily persuadable (and pre-disposed) elector-ates is unclear. In this article, we present a case study of the Facebook campaign activity of the Conservatives, Labour and the Liberal Democrats at the UK’s parliamentary general election in 2019. Utilising an analysis of the Facebook Ad Archive’s Graph Application Programming Interface (API), we ﬁnd that political parties do not consistently campaign to their easily persuadable electorates—of-ten chasing votes as much as they mobilise supporters. The evidence that parties campaign to speciﬁc echo chambers online is therefore, at best, mixed.


Introduction
As politics increasingly happens online, much has been written about the development of political echo chambers. The suggestion is that citizens either by design, underlying behavioural convictions or simply by accident only engage with pre-existing views that they agree with. On the supply side, it is suggested that either as a result of algorithmic consequence (Pariser, 2011) or decision making at the elite/party level (Gibson, 2020), messages are increasingly only being seen by certain subsets of an already pre-disposed populace. On the demand side, it is suggested that we choose to engage, more often than not, with those views with which we agree. However, there is a growing body of evidence that anxieties around the growth of the online echo chamber (in politics and wider society) is overstated (see Dubois and Blank, 2018).
Using data from the Facebook Ad Archive and Graph API, we investigate the existence of echo chamber campaigning at the UK parliamentary general election 2019 (UKPGE 2019). We focus on the three biggest parties (the Conservatives, Labour and the Liberal Democrats) and test whether their messaging was primarily designed around generic 'get out the vote' (GOTV) appeals. Where GOTV appeals are not apparent, we also test the extent to which the parties campaigned around areas that they have long-established ownership over, before looking at the demographics they targeted. We do this to ascertain to a greater degree whether political parties in this context acted as mobilisers (of existing support) or chasers (of potential support) (Rohrschneider, 2002). However, our results show a mixed picture. Whilst we see more than zero echo chamber campaigning by these parties, it is less than complete. We therefore add a UK perspective to the growing literature somewhat sceptical that online platforms (and political activity on said platforms) lead to the growth of echo chambers.

The rise of digital campaigning
In political campaigns, actors often adapt to the evolving capabilities of technology to make their messages heard-from radio in the early 1900s, to TV in the 1960s, through to the Internet in the 21st century. The growth of digital media and especially social media is fast moving. With this comes an array of new political communication strategies-such that academics now argue that we have entered a 'fourth phase' of political campaigns (Magin et al., 2017;Roemmele and Gibson, 2020).
The platforms that are now the prominent battlefields within this new era of political campaigning, therefore, have increasingly important effects on both the democratic process and democratic outcomes therein (see e.g. Vaidhyanathan, 2018;Van Dijck et al., 2018). And whilst we see many contemporary similarities with those that scholars highlighted in the 'third phase' (see Blumler and Kavanagh, 1999)-such as the use of consultants, targeted advertising and a reliance on data and market research companies-part of the reality of the fourth phase of campaigning is that it has changed the political landscape in two strategically significant ways. The cost of campaigning has been lowered considerably whilst at the same time offering vastly more precise targeting capabilities (Franklin Fowler et al., 2021).
The nuances of digital campaigning show that not every campaigner is a data expert nor are campaigns necessarily using data efficiently . However, the possibilities different advertising interfaces offer in terms of providing new ways to micro-target individuals and new ways to run targeted messaging are now being utilised (to varying degrees) in political campaigns. The Facebook interface, for example, allows advertising objectives to be set alongside audience, budget and format, whilst also offering A/B testing to gather a range of metrics that measure the effectiveness of different messaging strategies in real-time.
The 'challenge' digital campaigning presents, therefore, is twofold: we are presented with a (relatively) new and (relatively) unregulated campaigning arena, which may present (relatively) novel opportunities for campaigners to operate with impunity. This is then combined with a lack of transparency around how these campaigners are operating (see Dommett and Power, 2020), which merely serves to add to perceptions that actors are, indeed, operating with impunity.
This, the Digital, Culture, Media and Sport Select Committee (in the UK) argues, leads to the 'coarsening of public debate' but also, more worryingly, when 'brought to bear directly on election campaigns' threatens 'the very fabric of our democracy' (DCMS, 2019). Little wonder, then, that recent academic worktracing the rise of digital campaigning from the margins to the mainstream-has outlined that the changes hold 'profound significance, not just for the future of political parties and elections, but for the longer-term health and sustainability of the wider democratic project' (Gibson, 2020, p. 223).

An information crisis?
One of the ways in which these broad concerns manifest is in a focus on the role that 'echo chambers' play in current political discourse and political campaigns (Bakshy et al., 2015). The argument is that democracy as both an ideal and institution is harmed when citizens only attend to appeals from campaigns they support. This chimes with wider work in the field of governance and corruption which suggests that both tangible and, more often than not, intangible damage can be wrought in democracies when fundamental pillars-such as choice, debate and representation-are either bypassed, or are seen to be bypassed (Heywood, 2020).
Concerns surrounding echo chambers when tested empirically tend to occur on the supply side and at the level of platform and/or party (though for a more agent-centred interpretation, see Sunstein, 2017). Pariser (2011) suggests that platform algorithms filter the news and content that citizens engage with, such that people eventually do so only with those articles structured around their own preferences (Kitchens et al., 2020). This is exacerbated by the news environment on said platforms, which encourages popularity-based sorting and liking-effectively creating a path-dependent feedback loop where 'attitudinally congruent articles remain at the top of the feed' (Shmargad and Klar, 2020, p. 423).
At the party level, it is argued that this 'sorting' occurs via micro-targeting of message to a specifically identified (and tailored) partisan audience (Gibson, 2020;Roemmele and Gibson, 2020). For example, during elections themselves, it has been shown that fewer than half of the adverts on Facebook had, in any way, the goal of voter persuasion, focusing instead on playing to the base and generic calls to 'GOTV' (Franz et al., 2020;Franklin Fowler et al., 2021). These institutional-based explanations often implicitly cast the citizens that engage with these networks as 'trapped' by either algorithmic consequence or elite-driven campaigning decisions. It is argued, therefore, that we should view this era of political campaigning as either subversive (in which actors deliberately 'divide, demobilize and disinform citizens') or scientific (in which the 'narrowing of the message and target audience. . .are not characteristics of a healthy future') (Roemmele and Gibson, 2020, p. 607).
However, the debate around echo chambers is a live one and there is growing evidence that this narrative is overstated (Dubois and Blank, 2018;Guess, 2021). Whilst some predicted the perilous rise of echo chambers (Sunstein, 2002)-and countless media stories surrounding the inherent dangers remain-others suggest the inverse. That the increased opportunity for choice, debate and, indeed, representation, provided by platforms causes a higher level of exposure to diverse points of view (Feezell, 2018).
This optimism finds support in studies, which suggests that there is, in fact, considerable overlap in the accounts that those on opposite ends of the political spectrum follow-though conservatives are more likely to engage with leftleaning accounts than the inverse (Eady et al., 2019). Moreover, that social endorsements often 'trump' partisanship on Facebook in terms of news consumption (Messing and Westwood, 2014) and that most people hold generally 'moderate media diets' with a small group of partisans creating a disproportionate picture of the existence of echo chambers (Guess, 2021). Some have even suggested that this disproportionate picture of 'empirically marginal phenomena' merely serves to create 'moral panics' that distract from more fundamental structural transformations in the online world, which are far more deserving of attention (Jungherr and Schroeder, 2021).
Therefore, on the supply side, we see some empirical support that social media drives citizens into specific echo chambers via either platform or party-mediated design. On the demand side, however, we see a different picture. That citizens engage with moderate media, in moderation (sometimes encouraged by social cues), and those that do not disproportionately amplify any evidence of echo chambers. The picture, in other words, is mixed.
However, the supply and demand side explanations are not necessarily mutually exclusive. It might well be the case that-echo chamber or no echo chamber-party elites attend to the existence of them and campaign as such. In this sense, it would reflect a Thomas Theorem interpretation of behaviour, that 'if men define situations as real, they become real in their consequences' (Thomas and Thomas, 1928, p. 572), which has recently been applied to political decision making .
In this article, we turn to the most prominently used (online) landscape, Facebook, in the UKPGE 2019 (see Dommett and Power, 2020), to gain a better understanding of how political parties use this space and to further clarify the debate in the literature surrounding echo chambers. Elite cues matter in politics. This is important not just to ascertain whether echo chambers do exist online, but in considering their future development. In this sense, if we find support for our hypotheses, it could well be the case that if elites play to non-existent echo chambers (on the supply side), that over time echo chambers develop (on the demand side). To be clear, whilst this may well be a consequence of our argument (and of targeting), our article here is concentrated solely on the supply side, at the party level, and on Facebook. We do not extrapolate from our data the extent to which the voters targeted here exist in an echo-chamber in their wider use of social media or in their lives offline. Though as the above literature review suggests, the evidence points in both directions.

Hypotheses
To test the extent to which the parties at the UKPGE 2019 played to a (perceived or otherwise) echo chamber, we outline three hypotheses drawn from previous literature on both campaigning online (and Facebook), but also political campaigns more broadly. If we find support for these hypotheses, we would infer (on the supply side) an echo chamber effect. For H1, we take our cue largely from the work of Franklin Fowler et al. (2021) who, in the US context, found that the majority of communications on Facebook were centred around generic GOTV appeals and focused on 'purely promotional, valence-oriented ads aimed at mobilizing their base of existing supporters'. If this were repeated in our study, it would suggest a lack of interest in reaching beyond traditional bases of support.
H1: We are more likely to find purely promotional adverts aimed at generic GOTV appeals in party campaign material.
Where issues are discussed we have different expectations, formed primarily from recent work which has outlined that campaigning during the so-called 'fourth phase' is increasingly micro-targeted to specific audiences (Gibson, 2020;Roemmele and Gibson, 2020). In this instance, 'campaigning will increasingly focus on the easily persuadable 'perceived' electorate', which leaves the 'disengaged and 'unperceived' electorate out of scope and under-mobilized' (Gibson, 2020, p. 223).
We test this via both messaging and the target audience that received the said message. In terms of messaging, we expect parties to concentrate on traditional areas of strength and emphasise issues on which they have ownership. This is based in classic understandings in the political science literature under which a party develops a 'reputation for policy and program interests, produced by a history of attention, initiative and innovation toward problems, which leads voters to believe that one of the parties is more sincere and committed to do something' (Petrocik, 1996, p. 826). Of course, that parties concentrate their rhetoric on areas in which they are strong is a widely accepted narrative in our more general understanding of campaign behaviour and not necessarily suggestive of an echo chamber.
However, we suggest that when taken in concert with more general targeting strategies, we can better understand whether, in our case at least, political parties are reaching out beyond the 'easily persuadable electorate'. In terms of issue ownership/selective emphasis theory, we expect that the Conservatives will focus on 'issues related to national interests such as law and order, asylum/immigration, the EU' as well as concern surrounding 'the size of the state, such as taxes' (Seeberg, 2017, p. 488). Similarly, we expect Labour to concentrate on 'issues related to the welfare state such as health, education' and 'unemployment' (Seeberg, 2017, p. 488). In this particular election, and positioned as the 'true party of Remain' (Cutts and Russell, 2020, p. 108), we expect the Liberal Democrats to concentrate their messaging primarily around the EU and, in the sense that they 'co-owned' this issue with the Conservatives (on separate sides), we would also expect a focus on Brexit messaging from the Conservatives. In short, in terms of their focus on issues, we expect parties on Facebook to act as 'mobilisers' as opposed to 'chasers' (Rohrschneider, 2002).
H2: Where GOTV appeals are not utilised, we are more likely to find parties emphasising those issues on which they have traditionally held greater ownership.
In acting as mobilisers, we also expect that parties will target their messaging at specific demographics. There are limits to what the Graph API can show us in this respect (see below), but we can access broad age and gender targeting through these means. Many studies have recently shown that alongside level of education, age is becoming a crucial predictor of voting intention across many democracies; UK included (Ford and Jennings, 2020). With this understanding we expect parties, in acting as mobilisers, to concentrate on specific age brackets. Given that older voters gravitate towards the political right (Tilley and Evans, 2014)-and that these patterns were reproduced around Leave and Remain identities-we expect the Conservatives to target older voters, and Labour and the Liberal Democrats to target younger.
H3: We are more likely to find parties targeting their adverts at ageappropriate demographics.
If we find support for all three hypotheses (in the sense that we see ads targeted primarily in alignment with the above expectations), we can tentatively state that at the UKPGE 2019, the three main political parties acted much more like mobilisers than chasers. This would lend credence to those that suggest that parties interact with online platforms in ways that (re)produce echo chambers. However, any causal claims and generalisations should be understood as less robust than those studies which conduct large-N analysis across cases and context. Furthermore, even if support is found-the above still does not necessarily provide a definitive test of the hypotheses above. If, however, this study disconfirms the above hypotheses-it can be considered a somewhat more definitive outcome concerning the existence of (perceived or otherwise) echo chambers.
We now turn to a description of the method used in terms of the way in which we operationalise our case study, before engaging with a brief description of the Facebook Ad Archive itself and the way in which we used the Graph API to collect our data.

Methods
This article focuses on what the Facebook Ad Archive does, and does not show us about party activity on the platform. Over a period of three months (June-September) in 2020, we extracted and analysed Facebook Ad Archive data (via the Graph API) for insights covering the dates 1 November 2019 up to and including 12 December 2019 as representative of the period from the dissolution of parliament up to the day prior to the general election.
The case study itself is conducted in the tradition of Stake's (2008) understanding of the intrinsic and instrumental case study. Indeed, our study falls somewhere between the two. It is an instrumental case, because we think our findings provide insight into specific issues and may tentatively redraw or confirm certain generalisations from the literature surrounding echo chambers. However, we are clear to step back from making overly ambitious claims to generalisability from a single case study, of a single election, on a single platform. 1 Therefore, our case is intrinsic in the sense that whilst we can draw tentative conclusions about different cases and contexts, we have also (primarily) gained a greater understanding of this particular case. 1 It might also be the case that this election-with its dominant focus on Brexit-should further be considered an outlier. Though all elections, as they happen, are effectively special elections. As more elections happen with this kind of data available, we can increase our confidence with regards to the findings therein.
Given their dominance in terms of income generation and overall spending at elections , we expected the Conservatives and Labour to dominate spending on Facebook. However, when we conducted an initial search into the spending data that the Ad Archive provides, at first glance, it shows the Liberal Democrats outspending (at the upper range) Labour and the Conservatives on ads posted by their respective national party Facebook pages (see Figure 1). 2 That the Liberal Democrats may have outspent their more dominant rivals is an interesting finding in and of itself-though perhaps not surprising. Other studies have outlined that, in terms of total party income, the Liberal Democrats show a great degree of 'resilience' in terms of fundraising, despite periods in the electoral doldrums (Power, 2020, p. 84). 3 Taken together, and given the two-party dominance in terms of overall spend at elections, but the not insignificant spend of the Liberal Democrats on Facebook, we focused our study on these three parties. 4  Although as Dommett and Bakir (2020) have outlined, Labour (and to a lesser extent Conservative) spending is concentrated on both national and regional party pages. 3 Though the Conservatives and Labour do consistently raise considerably more money than the Liberal Democrats. 4 Beyond our three parties of study the Brexit party had the biggest spend (between £339,700 and £712,113), dwarfing that of other prominent minor or ethnoregionalist parties such as the Green Party (between £76,500 and £119,146) and the Scottish National Party (between £31,100 and £39,062). For more on how ad archive figures are calculated (and the methodological issues therein) see Dommett and Bakir, 2020, pp. 209-214.

The Facebook ad archive
In response to numerous pressures, the Facebook Ad archive was built to offer more transparency into political ads. It includes all ads deemed to be about 'social issues, elections or politics' in near real time. There remains little consensus-despite growing pressure from campaigners-around what constitutes a social, electoral or political issue and in the absence of any meaningful (if even flimsy) consensus, Facebook defines this as those that are: • made by, on behalf of, or about a candidate for public office, a political figure, a political party or advocates for the outcome of an election to public office; or • about any election, referendum or ballot initiative, including 'go out and vote' or election campaigns; or • about social issues in any place where the ad is being placed; or • regulated as political advertising. 5 Therefore, in the UK, Facebook defines social issues as those that include: civil and social rights, crime, economy, environmental politics, health, immigration, political values and governance, or security and foreign policy. Ads which fall into this category require a 'paid for by' disclaimer and are stored in the Ad Library for public view. When accessing ad archive data, there are two distinct interfaces to choose from: the Ad Library and the Graph API.

The Ad Library
The Ad Library offers a simple keyword search as a first stop for those who wish to look a little closer at ads on the platform. This basic search is free for anyone to use and pages that match your keywords come up as recommendations so you can skip directly to the ads placed by a particular page.
Ad results are displayed in a most recent format and you have to scroll down manually to see older ads. In terms of date-based enquiry, the search filter only allows: last day, last 7 days, last 30 days, last 90 days or across all time. You can also filter results by reach, region, platform, whether an ad is active or inactive and by the 'paid for by' disclaimer attached to all Facebook political ads.
Once a search is made you can look in closer detail at individual pages and ads. Each page has a summary of recent and all-time spending alongside a page transparency summary, which includes: the page owner, the page history (such as any name changes and/or the date created) and the primary region of those who 5 This category is somewhat notable given that many campaigners have raised the seeming lack of regulation in terms of online activity as an issue; especially as compared with analogous offline activity (though such regulation would nominally fall under the remit of the Electoral Commission). manage the page (e.g. UK). For each individual ad, you get an exact copy of the ad displayed as it appeared with all functioning links and for videos with playback available. Below the ad is an estimate of spend and reach with a button for more ad details. Within the ad details, you get a demographic and regional breakdown estimate of who saw the ad shown in a bar graph format.

The Graph API
The Graph API was used to extract data for analysis rather than the Ad Library because it is far more efficient for large numbers of ads. Unfortunately, the large number of ads being analysed mean that it is simply not feasible to look at every ad exactly as it would have appeared on Facebook. We instead assessed ad content on the text-based data that the Graph API provides. 6 Our API searches used a query which was entered to gather data for the three main party pages one by one. 7 The corresponding page ID field being the only aspect that changed for the three separate queries. This limited the search to Facebook only; inactive ads only; one page ID per query; UK only ads; stop and start date and time of ads; spending and impression estimates; all available content fields apart from the ad_snapshot_url; who funded the ad; and the demographic and regional audiences.
Data were initially taken from each of the pages of the three UK parties. We then reduced our results to include only the ads placed between 01 November 2019 and 12 December 2019. We then compared our data along a range of headings: number of ads, ad variants, spending, impressions, start and stop dates of ads, demographic audiences, regional audiences and observations related to selected keywords chosen that related to our hypotheses. To measure GOTV appeals (H1), we included keywords such as 'poll' and 'register', to measure issue ownership (H2) 'Brexit or Remain', 'Police', 'Immigration', 'NHS', 'Tax', 'School or Education' and looked at how many ads each party posted around these key words. To measure targeting (H3) we analysed each of the main party pages by demographic targeted (age and gender).

Limitations of the Ad Archive
Much has been written about the limitations of the Ad Archive (see Tromble, 2021) and whilst it brings benefits in the round, it can also provide a distorted picture of party activity. The most obvious is in Facebook's banded approach to 6 We did not, for example, analyse any embedded video content, pictures used, or other elements of these online ads that larger research teams, with more resources at their disposal, might choose to. 7 ads_archive?publisher_platforms¼FACEBOOK&limit¼500&ad_active_status¼INACTIVE&search_ page_ids¼ 330250343871&ad_reached_countries¼GB&fields¼ad_delivery_stop_time , ad_deliver-y_start_time , spend , impressions ,funding_entity. how much political parties spend, providing information in terms of 'lower' and 'upper' estimates. Whilst some researchers (see Dommett and Bakir, 2020) have traversed this issue by creating an average between the upper and lower bands (taking daily averages rather than period averages), the information remains largely the product of estimation rather than clear information.
This is also reflected in the targeting data both the Ad Library and Graph API provide for analysis. When placing a political ad on Facebook, there are a huge number of options for how ads will target an audience. This includes, but is not limited to, life event targeting, education level, recent location, interests and employer and consumer classifications. The data available for analysis, however, only allow us to see basic age and gender demographics and no more than England, Scotland, Wales or Northern Ireland for the location of where the ad was seen. With the audience data we cannot tell, for example, whether Wales was targeted for Brexit ads or whether different criteria led to more users in Wales seeing Brexit ads. The demographics may not have been the chosen targets at all and could in fact just be the consequence of some other targeting choices such as 'interested in rugby'. We therefore present the findings with these caveats in mind, taking care not to overlook the 'critical limitation in the data we so eagerly amass' (Tromble, 2021, p. 7). Table 1 shows the result of our analysis of the Ad Archive as a means to test our first two hypotheses. Here you can see the number of total ads placed during the election and how often our keywords were mentioned in said ads. In terms of campaigning around simple GOTV messaging, we only find support for H1 in the case of the Labour Party (61.26%). The Conservatives had no messaging around this and the Liberal Democrats only included these kinds of appeals in 5.27% of their adverts.

Results
In terms of H2, we again see a mixed picture. We do find that the polarised debate surrounding Brexit loomed large in the campaign-with the terms Brexit or Remain found in numerous ads from each party page (Labour 12.89%, Conservatives 23.26% and Liberal Democrat 41.12%). In particular, the Liberal Democrat focus on this does suggest that they played to this specific base of support, as did the Conservatives-though to a lesser extent. That Labour focused less attention on this issue, likewise, suggests a party with no clear messaging in this area (see Goes, 2020), continuing to prevaricate. Beyond Brexit, the Conservatives did appear to focus on traditional base issues, such as tax and police (both featuring in over 10% of ads placed). However, we also see the Conservatives firmly parking their tanks on Labour lawns in terms of campaigning on the NHS, which was mentioned in 28.89% of ads (and to a lesser extent  . That in terms of issues, Labour only really focused on the NHS would further support our hypothesis, but the data ultimately provide too mixed a picture to find support for H2 across all parties (Table 2). 8 To test our third hypothesis, we analysed each of the main pages for demographic and regional audiences. In Figure 2, we have grouped age ranges into three broad brackets (though the Ad Archive provides six) for greater parsimony, we have also left the gender targeting as provided by the Ad Archive in place. Here we find support for H3 in terms of the demographics that each party targeted. We can see that the Conservatives eschewed targeting the 18-34 age bracket concentrating instead on an audience mostly in the age range of 35-54 years and to a lesser degree in the 55þ range. Labour, on the other hand targeted most of their ads at the 18-34 demographic and a similar level at the 35-54 cohort. At the upper end of this age bracket is where we saw the crossover age (47)-the point at which a voter becomes more likely to support the Conservatives-at the 2017 general election (Tonge et al., 2020, p. 297). 9 That the Liberal Democrats primarily targeted the 18-34 age bracket also supports our hypothesis that they aimed their messaging at the most Remain supporting demographic.
In total, then, we present mixed findings for H1 and H2 and find support for H3. Whilst the Labour Party primarily conducted their advertising around a generic GOTV campaign and (to a lesser extent) focused on issues where they are traditionally strong (the NHS)-neither was entirely the case for the Conservatives nor the Liberal Democrats. In terms of targeting, however, we do see a very broad picture that adverts were targeted at those age groups we would expect for parties engaging in 'echo chamber' campaigning. However, the minimal audience data obscure further insight into how political actors are carrying out ad campaigns on the platform.

Discussion
This article presents both an intrinsic and instrumental case study. It is intrinsic in the sense that, in the first instance, it provides a greater understanding of the ways in which the main political parties engaged in echo chamber campaigning (or otherwise) at the UKPGE 2019. It has shown that in this instance the picture is unclear. Indeed, we find mixed evidence that Facebook led to a partisan focus in 2019. In 8 We conducted a final test to see the kinds of ads that were most used by each political party. This allowed us to understand, for example, what the Liberal Democrats were focusing their campaign on-given that our initial keyword search provided <50% coverage of their ad content. We did this by collating all the words that appeared in >4% of party ads on the page. Table 2 is simply provided as an expanded version of Table 1, as means to give a greater idea of the way in which the parties campaigned at the UKPGE 2019-not as a further test of any hypotheses presented here. 9 Though this fell to 39 in 2019.

Mobilizing or Chasing Voters on Facebook?
fact, what we found was the main governing party reaching out and meeting Labour on issues that they traditionally own, such as the NHS. Therefore, at least in terms of messaging, we present little support for the case that the 'fourth phase' of political campaigns necessarily leads to a greater focus on an easily persuadable electorate.
In terms of targeting, political campaigns do tend to focus on those demographics that broadly support either party or message. The Conservatives, for example, targeted the majority of their ads at those in the older age brackets, whereas the Liberal Democrats (as the party of Remain) aimed at younger (potential) voters. But this taken on its own does not lead us to conclude that the three main parties at this election particularly engaged in echo-chamber, voter mobilising campaigns. Whilst primarily an intrinsic case study, the findings presented here are also instrumental in the sense that they help to redraw generalisations and provide space for further research. We have added another sceptical voice to the literature which questions the concerns (and elite discourse) around the development of echo chambers in the online campaign landscape. We also question how 'new' these concerns are. Norris (1997), for example, argues that campaigns were 'transformed' in the early 1950s, 'by the decline of direct linkages between citizens and parties, and the rise of mediated linkages [emphasis added]'. Similarly, it has been shown that the widespread availability of television from the 1960s 'afforded less scope for viewers consistently to tune into their own side of the argument' (Blumler and Kavanagh, 1999, p. 212). Echo chamber concerns, then, are far more likely to be consistent, regardless of the political landscape. That citizens, 'always seeking reinforcement, prefer to wall themselves off from contradictory sources of information' is a concern that has 'remained a constant even as the underlying worries change with the times' (Guess, 2021).
However, this does not mean that trends in the 'fourth phase' are entirely without consequence. For example, the (more) established parties dominated the Facebook campaign. This suggests that Gibson's (2020, p. 13) hypothesis that the fourth phase of campaigning is likely to resemble one of 'normalization' in which 'the level of electoral competition drastically decreases, as only the biggest and best-resourced parties can survive', holds true-at least in terms of spending-in our study. 10 Further work might explicitly test this in other contexts, perhaps where there are a greater number of effective parties. One might also draw an analogy with the cartel party thesis and investigate the extent to which the online 10 And points towards the importance of recognizing broader structural concerns as opposed to 'moral panics' (Jungherr and Schroeder, 2021). world is reshaping party competition in these terms (Katz and Mair, 1995). Normative work might also build on that which suggests that the growth of the Internet creates social and democratic divides between the resource rich and poor, and the way in which these organisational realities raise concerns about fairness and equality (Norris, 2001).
The findings here also show that parties campaign in this space in different ways (much as they do offline) and we see considerable party-level variance. Therefore, it might be that different party families engage differently on these platforms. That the Labour Party fits all three hypotheses is instructive. If similar studies were conducted across a number of countries, might we find that leftleaning parties are more likely to 'mobilise', than 'chase'? Evidence from the US suggests that (on the demand-side) conservatives are more likely to engage with liberal voices than the inverse (Eady et al., 2019). And further studies conducted in the UK have shown that those that support the more 'left-leaning side of the debate have the most difficulty getting along with their opponents' (Duffy et al., 2021, p. 41) That said, it might also be the case that parties in opposition campaign differently on Facebook to those in government (much as they would offline). We also suspect that a large-N comparative study would find considerable variance across cases dependent on, amongst other things, party system and political culture.
However, our article also shows a general lack of transparency and, therefore, a lack of certainty around the way in which parties (and campaigners) use these platforms. We suggest that this is an important finding in and of itself. Calls for transparency in this area are not new (see Dommett and Power, 2020) nor should be seen as a panacea . However, clearer transparency measures which outline with a greater degree of granularity the ways in which targeting occurs on Facebook will allow for academics, journalists and interested members of the public to better understand these processes-and the ways in which these processes might run counter to electoral regulation as written. 11 The transparency 'problem', however, is not one that Facebook (or other platforms) should be expected to solve purely of their own volition. In fact, it might equally be said that the Facebook Ad Archive, with its attendant issues (which are legion), fills a hole where governments are falling short. If transparency is a democratic ideal that underpins politics (in the UK or elsewhere), legislation should follow which ensures that this ideal is upheld. We should not expect companies to uphold them on our behalf or out of a vague sense of goodwill-especially when it might run counter to commercial interests. Facebook, in other words, should be incentivised (or legally obliged) to adhere to minimum transparency 11 Though, as we have highlighted here, the steps one has to go through to gain access to the Graph API suggests this will largely be a job for interested journalists and academics. requirements as a basic cost of conducting business. As it stands, we have shown here that core questions remain surrounding exactly what was being spent, who was campaigning where, and who was seeing certain messaging (see also Dommett and Power, 2020).
To a kind reader, this conclusion might seem somewhat pessimistic. A harsher reading would be that such opacity negates the need for publication altogether. We, of course, disagree. In fact, this seeming weakness is also a strength. The limitations presented in the above are in many ways as important a finding as the empirics, especially in terms of highlighting transparency issues. Moreover, given the difficulties that we show are apparent when researching political activity on Facebook, the evidence presented here is an advance. We think the study has value precisely because we still know relatively little about Facebook activity. Therefore, it provides an important benchmark for future scholars. A foundation for others to build from.

Conclusions
Every study raises more questions than it answers, and this is no different. There are a number of caveats to the findings herein and multiple avenues for future research questions to answer. For example, further studies might more functionally compare online and offline campaigning in this area, perhaps linking ad archive activity with the Open Elections Project. Our study also raises important questions about targeting activity that could be investigated further. As the largest social media platform, Facebook likely has the most diverse audience yet is predominantly used by those between the ages of 25 and 49 years (see Ortiz-Ospina, 2019). Given that younger voters are more likely to use YouTube, Instagram and Snapchat, other studies might investigate the extent to which targeting occurs differently dependent on platform.
We might also question the utility of Labour or the Liberal Democrats spending so much money targeting younger voters on a platform they are less likely to use. Were the Conservatives simply more strategically wise to target the 'median' age user? Whilst it might seem counter-intuitive, targeting less represented groups on Facebook (as relative to other platforms) remains a good strategy. Facebook is, by some margin, the most popular social media site. Therefore, the number of young voters one can reach may well run to more than that on other platforms.
This raises larger questions, however, about whether targeting even cuts through in the first place. Whilst Facebook is the largest platform, with the largest number (if not proportion) of younger voters-there is good evidence that many voters (and especially the young) do not pay all that much attention to election content in the first place (see Fletcher et al., 2020). Therefore, if further studies find that parties do target their voters (to a minute degree of granularity) as 'mobilisers', such that it reflects echo-chamber campaigning on the supply sidethis need not be a cause for concern. We may still voice some scepticism that this will necessitate the eventual development of echo-chambers on the demand side, as an application of the Thomas Theorem might suggest.
Whilst many questions remain, we suggest that this article-despite being reflective of a single case, at a single election-represents an important contribution to fields as diverse as political communication, political campaigning and party organisation. Building on previous research, we cast doubt that parties campaign towards specific echo chambers and further highlight the diversity and complexity of the digital campaign environment. We think that transparency issues aside, our work serves to throw light onto these complexities and creates space for future research. This will allow for a greater understanding of how politics occurs in this space and further reflections as to the ongoing health of representative democracy.