‘A campaign strategy that is not data-driven is likely to fall behind’ (Aristotle 2022). This statement, made by the US data company Aristotle epitomizes recent orthodoxy about the requirements of a modern election campaign. Political parties and campaigning organizations across the democratic world are now often proclaimed to be engaged in increasingly scientific analysis whereby data and analytics are used to ‘produce individual level predictions about citizens’ (Nickerson and Rogers Reference Nickerson and Rogers2014: 51), resulting in sophisticated and targeted campaigns (Gibson Reference Gibson2020). Recent empirical scholarship has been devoted to mapping the uptake of these campaigning practices (Bennett and Bayley Reference Bennett and Bayley2018; Kefford et al. Reference Kefford, Dommett, Baldwin-Philippi, Bannerman, Dobber, Kruschinski, Kruikemeier and Rzepecki2022; Kruschinski and Bene Reference Kruschinski and Bene2021). The conclusion often drawn is that data-driven techniques provide a powerful new tool for efforts to maximize public support and secure desired outcomes, resulting in the diagnosis of a new ‘fourth era’ or ‘phase’ of election campaign – an ‘ideal type’ distinguished by the collection of data and use of analytics techniques for targeting and testing (Kruschinski and Haller Reference Kruschinski and Haller2017; Römmele and Gibson Reference Römmele and Gibson2020).
Alongside this focus of research attention and resource there have, however, been some efforts to question the prevalence of data-driven practices. In particular, Jessica Baldwin-Philippi (Reference Baldwin-Philippi2019) has diagnosed the ‘myth’ of data-driven campaigning, highlighting the tendency for campaign professionals and media commentators to overstate its extent and impact. Elsewhere, a growing body of work has begun to generate empirical evidence on the actual uptake and diversity of data-driven practices, finding significant variation in their extent and sophistication (Dommett et al. Reference Dommett, Kefford and Kruschinski2024; Kefford et al. Reference Kefford, Dommett, Baldwin-Philippi, Bannerman, Dobber, Kruschinski, Kruikemeier and Rzepecki2022; Macintyre Reference Macintyre2020). In this context, we ask, ‘To what extent is there evidence of a fourth era of data-driven campaigning in the UK?’, seeking to determine to what extent this latest ideal type defines current UK practices. In addition, we ask, ‘Are there differences between parties’ adoption of different campaign-era practices?’
In posing these questions we confront a particular challenge, as parties themselves disclose limited information about their campaign practice, and state-mandated transparency information tends not to provide detailed insight into campaign activity. Whilst it is possible to use standalone resources, such as the Facebook advertising archive, to assess elements of parties' campaign actions (Dommett and Power Reference Dommett and Power2023; Power and Mason Reference Power and Mason2023; Stuckelberger and Koedam Reference Stuckelberger and Koedam2022), determining the entire shape of a campaign is complex. To generate this insight, we exploit a hitherto unexamined aspect of a well-recognized data source – electoral transparency finance disclosures within the UK. Unlike many other countries, within the UK campaigns are required to upload a copy of an invoice when making spending declarations over £200, providing a unique source by which to examine what money is spent on. Presently unexplored within studies of campaign finance, by hand-coding the activities described in these invoices, we generate new data on the type and prevalence of different campaign activities, allowing us to characterize the nature of modern campaigns.
This article makes several contributions in support of understanding the nature of current election campaigns. First, methodologically, we offer a dataset that provides new insight into the campaign activity paid for during election periods, and a template for analysis of future and past campaigns, both within and beyond the UK. Second, empirically, we provide unprecedented comparative insight into the nature of modern campaigns conducted by UK parties, revealing the unique configurations of campaign activity different parties invested in during the 2019 general election. This provides an important counterpoint to characterizations based on campaigners' own depictions of campaign practice and reveals that whilst data-driven techniques are evident, they are by no means dominant and are being adopted by different parties to different degrees. Third, conceptually, our study also has implications for the analysis of parties and party campaigning more generally, as it indicates the value of resisting the well-established tendency to focus almost exclusively on new or innovative campaign practices and instead suggests that scholars should pursue a more holistic approach that overtly recognizes the evolution and blending of campaign practice. Cumulatively, this study seeks to reshape the way in which scholars within and beyond the UK approach the study of campaigns, showing the potential of a new methodological and analytical approach able to more acutely capture modern practice.
Literature review
The idea that political parties are engaged in data-driven campaigning has become an established feature of recent commentary on party organization and election campaign practice. It is widely contended that ‘modern technologies have fundamentally altered the dynamics of modern campaigning … providing new ways to broadcast relevant political information, to influence voters’ attitudes and behaviour, to encourage campaign donations and to more precisely engage networks of potential supporters' (Bennett Reference Bennett2016: 264).Footnote 1 In particular, data-driven digital technologies are seen to have made campaign interventions ‘more efficient’ (Nickerson and Rogers Reference Nickerson and Rogers2014: 54). Emerging scholarship has traced the use of these tools within international campaigns (Kruschinski and Bene Reference Kruschinski and Bene2021; Silva et al. Reference Silva, Santos de Oliveira, Andreou, Vaz de Melo, Goga, Benevenuto and Huang2020) and, arising from this work, Andrea Römmele and Rachel Gibson (Reference Römmele and Gibson2020) have diagnosed the emergence of a new era of campaign practice, identifying a fourth era of data-driven campaigning. As they outline:
Our core contention is that there has been a profound shift in the nature of political communication during the past decade, which marks a step-change into a new era of political campaigning. This new fourth era we define as ‘data-driven’ and it is, we contend, characterized by shifts in four key areas of campaign practice. First, and most obviously in the infrastructure and tools that are used to fight the campaign. Digital technology and data are now hardwired into the campaign organization and operation. Second, parties have moved away from the top-down ‘point to mass’ use of mainstream media channels to embrace a more devolved and networked approach to voter communication. Third, there has been a reformulation of the targets for those messages, with a focus now on producing a much more fine-grained and personality-based understanding of the persuadable electorate. Finally, campaigns have now become far more internationalized in terms of the range of actors that seek to participate and influence the outcome. Both ‘real’ and automated external actors are now engaged in seeking to influence, in both orthodox and highly unorthodox or illegitimate ways. (Römmele and Gibson Reference Römmele and Gibson2020: 597)
This approach has well-established antecedents, and scholars have long identified ideal types in campaign practice. Of particular interest to this study, scholars in political communication have identified different ‘phases’ of political campaigning (Aagard Reference Aagaard2016; Blumler Reference Blumler2016; Römmele and Gibson Reference Römmele and Gibson2020), demarcating ‘partisan-centred’ campaigns (phase 1), ‘mass-centred’ campaigns (phase 2), ‘target-group centred’ campaigns (phase 3) and now ‘individual-centred’ or ‘data-driven’ campaigns (phase 4) (Kruschinski and Haller Reference Kruschinski and Haller2017). Noting the overlap between diagnoses of phases and eras, we use the term era hereafter.
To briefly summarize, the first era of campaigning is often associated with the period 1850–1960 and can be characterized as ‘partisan-centred’ (Janda and Colman Reference Janda and Colman1998). Depictions focus on the existence of large party memberships and an ethos of civic participation (Bale et al. Reference Bale, Webb and Poletti2019; Gibson and Römmele Reference Gibson and Römmele2009; Norris Reference Norris2002). Cited campaigning tools include meetings, rallies (Kavanagh Reference Kavanagh1970), canvassing activity, political merchandise (Huckfeldt and Sprague Reference Huckfeldt and Sprague1992; Wielhouwer Reference Wielhouwer1999) and print media, such as leaflets or posters. Melanie Magin et al. (Reference Magin, Podschuweit, Hassler and Russmann2017) also point to the significance of ‘radio broadcasts, and posters’.
The advent and wider availability of new technology – particularly television – ushered in the second era of campaigning. Seen to be evident between 1960 and 1990, or characterized as ‘mass-centred’, this era has been associated by scholars with televised party-election broadcasts (Scammell and Langer Reference Scammell and Langer2006) and/or televised advertising (Abrams and Settle Reference Abrams and Settle1977) (dependent on context), the use of polling (Jacobs and Shapiro Reference Jacobs and Shapiro1994; Norris Reference Norris2002), news advertisements (Kruschinski and Haller Reference Kruschinski and Haller2017; Magin et al. Reference Magin, Podschuweit, Hassler and Russmann2017: 1701), billboard advertising (Panagopoulos and Ha Reference Panagopoulos and Ha2015), the use of external companies for design services (i.e. of leaflets) (Wring Reference Wring1996) and paid direct mail (Bale et al. Reference Bale, Webb and Poletti2019; Norris Reference Norris2002).
The third era of campaigning is associated with practices between 1990 and 2008 and has been characterized as ‘target-group centred’. Magin et al. (Reference Magin, Podschuweit, Hassler and Russmann2017: 1701) therefore point to the use of ‘party and candidate websites, banner ads, and direct mailing by e-mail’. Other scholars have outlined the use of phonebanking (Pattie and Johnston Reference Pattie and Johnston2003) and campaign emails (Bowers-Brown and Gunter Reference Bowers‐Brown and Gunter2002). The growth of campaign consultancy is also associated with increased specialization and the displacement of party personnel in favour of professional strategists, and an emphasis on external expertise and skills such as media training and PR designed to ‘better package the parties’ message’ (Gibson and Römmele Reference Gibson and Römmele2001: 33; see also: Negrine et al. Reference Negrine, Holtz-Bacha and Papathanassopoulos2007). New research techniques such as focus groups (Moran Reference Moran2008) are also associated with this period (Norris Reference Norris2002).
The final era, seen to have emerged since 2008, is commonly described as ‘data-driven’ but is also characterized as ‘individual-centred’. We prefer the former term because whilst there has been a focus on the collection and analysis of individual-level data to enable fine-grained microtargeting, empirical evidence has shown a prevailing focus on broad group-based targeting with limited use of two or more datapoints to enable individualized targeting (Votta et al. Reference Votta2023). In discussing this era, scholars have focused on the use of online (Bowers-Brown and Gunter Reference Bowers‐Brown and Gunter2002) and social media advertising (Römmele and Gibson Reference Römmele and Gibson2020), mobile application services (Baldwin-Philippi Reference Baldwin-Philippi2019; Nadler et al. Reference Nadler, Crain and Donovan2018: 12), databases (Hersh Reference Hersh2015; Savigny Reference Savigny, Flinders, Gamble, Hay and Kenny2009), data analytics (Kefford et al. Reference Kefford, Dommett, Baldwin-Philippi, Bannerman, Dobber, Kruschinski, Kruikemeier and Rzepecki2022; Römmele and Gibson Reference Römmele and Gibson2020; Simon Reference Simon2019) and message testing (Baldwin-Philippi Reference Baldwin-Philippi2019; McKelvey and Piebiak Reference McKelvey, Piebiak, Lalancette, Raynauld and Crandall2019: 10).
This ‘ideal-type’ approach to conceptualization and analysis intends, as Max Weber articulated, to demarcate a ‘utopia which cannot be found empirically anywhere in reality’ (Reference Weber, Schils and Finch1949: 90). In this way, for Matthijs Bogaards the ‘advantage of the ideal type is that it captures the full richness’ of a condition, such that ‘empirical phenomena will never fully correspond to it, nor should they be expected to do so’ (Reference Bogaards2000: 397–398). Ideal types should therefore be understood as ‘almost by definition crafted as “pure” concepts … not to reflect reality, but as heuristics that facilitate the analysis of complexity’ (Flinders et al. Reference Flinders, Judge, Rhodes and Vatter2022: 361). From this perspective, ideal types help scholars to organize understanding of complex and rapidly evolving contexts, and have been widely used to help us understand campaign activities. In this case, placing campaign techniques into ideal types defined by specific eras – related largely to technological developments – helps us understand key features of campaigns in each era that are new and distinct from the last. Consequently, our means of understanding campaigns can be adapted to include features drawn from the ideal types.
Despite the ubiquity of this approach, this mode of analysis has been criticized for becoming easily abstracted from any practical reality (Russell and Serban Reference Russell and Serban2022) and for creating an impression of ‘faux precision’ (Russell and Serban Reference Russell and Serban2021: 761) by drawing attention to some features whilst overlooking others. For scholars of party communication, it is well recognized, as Magin et al. contend, that actual campaigns ‘will hardly ever meet these ideal types. Rather, each campaign is an amalgamation of all campaign practices available at that time. The exact mixture of approaches will depend on what a campaign targets, who it addresses, and the relative importance it attaches to certain functions’ (Magin et al. Reference Magin, Podschuweit, Hassler and Russmann2017: 1701).
These arguments are of particular interest in the context of recent scholarship on data-driven campaigning, as within this work attention has focused almost exclusively on attempts to identify and characterize new campaign practice. In this way, Römmele and Gibson's four components focus on previously unseen attributes, whilst others have highlighted the availability of new forms of individualized digital data, or focus on more sophisticated forms of analytics (Gorton Reference Gorton2016). Often absent from such discussion – perhaps in the rush to diagnose novel ideal types – is, however, reflection on how these new features are used alongside previous practices. Whilst some have acknowledged the evolution of many of these techniques (Kusche Reference Kusche2020), it is somewhat rare for scholars to reflect on how, and to what extent, new and old techniques are being entwined within modern campaign assemblages (Nielsen Reference Nielsen2012). We argue this tendency is detrimental to our understanding of modern campaign practice and, considering what it obscures, advocate for a more holistic approach to the study of campaigns. To do so, we present a new dataset on modern campaign practice and ask, first, what evidence there is of a new campaign era, before turning to consider what a focus on such practices alone obscures.
Methods
The process of using financial disclosures is a well-established method of studying party activity and assessing the relative resources and behaviour of different party organizations. Previous studies have used financial data as a proxy to examine the importance that campaigns have attached to broadcast advertising (e.g. Ridout et al. Reference Ridout, Franz, Goldstein and Feltus2012) and to reveal the target districts campaigns identify as most important to their success (e.g. Fieldhouse and Cutts Reference Fieldhouse and Cutts2009). However, there is a relative dearth of work that analyses official campaign spend data to examine the resources that parties dedicate to different campaign activities – and the work that does this tends to be largely US-focused (Limbocker and You Reference Limbocker and You2020; Sheingate et al. Reference Sheingate, Scharf and Delahanty2022). This is because much publicly available data, even in countries with the most transparent financial reporting systems, are often insufficiently detailed to allow differentiation of spending by campaign activity. Within the UK, however, there is a requirement for all political parties to provide a copy of the invoice for any service supplied during the election period over £200. This invoice is uploaded to the Electoral Commission political finance database and can be accessed by any individual via the Electoral Commission's website. This requirement is not unique to the UK, and similarly permissive arrangements can also be found in Brazil and Mexico.
Within this article, we exploit this resource to gain unique insight into campaign activity in the UK. As presently reported, the political finance database currently disaggregates spending information about political parties under 10 categories. Using headings such as ‘advertising’, ‘unsolicited material to electors’ and ‘transport’, it gives an overview of the distribution of party spend. Whilst providing a useful barometer against which to monitor change over time, this resource is unable to offer granular detail into specific activities (see Dommett and Power Reference Dommett and Power2023). For example, the broad category ‘advertising’ tells us very little about the specific type of advertising taking place (e.g. on a social media platform, online or on a billboard). Likewise, a whole range of activities could fall under the banner of ‘overheads and general administration’, or ‘unsolicited material to electors’. Legislation, however, requires parties to upload invoices against each item of campaign spend of £200 or over. Accordingly, it is possible to open and review specific invoices, and to use these data to generate more information about the exact type of activity that a party is spending money on. Taking this approach, we inductively coded campaign activity into different ‘era’ activities, allowing us to isolate fourth-era activity and hence address our research questions:
• To what extent is there evidence of a fourth era of data-driven campaigning in the UK?
• Are there differences between parties' adoption of different campaign-era practices?
Our analysis focuses on the spending returns of national political parties at the 2019 general election. Electoral Commission returns contained a total of 22,720 separate items of expenditure declared, covering 6,396 invoices. We first conducted a sift which included only suppliers on which over £1,000 had been spent. This reduced the number of invoices to 5,770 whilst allowing for the analysis of £49.9m out of an overall £50m party spend at the election. To code these data, the authors of this article formed a coding team, opening each available invoice and inductively coding the activity described in the invoice. Engaging in blind double-coding, the team initially coded small samples of invoices to develop a set of categories and to ensure consistency of approach. Through a series of weekly meetings, coders would individually code and then compare and discuss allocated codes, working iteratively to produce a standard set of 50 eventual codes (9 main codes and 41 subcodes nested under these) that were used to classify the entire database (Table 1).
Table 1. Codes Identified within Database and Associated Description
Note: *Discrete items reported under this heading are included in the analysis below. Bold indicates main heading, with those following the subheadings in that category.
Simple rules for coding were established, such as conducting exhaustive coding (i.e. coding each separate item mentioned in an invoice separately – meaning multiple codes could be assigned for one invoice), non-duplicative coding (i.e. not assigning the same activity within an invoice more than one code) and dealing with ambiguous invoices (i.e. creating a series of subcodes to differentiate invoices that could not be coded for different reasons, such as a lack of clarity about service and invoices that appeared blank) (Power et al. Reference Power, Dommett, Macintyre and Barclay2023). Each invoice was opened and coded by two coders. To check inter-coder reliability, we allocated approximately 20% of each coder's invoices to another team member to measure consistency. The Cohen's Kappa score for each pair of coders was at no point below К = 0.709, indicating a high degree of internal reliability. Our data are made up of a total of 1,006 separate suppliers working for (at least) one of the political parties in the 2019 general election, with £49,904,074 of spend by these actors. Of this dataset, we were able to assign 86.2% to one of our 49 substantive categories, with 13.8% being coded as completely unclear (for a more detailed discussion of the implications of this finding for transparency see Dommett and Power Reference Dommett and Power2022; Power et al. Reference Power, Dommett, Macintyre and Barclay2023).Footnote 2
Conducting an ideal-type analysis of these data
Our dataset offers a wealth of information about the nature and focus of modern campaign activity, allowing analysis of the suppliers of campaign services and the expenditure on different campaign activities at an aggregate or individual party level. To answer our research questions, within this piece we deployed an ideal-typical analysis in which we consider each activity as associated with a certain era within the literature on the development of campaign practice. This allows us both to look at the degree to which new campaign techniques are in evidence, and to consider the insights to be gained from a more holistic approach that examines new practices relative to previous campaign eras. To do this we compiled a list of attributes and activities outlined within the existing literature précised above and found many instances where clear connections could be made between our categories and each particular era of campaigning (Kruschinski and Haller Reference Kruschinski and Haller2017: 5).
For other categories, however, such alignment was less clear-cut. To code these activities, we reviewed a wider range of literature (beyond that which focused on eras of campaigning) to find evidence of the use of each campaign activity. Rather than trying to identify when a particular activity was first used in this wider descriptive literature, we looked for evidence that a practice had been widely adopted and hence become associated with an era. This meant that ‘focus groups’, for example, were coded as 3rd era (1990–2008), despite being first employed in the 1960s (Moran Reference Moran2008). Several of our activities were not clearly connected to a particular era within existing literature; in these instances we used inductive coding to assign each activity to a different era. The full list of spending categories and their associated eras is detailed in Table 2.
Table 2. Classification of Codes Associated with Each Campaign Era
It is important to note that this coding approach sees each era as distinct, and this may raise some questions for those familiar with the idea of data-driven campaigning. This latest era of campaign activity often relies on the application of analytic tools to previous forms of campaign activity. For example, a multilevel regression and poststratification (MRP) model can be applied to Voter ID gathered via doorstep canvassing, survey cards or polling – activities each associated with earlier eras. While defining 4th era as a distinct category, we acknowledge this interaction and hence coded each invoice exhaustively to capture any evidence of analytics and data use occurring in relation to long-established campaign techniques. At times we expect that certain invoices simply fail to provide detail of data-driven campaign activity because of a lack of prescription on what such invoices have to include. This limitation (of both policy and strict ideal-type analysis) is something we return to discuss later in this piece.
Findings
Evidence of a new ideal type in the UK?
Given the prevalent focus in much recent analysis on the emergence of a new era of campaigning, we first addressed our research question: ‘To what extent is there evidence of a fourth era of data-driven campaigning in the UK?’ To do so we searched our database for evidence of invoices declaring 4th-era campaign activity. In total, we identified 47 suppliers delivering services in this space, and a spend of just over £9m (around 18% of the total spend).
Looking in more detail at the kind of activity being classified as 4th era, Table 3 shows the most common practices at the 2019 general election were social media and online advertising – accounting for 82.3% of total 4th-era spend, with this service purchased through eight and six suppliers, respectively. In addition, we can see six suppliers, and £163,731 being spent on data and infrastructure, with campaign databases also the fourth most prominent category associated with this era. There is also some, if somewhat minimal, expenditure on data services and message testing.
Table 3. Evidence of Fourth-Era Campaign Spending
* Notes: There are 42 completely unique suppliers, but some of these suppliers are coded in multiple categories, hence we report the data on unique suppliers in each category.
Percentages have been rounded to the nearest decimal point.
Studying the invoices associated with this era of campaigning in more detail, we found evidence of many developments discussed within the existing literature. In terms of social media, we can see significant spend on Facebook, but also on Instagram and Snapchat. The invoices go into considerably different levels of detail about what was spent, with some providing detailed breakdowns of the topic and spend devoted to particular ads and the number of impressions gained, whilst others provide barely any information. The invoices reveal some evidence of personalized, or at least group-based targeting, with some containing targeting details for Facebook adverts. An invoice to Plaid Cymru, for example, showed a ‘Get out the Vote’ message to be targeted at 18–40-year-olds. An invoice to Reform UK similarly shows the party running ‘Swing the Vote’ campaigns, targeted to lookalike audiences (a term which refers to the capacity to serve adverts to social media users who share characteristics within a particular group (Meta n.d.)) in the top 36 constituencies.
The invoices also reveal more about the specific types of data and infrastructure parties were investing in. Indeed, looking at invoices coded under this category, we can see one invoice to the Conservative Party from the company App Dynamics for just over £28.7k, whilst an invoice to the Labour Party from Tangent showed a spend of around £35k on a ‘Polling Day app’. Similarly, most parties were invoiced for work to maintain their own database – such as Foresight, who invoiced for services to support the Conservatives' Vote Source database, the Women's Equality Party, who paid ECanvasser for CRM system services, and the Liberal Democrats, who made a payment to NGP Van for their database services. We also found one from DataDat, showing a company in Estonia to be providing ‘bot-builder online software’ to the Labour Party.
Elsewhere, invoices showed evidence of investment in message testing. The Labour Party paid Data Praxis £4.6k for conference message development and testing, and £2.7k for general election message development and testing. It also paid DRG a total of just over £83k for services that included ‘10 message testing focus groups’. Meanwhile, Reform UK paid COR Research around £39k for ‘Daily Message Testing’, whilst the Conservative Party paid System1 Research just over £11k for a ‘Boris Speech Test’. Cumulatively, the invoices we designated as 4th era revealed evidence of the kind of practices often cited in the growing literature on this topic.
These findings suggest some evidence of data-driven campaigning, but it is important before drawing a clear conclusion to consider our second research question, about the differences between parties. For this reason, we turn our attention away from the aggregate data that report spending from all parties, to consider the extent to which different parties invest in the 4th-era activities we identified within our dataset. Specifically, we examined the amount of expenditure each party in our dataset devoted to each of the categories we labelled as indicative of 4th-era activity. Adopting this approach, we find variation between parties in terms of their investment, suggesting that the indicators detected above may not be present within each and every party.
As shown in Figure 1, larger parties (Labour, Conservatives, Liberal Democrats) are more inclined to invest in a broader range of activity than smaller parties (Greens, Scottish National Party (SNP), Reform UK). Investing in social media advertising is the dominant form of expenditure, but this is particularly true for the Greens, for whom it makes up over 90% of their 4th-era spend.
Figure 1. Spending on 4th-Era Categories.
By contrast, larger parties engage in a more diverse range of techniques which fall under the 4th-era umbrella, with the three main parties also utilizing a suite of data and digital services in addition to the money they give to social media and other online platforms which provide them with advertising space. For example, we have already seen that they use centralized campaign databases to store and operationalize the data they collect about voters, but they also commission data and data-consultancy services to augment their database in a way which smaller parties generally do not. For instance, the Conservatives spent over £130k on ‘data strategy' provided by Uplifting Data. Similarly, we also found that both Labour and the Conservatives invest significantly in testing and developing their messages before using them during the campaign. Interestingly, we find Reform UK is an exception in that it is a smaller party which engages in a relatively large number of 4th-era activities (albeit fewer than the main parties). However, it tends to invest in those practices which relate to digital campaigning, such as through social media consultancy and message testing, rather than operating a substantial data infrastructure.
Based on this review, there is certainly some evidence to support the idea that campaigning in the UK is data-driven. Our data do, however, suggest that parties are not adopting exactly the same practices associated with the 4th era. Whilst we have evidence that all parties are devoting budget to social media advertising, only the larger parties appear to simultaneously spend money on databases and data infrastructure used to store the electoral roll, Voter ID information and/or membership data (Dommett et al. Reference Dommett, Kefford and Kruschinski2024: ch. 5). On this evidence, there appear to be data-driven elements to UK elections, but not all parties are investing in the various practices associated with the 4th era.
In drawing this conclusion it is, however, not clear how far activities associated with data-driven campaigning have supplanted previous practices, or whether (and to what extent) they have been integrated into campaigns. It may, for example, be that the expenditure associated with these headings is dwarfed by spending on other, earlier campaign activities, raising questions about the degree to which parties are truly investing in these ideas and practices. In light of this idea, we now turn to discuss the insights offered by our wider database, presenting a more holistic analysis that explores the way in which campaign activities from different eras are blended, and the alternative strategies deployed by different parties.
The value of a more holistic approach
Using our dataset to understand the shape of modern campaigns, we begin by providing a descriptive overview of the relative prominence of our four eras. We discuss the specific activities associated with each era, and the party-level differences these data reveal. Exploring our dataset, we suggest that an understanding of novel practice becomes particularly meaningful when it is contextualized.
In Table 4 we report both the number of suppliers and the total amount of expenditure associated with each of our four eras. Interestingly, this reveals that most suppliers and party expenditures are primarily centred around traditional campaign practices – especially those associated with 1st- and 2nd-era activities. Indeed, we can see that 62% of expenditure was devoted to these practices. Significantly for our focus, it also becomes apparent that when it comes to 4th-era activities, relative to previous eras the number of suppliers and spending is far lower. Indeed, we see just 18.2% of campaign spending devoted to the 4th era, a figure overshadowed by the 36.2% spent on 2nd era, and the 25.8% expended on 1st era.
Table 4. Number of Suppliers and Expenditure by Era for All Parties
* Notes: This figure differs from the 47 given above as in Table 1 we report the number of unique suppliers within each subcategory coded as 4th era, whereas this focuses on unique suppliers under any 4th-era category.
Some suppliers performed services from multiple eras, or submitted some unclear invoices, so the ‘% of suppliers’ does not come to 100%.
Digging into these data first at an aggregate level, we explore the types and prominence of each of the activities associated with each era. Taking the 1st era, we identified nine subcategories (see Table 2). The vast majority of expenditure associated with this category fell under campaign material printing, which accounted for 70.3% of activity associated with the era. Looking at the invoices, indicative examples of what was captured under this heading included leaflets, brochures and party-branded correx boards, whilst under the separate category of merchandise we found spending on umbrellas, beer mats and Boris Johnson boxing gloves.
Second-era activity was associated with 14 subcategories. Looking at the distribution of spend across these categories we see paid leaflet delivery (59%) and polling (12.6%) as the dominant services that parties spent money on in this era. Under these headings, we see the use of national delivery services such as Whistl, alongside the altogether more localized South Devon Leaflet Distribution. We can see the same pattern in the polling services provided, with well-known research organizations such as YouGov and CTF Partners operating alongside individuals such as Phillip Myers conducting phone polling for the Sheffield Liberal Democrats.
For the 3rd era, we identified 15 categories. Looking at the distribution of spend we see a slightly more even distribution, with IT infrastructure and support (25.4%), consultancy (20.3%) and communication consultants (13.1%) the top three subcategories. IT infrastructure and support covers services such as the Labour Party's monthly platform fee for Promote (via Experian), digital support provided by Topham Guerin to the Conservatives, and threat-monitoring services and firewalls such as that provided by UKFast. Consulting, similarly to polling, runs the gamut from firms such as Rossbury Associates to (seeming) freelance individuals like S.B. Howell.
In addition to spend, we also examine the number of unique suppliers providing services associated with each era. Here, we find there to be a larger number of suppliers for activities within the earlier eras, and fewer suppliers for later eras. Indeed, we only detect 42 different suppliers which provide services in the latest era, compared to the 331 which provide 1st-era services. In practice, this means that different parties often rely on the same supplier to deliver 4th-era activities (notably social media companies), whereas in earlier eras parties rarely worked with the same supplier.
Exploring party-level variation
Returning to our second research question, we again turn to look at party-level variation in campaign practice. We can see first that there are large differences between parties in terms of their overall levels of campaign spend. Table 5 shows this for the six highest-spending parties. Our data reveal that the Conservatives operate the largest campaign budget (£16.4 million), whereas the Liberal Democrats (£14.3 million) and Labour Party (£12.2 million) also spend relatively highly. Conversely, the SNP (£1.0 million), Greens (£0.4 million) and Reform UK (£5.0 million) spent much less. This is key contextual information when considering how parties blend differing activities within their campaigns. It is likely, for example, that parties with higher budgets will be more able to invest beyond a smaller number of ‘core’ activities which are essential to all campaigns, irrespective of their resources.
Table 5. Total Campaign Spend by Party
Looking next at the blending of activities from different eras, we mapped party-level variation in the distribution of spending across our broad categories (Figure 2). Beginning with the three main parties depicted on the left of each era depicted in the diagram (Conservatives, Labour and Liberal Democrats), we detect that each invests heavily (between 30% and 45% of total spend) on 2nd-era activity, but they have different priorities with regards to spend in other eras. Notably, only Labour dedicates a large proportion of its spend to the 4th era. We observe similar asymmetries when examining smaller parties (depicted on the right of each era within Figure 2). The SNP spends a large majority of its overall budget on 1st-era activity, far more than any other party, and although the Greens and Reform UK each spend around 20–25% of their total on 4th-era activities, they display very different patterns when it comes to earlier forms of campaign activity.
Figure 2. Proportion of Spend by Era.
Exploring party-level differences further, we can examine the degree to which different parties deploy each of the campaign activities associated with each era.
In terms of activity in the 1st era, Figure 3 shows that whilst the printing of campaigning materials represents the lion's share of spending for each party, there are differences between parties in terms of the degree to which this is the dominant form of expenditure. For example, almost all of the Green Party and Liberal Democrat spend in this era was on campaign material printing. Conversely, the main two parties both spent around a quarter of their 1st-era budget on event costs, as did Reform UK, which spent only slightly less on events than it did on printing campaign literature (36%). Beyond these two categories – as we highlighted above – spending on other activities at the 1st-era level tended to be small.
Figure 3. Spending on 1st-Era Categories.
Turning to 2nd-era spend distribution by party (Figure 4), we find that all parties spend more on leaflet delivery than they do on any other category bar the SNP, whose largest outlay was £78,000 on billboards and other forms of public advertising. However, there is again variation in the degree to which this is the dominant mode of expenditure. The Liberal Democrats invest in almost all of the categories presented here, but £4 out of every £5 comes in the form of delivering campaign materials. Conversely, the Conservative and Labour parties spend between 40% and 50% respectively on leaflet delivery, whilst at the same time spending significantly more on paid advertising (Labour) and research, particularly through polling (the Conservatives). Once again, the Greens and the SNP are notable by not investing at all in a larger number of categories, suggesting that they are running more streamlined campaigns.
Figure 4. Spending on 2nd-Era Categories.
When it comes to 3rd-era categories, there is less of a pattern of parties coalescing in spending most of their resources on the same activities (Figure 5). It does seem that the larger parties, particularly the Labour Party, are considerably more likely to spend on IT infrastructure than other parties, who either spend nothing (Reform UK, the SNP, the Greens) or much smaller sums of money (the Liberal Democrats). Smaller parties, on the other hand, are more likely to invest resources in consultancy services, but here it is important to consider the context of the low levels of spend by such parties in this era. For example, whilst it is true that the Green Party concentrates a large majority of its spending amongst these categories on consultancy, this in fact represents only just over £3,000 to a single consultant. We also find that smaller parties are less likely to invest any money in those categories which facilitate individualized voter contact, such as phonebanking and Ordnance Survey street-level data.
Figure 5. Spending on 3rd-Era Categories.
Analysing the suppliers
In addition to exploring the distribution of expenditure, we can also look at the number of suppliers used by parties within each era. As shown in Table 6, there are once again differences between larger and smaller parties, with the Conservatives, Labour and the Liberal Democrats each working with suppliers numbering in the hundreds, and Reform UK, the Green Party and SNP working with relatively few. However, we also observe differences between the larger parties in the number of suppliers that they work with. For instance, the Liberal Democrats use more suppliers within the 1st era (141) than the Conservative Party uses in total across all eras (134). This may reflect each party's relative campaign structure, with the Liberal Democrats operating a more decentralized campaign than the Conservatives, reflecting their organizationally federal party structure. Indeed, we find some evidence for this idea when looking at invoices for the printing and distribution of campaign materials, as the Conservatives use eight suppliers for printing their campaign literature and six for its delivery, whereas the Liberal Democrats use 101 and 29 suppliers respectively for the same categories, including many examples of printing companies operating exclusively at the local level.
Table 6. Number of Suppliers Used by Each Political Party for Activity Associated with Each Era
Note: Numbers in parentheses indicate the percentage of the total n of suppliers within a given era.
Alongside these differences, we also find examples of parties being similar to one another. This is particularly true when looking at the proportion of suppliers that they use within each era. All parties use relatively few suppliers in the 4th era, whereas between 41% and 48% of suppliers for all parties (bar the Conservatives) provide 1st-era services. However, we find there to be some differences between the larger and smaller parties in terms of the proportion of suppliers that they use from the 3rd era. Reform UK, the Greens and the SNP are all in the single figures for suppliers in this era, representing at most 14% of their total suppliers (SNP). Conversely, at least 21% of the suppliers used by the larger parties can be found in the 3rd era, and for the Conservatives this is only just fewer than two in five (39%). This tendency predominantly reflects what we show in Figure 5, where the smaller parties run skeleton campaigns which involve trivial or no expenditure in categories relating to their IT infrastructure, website and telecommunications.
Overall, our data reveal there to be a small number of core activities that all parties invest heavily in, certainly in terms of printing and delivering campaign materials. However, we also detect differences in how parties spend outside of these areas. In particular, we find that larger parties invest in a broader suite of services than small parties do. Labour, the Conservatives and the Liberal Democrats spend at least some money in a large majority of categories. By contrast, the smaller parties tend to concentrate their resources within a narrower range of activities. This is most obviously the case for the Green Party, who we found only spent in 14 categories, which is considerably fewer than half of the number of categories that the Liberal Democrats were active in (39). Importantly, these findings apply to each of our four eras, suggesting that the trends found in the 4th era mirror activity within the wider campaign.
Discussion
Collectively, these data reveal important nuances in our understanding of modern campaigns. Significantly, it helps to reveal variation not only in the extent to which different parties draw on campaign activities associated with different eras, but also in the precise activities they decide to invest in, and in their relationship with external suppliers. Most importantly for our analysis, these insights reveal the very different ways in which new tools are being integrated into modern campaign activities. It shows that, in actuality, 4th-era activity is relatively sparse – both in terms of the number of suppliers and expenditure. We can also see that the adoption patterns found in relation to 4th-era practices mirror practices associated with earlier eras. Indeed, we have shown that certain smaller parties are likely to invest heavily in one or two types of campaign activity, whilst other – primarily larger parties – invest in multiple activities associated with each era. Such findings suggest there are material constraints affecting the way in which data-driven tools are being adopted.
Some work has already suggested that available resources may constrain the use of data-driven tools (Dommett et al. Reference Dommett, Kefford and Kruschinski2024; Kruschinski and Haller Reference Kruschinski and Haller2017). We know (see Fowler et al. Reference Fowler, Franz, Martin, Peskowitz and Ridout2021) that some 4th-era practices involve the use of relatively cheap – and effectively off-the-shelf – services (like Facebook advertising), whilst others are considerably more expensive and require the purchase (and construction) of complex data infrastructures. We think parties' use of these tools reflects resource profligacy versus resource efficiency. Bigger and more established parties have a large disposable income, such that they can simply afford to spend on a wide array of activities – inclusive of those in the latter eras. Less well-resourced parties are somewhat more careful and conduct directed spending on latter-era activities they consider the most effective (in terms of outcome) and efficient (in terms of spend).
This chimes with existing work on campaign spending which shows considerable variation in campaign effectiveness and focus (Denver and Hands Reference Denver and Hands1997; Fieldhouse et al. Reference Fieldhouse, Fisher and Cutts2020; Fisher et al. Reference Fisher, Cutts, Fieldhouse and Rottweiler2019). However, our findings hint that it may not only be money that affects the form of a campaign. The Labour Party, for example, was notable in spending less money on paid leaflet delivery than the Conservatives and Liberal Democrats – something that is likely a product of their more extensive activist base compared to other parties, which may need to pay to distribute their campaign materials (see, for example, Bale et al. Reference Bale, Webb and Poletti2019). It also might be related to the less strategic approach the Conservatives have been shown to take at elections (Fieldhouse et al. Reference Fieldhouse, Fisher and Cutts2020; Fieldhouse and Cutts Reference Fieldhouse and Cutts2009), where they simply spend money where they have the resources. This is as opposed to the Liberal Democrats, who focus their fire on specific constituencies where they have strong local support, and by association electoral prospects (Russell and Fieldhouse Reference Russell and Fieldhouse2005).
Our findings also hint at possible alternative strategies in the use of campaign tools. The Green Party and Conservative Party, for example, spend similar percentages on 4th-era activities, but focus their money on different services. Whilst the Greens concentrated on social media advertising, but showed little evidence of other forms of online advertising, the Conservatives spent on both these advertising forms almost equally. Such examples suggest that parties are deciding to invest in different activities, making different strategic calculations about where to deploy their available budgets.
Taking our findings together and applying them to understand recent trends in data-driven campaigning, we do find evidence of it in the UK, but fundamentally we show that this form of activity does not define campaigning, but is rather a supplement to existing practices. Moreover, we can see that certain parties are more invested than others, and that some – especially the Conservatives, Liberal Democrats and SNP – invest a lower proportion of their spending on activities associated with this era. These conclusions demonstrate the value of a holistic analysis and suggest that in order to understand modern campaigning it is important not to focus in isolation on the presence of the new, but rather to consider campaigns as assemblages that draw on different eras (differently). Such analysis has the potential to reveal the presence of different ‘types’ of campaign activity, but also the considerable diversity that can characterize the modern campaign.
Perhaps most importantly, our analysis raises questions about the prevailing tendency to focus on new practices and ideal types in isolation. Indeed, our analysis suggests that no single ideal type captures the reality of how parties in the UK are conducting their campaigns. Even allowing for Bogaards's (Reference Bogaards2000) contention that ideal types exist to capture the ‘full richness’ of an idea which empirical reality is unlikely to achieve in practice, our findings still suggest that British campaigns cannot be fairly summarized as representing a particular era. Although we recognize the incentive to typologize new practice, and appreciate the value of an ideal type as a metric of analysis, we argue that it is important to pursue alternative forms of analysis. In this we follow Magin et al. (Reference Magin, Podschuweit, Hassler and Russmann2017) and call for research both in the UK and elsewhere that explores how these eras are amalgamated to construct unique campaign configurations.
The analysis we have advanced and the dataset we present facilitate this analysis – allowing us to appreciate the way in which campaigns are constructed by different parties and the constraints and strategic decisions that guide the form of a modern campaign. Such analysis is a vital supplement to an ideal-type approach, and hence we call for other scholars interested in the UK to draw on our resource, but also to extend and replicate this approach in other comparative contexts – most notably in Brazil and Mexico, where similar financial disclosure regimes exist. We therefore raise the possibility of future comparative work that is able to consider how systemic and regulatory contexts affect the form of modern campaigns.
In drawing these conclusions, it is important to note a number of limitations with our work that affect our findings. In seeking to explore the shape of modern election campaigns we exploited a hitherto unexamined aspect of financial disclosure. Whilst offering unprecedented insight, this source does not capture all campaign activity. Specifically, it does not account for unpaid activity, or that which costs below £200 (and hence was not invoiced), or spending that occurred outside the formal campaign period (when disclosure is not required to the Electoral Commission). Furthermore, campaigns and suppliers are given little instruction about what to include within invoices, meaning that disclosures may not always contain evidence of the exact type of campaign activity taking place. There is accordingly much we do not know about the campaign actions parties utilize and the degree to which unpaid or undisclosed activities mirror the trends we uncover. Recognizing this limitation, we have suggested elsewhere that there are several reforms that could be made to improve transparency (Power et al. Reference Power, Dommett, Macintyre and Barclay2023); nevertheless, we contend that our analysis offers the most comprehensive insight to date.
It should also be noted that our analysis focused on the 2019 general election and as such examines a particular context. This election was a ‘snap election’ called at short notice by the Conservatives and may therefore differ from other elections. In essence, when parties have more certainty about the electoral timetable – something the Fixed-Term Parliaments Act intended to deliver – they may construct their campaign differently from when an election needs to be fought at short notice. Parties operating at a snap election may, for example, have less capacity to experiment with data use, rendering them more likely to focus on campaign activities associated with earlier eras. Comparative analysis is required to facilitate further insight, a task that will be essential for future research to verify our claims.
That said, in revealing the shape of modern campaigning and showing the degree to which parties draw on long-established campaign mechanisms, our findings challenge the tendency to concentrate on the most novel campaign techniques. Rather, they show the importance of considering how parties blend and combine different tools, an approach that we argue has value for scholarship both within and beyond the UK.
Conclusion
In this study, we set out to answer the questions: ‘To what extent is there evidence of a 4th era of data-driven campaigning in the UK?’ and ‘Are there differences between parties’ adoption of different campaign-era practices?’ Presenting a new analysis of financial transparency data, we found some evidence of data-driven campaigning but showed that parties are not adopting exactly the same practices associated with the 4th era. Whilst all parties devote budget to social media advertising, only the larger parties simultaneously spend money on databases and data infrastructure.
More fundamentally, however, our article questioned the value of focusing solely on ‘novel’ practices. Exploring our new dataset, we have shown the ongoing importance of activities associated with earlier eras and have revealed areas of commonality and divergence in how parties campaign. In particular, we have shown there is considerable spend on leaflet production, merchandise, postal delivery and polling. The precise configuration of spend associated with each era is not, however, uniform, suggesting that each campaign exercises a degree of agency over where and how to invest resource.
Offering these findings, this article provides hitherto unprecedented insight into the nature of modern campaigning and has revealed the different configurations of parties' campaign activity. These findings have interesting implications for broader discussions of campaign effectiveness and strategy. Whilst media coverage often focuses on new technology, it is not clear whether parties' uneven adoption of these techniques reflects their view of perceived effectiveness. Is data-driven campaigning equally effective for all parties, and is a lack of usage a consequence of party-level attributes or other concerns? Although some attempts have been made to address such questions (Dommett et al. Reference Dommett, Kefford and Kruschinski2024), further analysis of practice and perceived effectiveness is required.
In prompting such questions, we challenge the prevailing tendency to conduct ideal-type analysis, and have instead shown the value of a more contextualized, holistic approach. In doing so, we have not sought to dismiss idealized studies, but have instead suggested that there is a need to go beyond them, exploring different ways of thinking which allow a more rounded study of campaigns. To facilitate these efforts, we offer our own dataset for investigation by other scholars and call for future studies to conduct similar work via comparative datasets, and to extend our own through historical and future analysis of electoral transparency data yet to come.
Data availability statement
The dataset and codebook for this article can be found at: Katharine Dommett, Sam Power, Andrew Barclay and Amber McIntyre (2023), Codebook and Data: 2019 General Election Party Expenditure Coding (University of Sheffield), dataset, https://doi.org/10.15131/shef.data.24884940.v1.
Acknowledgements
We would like to thank the reviewers for their valuable and highly constructive feedback on this article as well as the editors for an excellent production experience. In addition, we would like to thank colleagues at the University of Manchester Democracy and Elections Group, Chris Butler and Paul Webb, for their feedback on previous versions of this article, as well as attendees of the 2022 annual conferences for the European Consortium for Political Research and the Elections, Public Opinion and Parties specialist group of the Political Studies Association. We would also like to acknowledge that this project was supported by the NORFACE Joint Research Programme on Democratic Governance in a Turbulent Age and co-funded by ESRC, FWF, NWO and the European Commission through Horizon 2020 under grant agreement no. 822166.
Open-access statement
For the purpose of open access, the author has applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising from this submission.
Open access
