The centrality of pledge awareness

A key precondition for pledge-based voting according to mandate logic is that voters are aware of the pledges parties make before an election and whether or not they have fulfilled them after the election. This is illustrated in Figure 1. Taking the lead from Thomson (Reference Thomson2011), a number of studies have investigated how many pledges voters are able to correctly identify as either kept or broken (eg Naurin and Oscarsson Reference Naurin and Oscarsson2017; Thomson and Brandenburg Reference Thomson and Brandenburg2019). The common approach in these studies on post-election awareness has been to ask respondents at the end of an election cycle (or at the beginning of the next) to recall whether a given pledge was fulfilled or not. The broad conclusion is that there is a positive correlation between actual pledge fulfillment and voters’ beliefs about fulfillment, but that interest in politics generally improves respondents’ ability to correctly evaluate the fulfillment status of pledges.

Figure 1.

The centrality of pledge awareness in mandate logic.

Figure 1. Long description

Flowchart illustrating the centrality of pledge awareness in mandate logic. The process begins with policy program-making, where a pledge is made. This leads to mandate giving, where a selection is made. Finally, the process moves to evaluating, where accountability is assessed. Pre-election pledge awareness influences the selection process, while post-election pledge awareness affects accountability. The flowchart highlights the importance of pledge awareness in the democratic process.

Their qualities aside, none of the existing studies explore pledge awareness before the election. Yet we argue that pre-election awareness is essential in the mandate policy chain because this knowledge can help voters keep track of a pledge’s fulfillment status after the election. We define pre-election pledge awareness as voters’ ability to perceive (ie being exposed to and notice), store (ie remember), and retrieve (ie to recognize or recall) election pledges that parties make as part of an election campaign (resembling Zaller’s classic conceptualization of political awareness, see Zaller Reference Zaller1992). Given individuals’ scarce attention and weak learning capacity for complex information (eg Graber Reference Graber2001; Simon Reference Simon1967), pledges do not easily enter a voter’s memory, but instead need to, for instance, receive a great deal of attention in the media, be highly relevant for the individual, or align with the individual’s pre-dispositions (eg partisan identity, see Campbell, Converse, Miller et al. Reference Campbell, Converse, Miller and Stokes1960). Crucially, voters must not only be aware of a pledge but also correctly assign it to the pledge’s sponsor to select a representative in line with the mandate logic (the left part of Figure 1). Yet when a pledge has been picked up by the voter and assigned to the correct sponsor, it can serve as an asset when holding parties accountable after the election (the right part of Figure 1). Specifically, we theorize two mechanisms for how pre-election pledge awareness becomes relevant for post-election pledge awareness.

First, pre-election awareness of a pledge and its sponsor enables post-election awareness because it means voters know what to look for in the post-election news stream (the importance of these so-called surveillance-policy facts has been highlighted by Barabas, Jerit, Pollock et al. Reference Barabas, Jerit, Pollock and Rainey2014). Similar to pre-election pledge awareness, post-election pledge awareness is difficult to achieve (perhaps even more so), as politics in general and pledges in particular receive less media coverage outside election campaigns (De León and Vermeer Reference De León and Vermeer2023; Falasca Reference Falasca2014; Müller Reference Müller2020; Peter and De Swert Reference Peter and De Swert2009). Having previously acquired information about a specific pledge gives voters an informational advantage during these less intense political periods (constituting a significant resource to these voters, for a similar argument, see Thomson Reference Thomson2011). Graber (Reference Graber1984) argues that citizens frequently use previous information (stored as so-called schemas) when they access, process, and store political news in their memories. If voters do not have a mental schema to draw upon when being exposed to a news story, it is less likely for this to be stored in their memory (Bøggild, Aarøe, and Petersen Reference Bøggild, Aarøe and Petersen2021; Graber Reference Graber1984; Aarøe and Petersen Reference Aarøe and Petersen2020). Using this argument, obtaining information about the fulfillment of a pledge should be more difficult when voters have not already stored information about the pledge. This logic becomes clear when the media covers the adoption of a new bill without making a reference to the pre-election pledge from which the policy originated. If the media only reports that ‘Party A implemented Policy B’, the voter needs to connect this policy to a pledge made by the party in the previous campaign to update its fulfillment status. Even when the media explicitly makes this connection, voters may fail to integrate the information into memory if they lack a cognitive schema to which the information can be meaningfully linked (Graber Reference Graber1984).

Second, pre-election awareness motivates post-election awareness if we assume that pre-election awareness reflects the importance that voters assign to a pledge, or at least that they find it noteworthy (on the importance of individual-level saliency and personal relevance for attention to political information, see Ryan and Ehlinger Reference Ryan and Ehlinger2023; Thomson Reference Thomson2011). The personal motivation that drives certain people to acquire information about a given pledge before the election can constitute the same drive that leads people to seek out and be attentive to information about the progress of pledge fulfillment after the election. When voters care about a pledge, they should also follow it more closely after the election. Of course, the media plays a key role as voters receive the most relevant political information from media sources (Norris Reference Norris2000; Strömbäck Reference Strömbäck2008), but voters might also search more actively for the status of a given pledge if they find it personally relevant (eg seeking information directly from party platforms, interest organizations, or friends and family). While the first mechanism emphasizes the importance of pre-election pledge awareness as a cognitive resource, the second mechanism highlights that voters might also be eager to know the fulfillment status of a pledge after having become aware of it during the campaign.

The overall argument presented here follows the same line of thought as Delli Carpini and Keeter (Reference Delli Carpini and Keeter1996), namely that informed voters are better citizens from a democratic perspective. Specifically, we have argued that voters with pledge-specific knowledge should, in short, be more capable of holding parties accountable for their promises at the next election. The problem of information scarcity that prevents people from learning about the fulfillment of a pledge can be reduced by information acquisition during the campaign period, when voters and the media follow politics more closely (Andersen, Tilley, and Heath Reference Andersen, Tilley and Heath2005; Gelman and King Reference Gelman and King1993). In this way, pre-election awareness may be considered a distinct source of post-election awareness above and beyond general interest in politics.

Guiding low-interest voters

We might, however, expect that pre-election awareness is most important for those voters who are not very interested in politics outside the intense period in the run-up to an election. These voters are more likely to experience information scarcity, as they do not monitor politics on a regular basis between elections and are therefore unlikely to know whether a pledge has been fulfilled or not.

Yet during elections, some pledges emerge as very salient or defining of the campaign, such that even voters with limited political interest absorb these messages. Alternatively, voters with a low interest in politics might become aware of specific pledges that relate to aspects of life that are important to them (similar to the argument of issue publics, see, for instance, Rossiter and Harden Reference Rossiter and Harden2025; Ryan and Ehlinger Reference Ryan and Ehlinger2023). For low-interest voters, knowledge about these pledges will be one of the few pieces of relevant political information they bring into the post-election period. The enabling and motivating mechanisms described above should thus be particularly activated for these types of voters, while voters with a high interest in politics are more likely to receive information about the fulfillment statuses of pledges, regardless of whether or not they became aware of the pledges before the election. Awareness of pledges, in short, may make otherwise disinterested citizens ‘participate intelligently in governmental affairs’ (Schudson Reference Schudson1998: 310).

Case selection and pledge collection

We study pledge awareness in the context of the 2019 Danish national election. Denmark is, in many ways, a typical Western European democracy with a proportional representative electoral system, which generates fierce multiparty competition. For the 20 years or so before the 2019 election, the two main contenders for government were the Liberal Party (the incumbent at the time of the election) and the Social Democrats (who ended up winning the election). Together, these two parties normally receive around 50per cent of all votes, implying that they are the natural providers of the prime minister but also that they depend on smaller coalition partners to win a majority. This is typically the situation across Europe, where single-party majorities are rare.

Since the Social Democrats and the Liberal Party have been the only prime minister parties in recent decades, we focused on voter awareness of pledges made by these two parties (previous studies also focus on government pledges when studying fulfillment awareness, see Naurin and Oscarsson Reference Naurin and Oscarsson2017; Thomson Reference Thomson2011; Thomson and Brandenburg Reference Thomson and Brandenburg2019). To identify real pledges made by these two parties in the run-up to the 2019 election, we searched campaign-related information on the parties’ webpages and Facebook profiles (spokespersons from both parties confirmed that these were the two main channels for announcing pledges). We used those political statements that met the conventional scholarly definition of a pledge as ‘a statement committing a party to one specific action or outcome that can be clearly determined to have occurred or not’ (Thomson, Royed, Naurin et al. Reference Thomson, Royed, Naurin, Artés, Costello, Ennser-Jedenastik and Praprotnik2017: 532).

In total, the Liberal Party and the Social Democrats issued 89 and 112 pledges, respectively. This aligns very well with what other parties tend to produce. According to Thomson, Royed, Naurin et al. (Reference Thomson, Royed, Naurin, Artés, Costello, Ennser-Jedenastik and Praprotnik2017), in 12 Western democracies and 57 elections, parties make an average of 105 pledges per election (with a standard deviation of 63). The Social Democrats ultimately won the election in 2019 and formed a single-party government. They kept 72 per cent of their election pledges (at least partially), which is higher than the approximately 60 per cent average pledge fulfillment rate among government parties (based on a study with 12 countries, see Thomson, Royed, Naurin et al. Reference Thomson, Royed, Naurin, Artés, Costello, Ennser-Jedenastik and Praprotnik2017). This difference might reflect that the Social Democrats led a single-party government (ibid.). We discuss this case selection and its influence on our results in the discussion section of the paper.

Three voter studies: pre-election study, panel study, and post-election study

Testing pledge awareness in the full chain of mandate theory is demanding as it requires a design setup that allows for studying voters’ information levels throughout the electoral cycle. To achieve this, our project included three studies: a panel study with three subgroups and two pooled cross-sectional studies (one was conducted pre-election and the other was conducted post-election). Figure 2 provides an overview of these three studies. The two pooled cross-sectional studies were developed to measure voters’ awareness of the full universe of pledges made by the Social Democrats (and the Liberal Party in the pre-election study) and used a crowdsourcing design with a selected number of pledges distributed to each respondent (to avoid survey fatigue). The panel study was developed to capture the individual-level importance of pre-election pledge awareness for post-election fulfillment awareness and focused on the four most central election pledges as identified during the pre-election study.

Figure 2.

Overview of the three voter studies.

Notes: In the online Appendix, we show the exact start and end dates of each survey round. The post-election study was conducted just before the subsequent election (November 1, 2022).

Figure 2. Long description

A timeline illustrating the structure of three voter studies conducted around an election. The timeline begins three months before the election campaign starts on May 7, 2019, and extends to forty months after the election held on June 5, 2019. The timeline is divided into three main phases: the pre-election study, the panel study, and the post-election study. The pre-election study includes Group 0 surveyed three months before the campaign starts. The panel study is further divided into Wave 1, Wave 2, and Wave 3. Wave 1 includes Group 1 surveyed four weeks before the election, Group 2 surveyed two weeks before the election, and Group 3 surveyed one week before the election. Wave 2 includes Group 1 surveyed one week after the election, Group 2 surveyed one month after the election, and Group 3 surveyed four months after the election. Wave 3 includes Group 1 surveyed nine months after the election, Group 2 surveyed twelve months after the election, and Group 3 surveyed eighteen months after the election. The post-election study includes Group 4 surveyed forty months after the election.

The panel setup was necessary to measure voter knowledge while the relevant political events were unfolding. Otherwise, we might underestimate voters’ information levels. If we ask voters about a pledge long after the electoral campaign, some voters might not be able to remember it, even though they were aware of it during the campaign (people frequently forget about pledges, as we show in the analysis). Given that many voters do not even recall whom they voted for (eg Belli, Traugott, Young et al. Reference Belli, Traugott, Young and McGonagle1999), this is a real source of error if unaccounted for. In other words, we are not interested in voters’ ability to recall the election campaign and the specific promises made long after the fact; rather, we focus on the level of pledge-specific knowledge when this information was most important to them (namely, right before they were to vote on election day).

Furthermore, establishing a relationship between voters’ ability to link a party to a pledge and their ability to correctly evaluate whether the pledge has been fulfilled might not be very surprising if voters are asked at the same point in time, while it is much less trivial to show that voters can use information that they obtained during an election campaign several months later when the pledge is being fulfilled. Related to this, an ambition of the panel study was also to track whether voters’ evaluations of pledge fulfillment changed relatively soon after a pledge was fulfilled. Here, it would be insufficient to only survey respondents at the end of the following election period, as the fulfillment of the pledge might have occurred at the beginning of the period.

To address these measurement challenges related to voters’ pledge-specific knowledge, we used the panel survey to interview voters before the election and to re-interview them several times after the election, up to one and a half years later (see Wave 1, Wave 2, and Wave 3 in Figure 2). Each wave consisted of three survey samples interviewed at different points in time within each wave (see Group 1, Group 2, and Group 3 in Figure 2). From the outset, each group included approximately 2,000 respondents (total N = 6,233) and was nationally representative of the electorate in terms of gender, age, education, and geographical location.Footnote ¹ Our panel setup resembles a so-called rolling re-interview design (Leeper and Slothuus Reference Leeper and Slothuus2016) and enables us to track voters’ knowledge about individual pledges and observe changes when a pledge is fulfilled (if fulfillment occurs during the observation period).

The pre-election and post-election studies included 2,000 and 2,089 respondents, respectively. In both studies, respondents were representative of the entire electorate (again, matching the population on age, gender, education, and geographical location).

Measuring pre-election pledge awareness

In the pre-election stage, our independent variable concerns voters’ awareness of campaign pledges. Based on the definition above, we need a measure that captures whether voters have heard and remembered real pledges made by political parties that they are then able to later recall. Simply asking respondents whether they have heard of a specific pledge might be problematic, as voters can easily respond affirmatively without truly having been aware of the pledge (ie intentionally or unintentionally lying). It would also be insufficient to experimentally treat respondents with information about a given pledge, as this would merely capture pledge exposure, while awareness requires that voters notice and store this information, which likely requires more persistent exposure in the news, for instance (several studies show how respondents rapidly forget about information that they receive in experiments, see Chong and Druckman Reference Chong and Druckman2010; Coppock Reference Coppock2017, Reference Coppock2021; De Vreese Reference De Vreese2004; Gerber, Gimpel, Green et al. Reference Gerber, Gimpel, Green and Shaw2011; Kalla and Broockman Reference Kalla and Broockman2018; Tappin and Hewitt Reference Tappin and Hewitt2023). Instead, our primary measure focuses on voters’ ability to correctly attribute actual pledges to the party that made them. Correctly linking a pledge to its sponsor is necessary in order for voters to base their vote on this pledge on election day. We focused on pledges made by the Social Democrats because they entered government alone after the election, giving them the greatest chance of implementing their pledges (Thomson, Royed, Naurin et al. Reference Thomson, Royed, Naurin, Artés, Costello, Ennser-Jedenastik and Praprotnik2017).Footnote ² To avoid survey fatigue and the risk of ‘straightlining’ (ie respondents providing identical responses to multiple items), we did not ask each respondent to consider all 112 election pledges made by the Social Democrats (Dillman, Smyth, and Christian Reference Dillman, Smyth and Christian2014; Hillygus and LaChapelle Reference Hillygus and LaChapelle2022). Instead, we selected a subset of promises, using the results of the pre-election study to choose the most salient pledges according to the electorate, as they should also matter most when voters select their party on election day (selecting a subset of pledges based on their saliency is also used by previous studies, see Thomson Reference Thomson2011; Thomson and Brandenburg Reference Thomson and Brandenburg2019).

In the pre-election study, we included all 201 pledges by the Social Democrats and the Liberal Party in a crowdsourcing design, in which each respondent was randomly assigned five pledges and asked different questions for each pledge (we included both parties, as we naturally did not know beforehand who would win the election; see footnote 2). Importantly, we asked respondents the following question: ‘During an election campaign, political parties make many statements. If you heard this particular statement from a party during a campaign, how likely is it that it could influence your vote choice?’ with response categories ranging from 1 (‘Very unlikely’) to 7 (‘Very likely’) as well as a ‘Don’t know’ category. Based on a sample average of this question, we selected four pledges that were among the most salient and that also pertained to four different key issues discussed in the run-up to the election (see the online Appendix for a list of all pledges and a more detailed discussion of this selection). Thus, the four pledges covered four different topics that were all highly salient during the campaign: pensions, immigration, health care, and the environment (see Hansen and Stubager Reference Hansen and Stubager2021). The four pledges were the following:

1. (i) ‘We want to introduce a new right to early retirement’ (pension pledge).

2. (ii) ‘We want to tighten eligibility for immigrants’ reunification in Denmark’ (immigration pledge).

3. (iii) ‘We want to introduce targeted, preventive health care screening to cure cancer and other chronic illnesses in their early stages’ (health care pledge).

4. (iv) ‘We want to reduce CO₂ emissions from agriculture. We will establish a council including the agricultural sector to present reasonable goals for reducing CO₂ emissions among farmers’ (climate pledge).

These four pledges were rated among the most important by voters and corresponded to the most salient issues during the election (Bøggild and Jensen Reference Bøggild and Jensen2023; Hansen and Stubager Reference Hansen and Stubager2021), and they are thus the most relevant ones for the purpose of studying mandate logic and promissory representation. While the least salient pledges (such as ‘We will increase tax deduction for research and development by 130 percent’) might still serve a purpose for the Social Democrats and be important for a small subgroup of society, these pledges were likely not decisive for the overall election result.

For pledges to work for prospective voting, it is necessary that voters know whether or not a specific party, such as the Social Democrats, has made a particular pledge, especially when it is among the most popular pledges of the election. Thus, for each of these four pledges, we asked respondents which of the 13 parties running in the election had made it (respondents could pick more than one party). Respondents knew that the four pledges were genuine but had to be able to link them to the Social Democrats (other parties might have made similar pledges, but this study is interested in whether voters knew that the Social Democrats, one of the main contenders for prime minister, had made particular pledges). In this way, our measure uses a binary approach to study pre-election pledge awareness. Alternatively, we could have implemented a more continuous measure of pledge awareness (asking respondents to say how certain they are that Party A made Pledge B). Yet we are mainly interested in the distinction between whether a voter is confident enough to select a specific party as the pledge sponsor or not, as we believe that this certainty threshold is the important one for it to matter after the election.

Figure 3 illustrates pre-election awareness of the Social Democrats’ pledges in the electorate at large. As Panel A illustrates, the level of awareness is generally low, especially considering that we focus on the most salient pledges to begin with. One notable exception, however, is the pledge to introduce a new right to early retirement (pension pledge). More than half of the respondents (53.1%) knew that the Social Democrats had made that pledge. In contrast, only around one in twenty (5.4%) knew that the party had made the pledge to tighten eligibility for immigrants’ reunification in Denmark (immigration pledge). For the remaining two pledges, the health care pledge and the climate pledge, awareness is somewhere in between these two poles but towards the lower end, with 16.3 per cent and 11.4 per cent aware that the Social Democrats had made each of these pledges, respectively. Nevertheless, while awareness is low, it is better than chance (marked by the dashed line). Randomly picking the Social Democrats among the 13 parties in total would yield around 7.69 per cent, meaning that the electorate performs better than chance for three of the four most salient pledges from the Social Democrats.

Figure 3.

Pension pledge well-known among voters.

Notes: N = 6,233. 95% confidence intervals. Panel A shows the share of voters who knew that the Social Democrats had made each of four salient pledges during the 2019 Danish national election. The dashed horizontal line shows the chance of randomly selecting the Social Democrats among the 13 parties (if only selecting one party and assuming that each party has an equal likelihood of being picked). Panel B shows the share of voters who mentioned a pledge related to each of the four issues and matched this pledge with the Social Democrats.

Figure 3. Long description

The bar graph consists of two panels, A and B. Panel A shows the percentage of pledge sponsor selection for Pension, Health, Climate, and Immigration. Pension has the highest selection rate at approximately fifty-two percentage, followed by Health at around twenty-two percentage, Climate at about fifteen percentage, and Immigration at roughly five percentage. Panel B displays self-reported pledge recall for the same categories. Pension again leads with approximately fourteen percentage, Health follows at around seven percentage, while Climate and Immigration both have recall rates below three percentage. Error bars indicate variability in the data. All values are approximated.

These results clearly suggest that although awareness in general is low, pledges can reach the public consciousness. Among the Liberal Party’s pledges, a similar picture appears, although no pledge receives the same amount of recognition as the pension pledge by the Social Democrats (see the Appendix). Despite the advantages of using this measure of linking pledges to their sponsor, it should be noted that people might respond correctly by making a qualified guess rather than being aware of the pledge according to the definition above (ie people guess that the Social Democrats made the pledge to increase public spending on pensions because this party is more strongly associated with pro-welfare state policies than, say, a liberal party). Voters may also be misled when they rely on heuristics (Colombo and Steenbergen Reference Colombo and Steenbergen2020; Dancey and Sheagley Reference Dancey and Sheagley2013; Lau and Redlawsk Reference Lau and Redlawsk2001), for example, by associating climate-related pledges with green parties and immigration-related pledges with right-oriented parties (these parties usually have issue ownership, see Abou-Chadi Reference Abou-Chadi2016; Petitpas and Sciarini Reference Petitpas and Sciarini2022; Seeberg Reference Seeberg2017). As the use of these heuristics can lead to an overestimation and an underestimation of real pre-election pledge awareness, we do believe that our measure mirrors well the complexity that voters experience in real life.Footnote ³

However, to supplement the measure of pre-election pledge awareness, we also used open-ended questions to ask respondents whether they could name any election pledges from the campaign (they were given the option to write up to ten pledges). These open-ended questions were posed before respondents were shown the selected campaign pledges (for more information about the open-ended measure, see the Appendix). Many voters did not write any pledges (67.1%). While this result seems to support the impression that pledge awareness is, on average, relatively low, it is difficult to say whether the large share of respondents without any pledge mentions reflects a lack of awareness or a lack of willingness to put in the extra effort of writing the pledges (open-ended questions generally require greater cognitive burdens, see Hillygus and LaChapelle Reference Hillygus and LaChapelle2022). Even those who did write a response often used relatively broad single-word answers like ‘Pension’ or ‘Climate’ (this imprecise and expansive understanding of what counts as a pledge has also been shown by Naurin Reference Naurin2011b). To validate the selection-based measure, Panel B in Figure 3 shows the share of respondents who mentioned a pledge related to one of the four issues and linked this self-reported pledge to the Social Democrats. Reassuringly, the distribution of the self-reported measure follows the same pattern as the distribution of the selection-based measure (compare Panel A and Panel B of Figure 3). This reinforces our confidence that the primary measure of pledge sponsor selection reflects actual pre-election pledge awareness (and that this is the most relevant measure to use, as the self-reported measure clearly underestimates real knowledge – compare the share of respondents reporting awareness of the pension pledge in the two panels of Figure 3). We do, however, perform robustness tests in the panel analysis with the self-reported pledge recall measure, finding substantially similar results.

In the Appendix, we use the pre-election study data to show that this relatively low level of pledge awareness is not an artifact of selecting these specific four pledges but corresponds well to overall awareness across all pledges made by the Social Democrats (the average share of correct answers is 15.4%).

Measuring post-election pledge fulfillment awareness

In the post-electoral stage, our primary variable of interest is related to voters’ ability to correctly evaluate whether each of the four pledges chosen for the panel survey has been fulfilled or not (see question wording in the Appendix). Answer options included ‘Fulfilled’, ‘Not fulfilled’, or ‘Don’t know’. One of the pledges (the pension pledge) was fulfilled by the Social Democrats during the post-election panel survey period, meaning that we could track whether voters noticed this change in fulfillment status, and we could also compare evaluations of this pledge with the remaining pledges that were not fulfilled.

Analysis design

Our analytical model exploits our sophisticated research design to enhance the causal validity of our empirical findings. First, our design makes problems of reverse causality implausible since pre-election pledge awareness clearly occurs chronologically prior to post-election fulfillment of pledges. Second, we exploit our rolling re-interview design in the panel survey to capture the ability to correctly update the fulfillment status of the pension pledge (as this pledge was fulfilled within the panel survey period). Furthermore, we also perform additional analyses that study whether pre-election pledge awareness endures from Wave 1 to Wave 2 and then study the change in pledge fulfillment updating from Wave 2 to Wave 3 (holding individual-level factors constant by resembling a difference-in-difference design). Third, we use the selection of multiple pledges to conduct placebo tests to reinforce our confidence in the importance of pre-election awareness of specific pledges as a causal factor. Finally, we use the selection-based measure and the self-reported pledge recall measure in the main analysis (as explained above).

In addition to setting up the design to rule out problems of reverse causality and individual-level confounding variables, we also control for variables that have been highlighted by previous research as important factors. In the main analysis, we use four different models: a baseline model without any controls, a model with demographic variables, a model with political control variables, and a model with demographic and political control variables. Gender, age, and education have all been found to affect political knowledge and might therefore also affect how informed voters are about election pledges (Delli Carpini and Keeter Reference Delli Carpini and Keeter1996; Frazer and Macdonald Reference Frazer and Macdonald2003; Grönlund and Milner Reference Grönlund and Milner2006; Jennings Reference Jennings1996; Lambert, Curtis, Kay et al. Reference Lambert, Curtis, Kay and Brown1988; Mondak and Anderson Reference Mondak and Anderson2004; Rasmussen Reference Rasmussen2016). Like previous studies on respondents’ evaluations of pledge fulfillment (Naurin and Oscarsson Reference Naurin and Oscarsson2017; Thomson and Brandenburg Reference Thomson and Brandenburg2019), we therefore control for these variables. Additionally, we also include region and children under 18 in the household as control variables. These are the five demographic control variables (see operationalization of all five in the Appendix).

For the political control variables, we first include political trust, as this has been found to matter for voters’ pledge evaluations (Thomson and Brandenburg Reference Thomson and Brandenburg2019). Party belonging is another relevant covariate, as it might affect the way voters perceive and interpret the political world (Campbell, Converse, Miller et al. Reference Campbell, Converse, Miller and Stokes1960; Johnston Reference Johnston2006). Respondents might thus be more likely to associate an election promise with a party that they like. After the election, voters might also evaluate the fulfillment status of government pledges more positively if they have stronger ties to the party in government (as argued in the motivated reasoning literature, see Taber and Lodge Reference Taber and Lodge2006). In line with previous pledge research, we therefore also control for party belonging using a vote measure (whether a respondent voted for the Social Democrats in 2015 or not) and sympathy towards the Social Democrats. We also control for ideology and pledge agreement, as voters who like a pledge might be more prone to associate it with parties they like and to later interpret the fulfillment of this pledge in a biased way. Another pledge-specific variable is the salience of the pledge. Thomson (Reference Thomson2011) finds that the personal relevance of a pledge matters for voters’ likelihood of knowing its fulfillment. In a similar vein, finding a pledge important should matter for voters’ ability to correctly identify the sponsor of the pledge and its fulfillment after the election. We measure pledge saliency as the individual importance of a given pledge compared to the other pledges that respondents were asked about. As an alternative measure of saliency, we also include the importance of the issue related to a given pledge as a control variable.

Political interest and news interest are two additional relevant control variables that have also been included in previous research on election pledges (Naurin and Oscarsson Reference Naurin and Oscarsson2017). People following politics closely in the media and elsewhere have a higher chance of hearing about pledges during the campaign and after the election, especially given the mediatization of politics (Strömbäck Reference Strömbäck2008). This should make high-interest voters more likely to know about the sponsor of a pledge before an election and the fulfillment of the pledge afterwards. Voters with high political interest might also be better at guessing, because they are more familiar with the political profiles of the parties and the political context. We also included political efficacy as a control variable, leaving us with ten political control variables in total (see the operationalization of all ten in the Appendix).

Our estimation model relies on OLS regression. We apply standard errors clustered at the individual level in all models. Below, we show the regression equation for our main analyses, ie Figure 5 (Panel A, Model IV) and Figure 7:

(1) ${Y_i} = {\beta _0} + {\beta _1}{K_i} + {\rm{B}}_i^ \top \gamma + {\rm{W}}_i^ \top \delta + {\varepsilon _i}$

(2) ${Y_i} = {\beta _0} + {\beta _1}{K_i} + \;{\beta _2}{P_i} + {\beta _3}({K_i} \times {P_i}) + {\rm{B}}_i^ \top \gamma + {\rm{W}}_i^ \top \delta + {\varepsilon _i}$

where i indexes individual respondents in Group 3 of the panel survey. ${Y_i} = 1\;$ if respondent i correctly identifies the pension pledge’s fulfillment status in Wave 2 (not fulfilled) and Wave 3 (fulfilled), and ${Y_i} = 0\;$ otherwise. ${K_i} = 1$ if respondent i correctly identifies that the Social Democrats made the pension pledge in Wave 1, and ${K_i} = 0\;$ otherwise. ${{\rm{B}}_i}\;$ is the vector of covariates measured pre-Wave 1 (gender, age, education, children under 18, region, news interest). ${{\rm{W}}_i}$ is the vector of covariates measured at Wave 1 (political trust, ideology, 2015 Social Democrats vote, Social Democrats sympathy, pledge saliency, issue saliency, pledge agreement, political efficacy). In the first equation, political interest, ${P_i}$ , is included in ${{\rm{W}}_i}$ , while it is part of the constitutive terms, ${\beta _2}{P_i},$ , and the interaction term, ${\beta _3}({K_i} \times {P_i})$ , in the second equation. ${\beta _0}$ is the intercept, and ${\varepsilon _i}\;$ is an idiosyncratic error term.