A transformation of the political landscape

In many Western democracies, political parties have witnessed a significant transformation of the political landscape, especially within recent decades. Several new parties have emerged, typically with a niche focus on issues such as immigration, climate or regional independence (De Vries & Hobolt, Reference De Vries and Hobolt2020; Meguid, Reference Meguid2008). The old parties have experienced that their traditional voters have disappeared or are much less loyal, and the historical linkages between social class and vote choice have weakened (Dalton, Reference Dalton2002; Inglehart, Reference Inglehart1997; Kriesi et al., Reference Kriesi, Grande, Lachat, Dolezal, Bornschier and Frey2006; Stubager, Reference Stubager2013; Thomassen, Reference Thomassen2005).

New connections between voters and parties may have appeared, but voters are still much more flexible in their party choice than previously. They change party from election to election (Dalton et al., Reference Dalton, McAllister, Wattenberg, Dalton and Wattenberg2000; Dassonneville, Reference Dassonneville2022; Mair, Reference Mair2013), and they do not decide their vote choice before the election campaign starts and oftentimes not until the very last days of the campaign (Box‐Steffensmeier et al., Reference Box‐Steffensmeier, Dillard, Kimball and Massengill2015; Dalton et al., Reference Dalton, McAllister, Wattenberg, Dalton and Wattenberg2000; Irwin & Van Holsteyn, Reference Irwin and Van Holsteyn2008). Furthermore, parties increasingly compete on issues in an attempt to set the common political agenda (Green‐Pedersen, Reference Green‐Pedersen2007). Parties can try to raise different issues on which they have an advantage, but they also have to emphasize issues that are outside their primary focus because it is on the agenda of the other parties or the electorate (Green‐Pedersen & Mortensen, Reference Green‐Pedersen and Mortensen2010; Wagner & Meyer, Reference Wagner and Meyer2014).

These empirical developments have changed the arena in which political parties operate. The consequences of having more parties, more flexible voters and more issue competition are twofold: The competition for votes has increased, and election campaigns have become more important. How can parties respond to these developments in a way that attracts voters? Below, I argue that the policy‐committing strategy is particularly useful in this new political landscape.

A costly signal

A statement includes a commitment when a party takes high responsibility for a future policy. The party links itself to the provision of a policy by saying that it will make sure that the policy will be implemented.Footnote ² Commitments can increase the credibility of a party's policy programme because they help voters hold the party accountable at the next election. This should enhance voters’ confidence that the party will keep its word and pursue its stated policy plans. Empirically, voters do increase their expectations of policy implementation when parties have made a promise on it (Bonilla, Reference Bonilla2022), and parties, generally, gain votes by making promises (Bøggild & Jensen, Reference Bøggild and Jensen2023; Elinder et al., Reference Elinder, Jordahl and Poutvaara2015). A commitment should be even more valuable to voters today than previously as post‐election government negotiations have become more difficult and unpredictable (for instance, the number of days until a government is formed is higher today, see Online Appendix B). By making a commitment, a party signals to voters that they are willing to prioritize the policy in these complex negotiations.

Even though parties can profit by making commitments in the short term, they limit their flexibility in the long term because they will have to fulfil them after the election. Predicting a party's likelihood of keeping its promises is notoriously difficult, as there are many steps between the formulation of a promise and actual policy implementation: Election results are unknown as is the outcome of government formation, especially if it includes a coalition of several parties (e.g., De Winter & Dumont, Reference De Winter, Dumont, Strøm, Müller and Bergman2008; Golder, Reference Golder2010). In the legislative arena, a party often has to negotiate with other parties and make compromises, which eventually means breaking its pre‐electoral promises (Thomson et al., Reference Thomson, Royed, Naurin, Artés, Costello, Ennser‐Jedenastik, Ferguson, Kostadinova, Moury, Pétry and Praprotnik2017). Finally, unforeseeable crises can make specific policies more demanding to implement than a policy promised before an election (Praprotnik, Reference Praprotnik2017b).

If parties break their promises, they are likely to attract negative media coverage (Duval, Reference Duval2019; Müller, Reference Müller2020) and risk that voters punish them in the next election. Empirically, Matthieß (Reference Matthieß2020) finds that parties lose a higher share of votes the more pledges they break. Voters also evaluate parties more negatively when pledges are broken (Naurin, Soroka, et al., Reference Naurin, Soroka and Markwat2019). While electoral punishment does not necessarily translate into a loss of seats, policymaking power or government offices (this depends on party characteristics and institutional settings such as the electoral system and party system; see Kam et al., Reference Kam, Bertelli and Held2020), we should expect that losing votes, all else equal, is harmful to a vote‐seeking party. Making promises is thus potentially electorally costly in the long term, and parties should save the strategy for situations in which they are in high need of credibility in the short term.

In sum, the policy‐committing strategy provides prospects and perils to parties that need to adapt to changes in their environment. While promises limit the parties’ manoeuvrability after the election, the pressure to make commitments has grown in recent decades due to the developments mentioned above. Making commitments can be a way for a party to attract voters by making its policy programme more credible. Therefore, I expect a time trend in the use of the policy‐committing strategy:

However, the pressure to commit has not necessarily increased homogenously across contexts but may depend on the specific party system, party type and issue on which a party campaigns. Below, I elaborate on these factors and their influence on parties’ use of the policy‐committing strategy.

System‐level differences: The importance of commitments in complex negotiation environments

The number of political parties has generally increased over time, usually measured as the effective number, which takes each party's relative strength into account (Best, Reference Best2013; Dalton et al., Reference Dalton, McAllister, Wattenberg, Dalton and Wattenberg2000; De Vries & Hobolt, Reference De Vries, Hobolt, Velasco and Bucelli2022). Particularly, this development has happened in countries with proportional voting systems as they are more permissive than systems with plurality voting (Duverger, Reference Duverger1959).

When the number of parties in a political system increases, the competition for votes intensifies, all else equal. Moreover, a higher number of parties also has consequences for the period after the election. For instance, the number of parties in a system is positively associated with the level of fractionalization within a government (Valentim & Dinas, Reference Valentim and Dinas2024). Furthermore, the negotiation period is typically more complex and lasts longer until a government is formed because parties have to get to know each other's preferences and more parties might disagree (Bäck et al., Reference Bäck, Hellström, Lindvall and Teorell2024; Golder, Reference Golder2010). Consequently, the successful implementation of a party's policy programme also becomes more uncertain. For instance, if parties are trying to form a coalition, and they disagree on an issue, they may use the written agreement to make compromises (Eichorst, Reference Eichorst2014; Klüver et al., Reference Klüver, Bäck and Krauss2023; Martin & Vanberg, Reference Martin and Vanberg2011). In a coalition government, parties also have to allocate ministerial offices, and a minister from the other party may pursue policies that are against the coalition partners’ interests (Laver & Shepsle, Reference Laver and Shepsle1996; Martin & Vanberg, Reference Martin and Vanberg2011).

From the electorate's perspective, voters should value policy commitments more when implementation of a policy programme becomes more uncertain in a system with more political parties. The party signals that despite the complex political environment, it is very determined to pursue its policies, to prioritize the policy and to put its trustworthiness at stake. Parties can thus make their election programme more credible by using the policy‐committing strategy.Footnote ³

To sum up, since a larger number of parties increases competition before elections and uncertainty about policy implementation after, parties should be more likely to make commitments in a highly fragmented party system to attract voters:

A particular type of party, usually referred to as niche parties (e.g., Meguid, Reference Meguid2005), has emerged in these increasingly fragmented party systems and challenges established mainstream parties. Below, I argue that the pressure to commit is uneven across the two types of parties.

Party‐level differences: How a party's mainstream status affects the usefulness of the policy‐committing strategy

During the 1980s, green parties and anti‐immigration parties gained prominence among voters at the expense of mainstream parties such as conservatives, social democrats and liberals (De Vries & Hobolt, Reference De Vries and Hobolt2020). Green parties and anti‐immigration parties are often defined as niche parties as they mainly focus on a specific niche issue (Meguid, Reference Meguid2005). Their advantage is that they can appear very determined and principled on their core issue and attract voters who find this issue important as well (Petrocik, Reference Petrocik1996; Wagner & Meyer, Reference Wagner and Meyer2014). Typically, they are more extreme on their positions, follow the preferences of their supporters and attract voters by appearing principled and motivated by policy (Adams et al., Reference Adams, Clark, Ezrow and Glasgow2006; Ezrow et al., Reference Ezrow, De Vries, Steenbergen and Edwards2011; Klüver & Spoon, Reference Klüver and Spoon2016). In contrast, mainstream parties usually cover several issues and may appear less determined on specific issues, yet they appeal more broadly (Adams et al., Reference Adams, Clark, Ezrow and Glasgow2006; Bergman & Flatt, Reference Bergman and Flatt2020; Greene, Reference Greene2016; Van Heck, Reference Van Heck2018). While several studies have shown that mainstream parties may respond to the emergence of niche parties at the issue level (Abou‐Chadi, Reference Abou‐Chadi2016; Meguid, Reference Meguid2008), I argue that mainstream parties can use a strategy at the programmatic level, namely the policy‐committing strategy. By making commitments, mainstream parties can try to appear credible and determined in their policy programme and thus counterbalance the positive reputation of niche parties as more principled and policy‐seeking.

Another argument for making commitments is that the mainstream party highlights its own status as government potential. Mainstream parties are typically government actors, while niche parties usually stay in opposition (Meguid, Reference Meguid2005). By making commitments, a government‐aspiring party signals to voters that if they vote for the party, its policy preferences are more likely to be implemented as their support will translate into a potential future government party.

In sum, I expect that mainstream parties are more likely to use the policy‐committing strategy as a response to the relatively new phenomenon of niche parties; both to challenge the status of niche parties as more principled and to highlight their own status as a more likely government party:

As already hinted in the previous section, issue emphasis is an important way parties compete against each other and a way for scholars to distinguish between parties. However, to understand parties’ pressure to commit, we also need to focus on specific issues and the general dynamics of issue competition. This is the purpose of the next section.

Issue‐level differences: How parties need credibility when changing issue emphasis

Political parties increasingly compete on political issues (Greene, Reference Greene2016; Green‐Pedersen, Reference Green‐Pedersen2007). Even niche parties are broadening their issue appeals (Bergman & Flatt, Reference Bergman and Flatt2020; Meguid, Reference Meguid2022; Meyer & Wagner, Reference Meyer and Wagner2013). Parties will try to force their competitors to focus on their issue agenda and at the same time be constrained by the common party system agenda (Green‐Pedersen & Mortensen, Reference Green‐Pedersen and Mortensen2010). Increased issue competition implies that parties sometimes have to emphasize issues they did not talk much about previously at the expense of issues they used to prioritize.

A major cost of changing issue priorities is the loss of credibility. A party's credibility comes from many years of dedication to an issue (Budge & Farlie, Reference Budge and Farlie1983; Petrocik, Reference Petrocik1996). When a party shifts attention from one election to the next, it loses its political identity (Janda et al., Reference Janda, Harmel, Edens and Goff1995) and has no reputation on its new issue agenda (Walgrave et al., Reference Walgrave, Tresch and Lefevere2015). Changing the political package can therefore be harmful to the party (Meyer, Reference Meyer2013, p. 10; Volkens & Klingemann, Reference Volkens, Klingemann, Luther and Müller‐Rommel2002, p. 145). Voters are sceptical about the party's ability and willingness to solve the problems they suddenly begin to worry about and will eventually perceive its change in viewpoints as opportunistic and dishonest (Tomz & Van Houweling, Reference Tomz and Van Houweling2012). Voters prefer parties that are consistent and reliable, so when a party changes its profile voters may respond negatively (Ansolabehere & Snyder, Reference Ansolabehere and Snyder2000; Groseclose, Reference Groseclose2001).

Therefore, a party mainly changes its emphasis if it is forced to, for instance, if it has lost an election or is trailing opposing parties and wants to be more responsive to the voters’ priorities (Damore, Reference Damore2004; Janda et al., Reference Janda, Harmel, Edens and Goff1995; Meyer & Wagner, Reference Meyer and Wagner2013; Spoon & Klüver, Reference Spoon and Klüver2014). Alternatively, the party may have to focus on new issues if other parties have placed them on the agenda (De Vries & Hobolt, Reference De Vries and Hobolt2020; Green‐Pedersen & Mortensen, Reference Green‐Pedersen and Mortensen2010; Meguid, Reference Meguid2005). If some problems become too large to ignore, parties may have to respond by increasing their focus on the issue (Baumgartner & Jones, Reference Baumgartner and Jones2010; Dalmus et al., Reference Dalmus, Hänggli, Bernhard, Van Aelst and Walgrave2017; Kristensen, Reference Kristensen2020; Seeberg, Reference Seeberg2021). In these cases, the party must convince voters that it can deliver credible solutions to problems it did not pay much attention to before. Voters might be sceptical about the party's intentions and be unsure whether it will pursue the policies after the election because it has no record on these matters. The party must compensate for a lack of credibility by signalling that it is trustworthy on its new political agenda, and this is where the policy‐committing strategy becomes important.

By making commitments, parties signal to voters that their new priorities are credible. While they risk losing credibility when they abruptly focus on political problems they previously ignored, parties can attempt to regain the electorate's trust by constraining their future behaviour. They commit themselves to solving the problems they pay attention to. When parties use the policy‐committing strategy to rebrand their political images, they exchange flexibility in the future policymaking phase for voter credibility in the current campaign phase:

To test the four hypotheses, I developed a large cross‐national dataset that identifies parties’ use of commitments in 11 countries. In the next section, I will elaborate on the data collection process.

Data source and case selection

Parties’ election promises have been identified in research, but they have never been measured within parties’ issue agendas in a cross‐country design (for one‐country studies, see Praprotnik, Reference Praprotnik2017a; Praprotnik & Ennser‐Jedenastik, Reference Praprotnik and Ennser‐Jedenastik2023), perhaps because it is difficult to identify promises systematically (Naurin, Royed, et al., Reference Naurin, Royed and Thomson2019).Footnote ⁴ To test my argument, I developed and tested a coding scheme that produces high‐quality human coding (see more details about this part in Vestergaard, Reference Vestergaard2023) that could subsequently be used to train a language model to code unknown material and allowed me to use machine coding to include several countries. The final dataset gives me an overview of parties’ use of policy commitments across 11 countries covering four decades. I will explain the coding process in detail later, but first, I will describe the data source and case selection.

I tested and used my coding scheme on election manifestoes, which are, in a sense, contracts between parties and voters. Parties also make commitments on other platforms, such as social media, websites, political ads and television, but they announce their political visions in the manifesto, including their views on the most pressing problems and their solutions to them, and voters indirectly sign the contract by their vote on the election day. There are many steps between an election campaign and actual policy enactment after an election (for instance, the election itself, negotiations between parties, government formation and parliamentary debates about policy bills), and the manifesto works as a written agreement that voters can use to remind the party about its promises. In this way, manifestoes establish an important link of accountability between parties and voters and are used in most studies on election promises (e.g., Artés & Bustos, Reference Artés and Bustos2008; Naurin, Royed, et al., Reference Naurin, Royed and Thomson2019; Thomson, Reference Thomson2001). Another benefit is that I can use manifestoes to track parties’ political agendas across countries and over time and test the issue‐level hypothesis. In particular, I can measure party agendas and commitments from the same data source, which is advantageous since I assume that parties base their policy‐committing strategy on a total set of considerations, including how salient they find different issues.

The analysis includes 11 countries with different institutional characteristics: United States (US), United Kingdom (UK), Australia (AU), Canada (CA), New Zealand (NZ), Switzerland (CH), Germany (DE), Luxembourg (LU), Austria (AT), Ireland (IE) and Denmark (DK). Diversity in terms of political setting is crucial because it affects parties’ incentives to make commitments as argued in the theory section. Most importantly, the countries differ regarding the number of parties normally represented in the parliamentary arena, ranging from a pure two‐party system like the United States to a pure multi‐party system like Denmark. Moreover, the countries vary in terms of the typical type of government and the executive–legislative relation, which are added to the analysis as control variables (see below). Online Appendix C provides an overview of differences and similarities between the 11 countries on these three institutional characteristics.

By including a diverse set of countries in the analysis, I also increase the likelihood that the results from this study can be generalized to countries with varying types of party dynamics and institutional setups. I include all available election manifestoes from the 11 countries in the corpus of the Comparative Manifesto Project (Burst et al., Reference Burst, Krause, Lehmann, Lewandowski, Matthieß, Merz and Zehnter2020); see Online Appendix D.

Dependent variable

Measuring the policy‐committing strategy

After having identified commitments in election manifestoes, I calculated how committed the party was. This could be done at the manifesto and issue level. For the first three hypotheses, both measures were used, but only the issue‐level measure made sense for the fourth hypothesis to test the expected issue dynamics.

At the manifesto level, I simply calculated the share of commitments relative to the total number of quasi‐sentences. I measured this as a percentage. The formula that measures the share of commitments (SC) is

$SC=\frac{C}{N}\times \mathrm{>100},$

where C is the number of quasi‐sentences that includes commitments and N is the total number of quasi‐sentences in the manifesto.

At the issue level, I made this calculation within issues. The formula then becomes

$S{C}_i=\frac{C_i}{N_i}\times \mathrm{>100}$

where i is the issue, C is the number of commitments and N is the number of quasi‐sentences.

The Comparative Manifesto Project has already coded each quasi‐sentence on the issue dimension, so I used their coding to measure the share of commitments at the issue level (Volkens et al., Reference Volkens, Burst, Krause, Lehmann, Matthieß, Merz, Regel, Weßels and Zehnter2020). Their dataset has 56 main categories and several subcategories. To avoid too many issue categories, I grouped the codes (e.g., per104 ‘Military: Positive’ and per105 ‘Military: Negative’ in one military issue, and per303 ‘Governmental and Administrative Efficiency’ and per304 ‘Political Corruption’ in one bureaucracy issue) into 24 issues: agriculture, bureaucracy, centralization, civic‐mindedness, constitutionalism, culture, democracy, EU, economic groups, economy, education, environment, equality, foreign policy, internationalism, law and order, military, morality, multiculturalism, national way of life, non‐economic groups, protectionism, technology and infrastructure and welfare. In Online Appendix E, I show which codes from the Comparative Manifesto Project I grouped into the broader issue categories. In Online Appendix H, I tested the robustness of my findings with a more minimal grouping of issue codes (43 issue categories).

The descriptive statistics of the variable reveal that around 14 per cent of the quasi‐sentences within an issue contain policy commitments (see Table 1 for more details). This suggests that commitments are indeed a costly signal that parties are hesitant to use and that parties also use election manifestoes for other purposes than making commitments.

Table 1. Descriptive statistics of variables

Note: The descriptive statistics are based on the observations included in Model II in the analysis (measuring the share of commitments at the issue level).

Primary independent variables

I have four independent variables, one for each hypothesis. To measure the potential time trend, I used the year of the election as the main variable.

To measure the effective number of parties, I relied on a variable from The Comparative Political Data Set that measures the number at the vote level (Armingeon et al., Reference Armingeon, Engler, Leemann and Weisstanner2022). They follow the formula proposed by Laakso and Taagepera (Reference Laakso and Taagepera1979) that takes each party's vote share into account when calculating the total number of parties.

The third main independent variable is related to the status of a party as either mainstream or niche. Traditionally, this has been done by categorizing parties based on their party family. Even though more recent accounts base the decision on a more nuanced measure of a party's communication style and issue profile (Bischof, Reference Bischof2017; Greene, Reference Greene2016; Meyer & Miller, Reference Meyer and Miller2015; Wagner, Reference Wagner2012), the party family approach is still largely used (e.g., Gunderson, Reference Gunderson2023; Spoon & Klüver, Reference Spoon and Klüver2019) and was chosen here to not confuse the party‐level variable with the fourth independent variable of saliency change. Thus, mainstream parties belong to the party family of Christian democrats, conservatives, liberals or social democrats, while niche parties are all other party families (the categorization into party families is from the Comparative Manifesto Project; see Volkens et al., Reference Volkens, Burst, Krause, Lehmann, Matthieß, Merz, Regel, Weßels and Zehnter2020).

Regarding the fourth independent variable, I measured saliency change at the issue level making use of the Comparative Manifesto Project's coding of election manifestoes (Volkens et al., Reference Volkens, Burst, Krause, Lehmann, Matthieß, Merz, Regel, Weßels and Zehnter2020). As a first step, I calculated the number of quasi‐sentences a party devoted to a specific issue relative to the total number of quasi‐sentences in the manifesto. Specifically, I used Lowe et al. (Reference Lowe, Benoit, Mikhaylov and Laver2011)’s formula to calculate saliency (Sa):

$S{a}_i=\log \left(\frac{S_{e_i}+\mathrm{>1}}{N}\right),$

where i is the issue, Se is the number of quasi‐sentences within an issue i and N is the total number of quasi‐sentences in the manifesto.

By using the logarithmic measure, I take into account that an increase in saliency means more when a party puts less emphasis on an issue (Lowe et al., Reference Lowe, Benoit, Mikhaylov and Laver2011). Increasing emphasis on an issue from 1 per cent to 2 per cent of the manifesto represents a larger boost in saliency than increasing the emphasis from 50 per cent to 51 per cent. This logarithmic measure of saliency is scaled in such a way that the less negative it is, the more salient the issue is to the party.

As a second step, I measured the change in saliency from one election to the next by subtracting the saliency of an issue at the current election from the saliency at a previous election. In most cases, saliency is available for the previous election, but, if not, I take the most recent available election (except if the previous election was more than 10 years ago). Thus, positive values mean that a party increased the saliency towards the current election (the more positive, the more salient), and negative values mean that a party decreased the saliency (the more negative, the less salient).

Estimation strategy and control variables

In the analysis, I used negative binomial multivariate regression models with robust standard errors, which are suitable when the dependent variable has a count‐like structure,Footnote ⁵ follows a negative binomial distribution and is over‐dispersed (i.e., the mean is considerably lower than the variance). I ran two models, one using the policy‐committing strategy at the manifesto level and one at the issue level. In both models, I included government status (based on ParlGov, see Döring & Manow, Reference Döring and Manow2021) before the election as a control variable. Government experience matters both for parties’ issue agendas and the extent to which they communicate prospectively about policies and make promises in their manifestoes (Müller, Reference Müller2022; Praprotnik & Ennser‐Jedenastik, Reference Praprotnik and Ennser‐Jedenastik2023; Van Heck, Reference Van Heck2018).

Regarding country‐level controls, I added, first, a variable on the executive–legislative relations within the country (i.e., a presidential system or a semi‐presidential/parliamentary/hybrid system) and, second, a variable on the type of government at a given election (single‐party majority, minimal winning coalition, surplus coalition, single‐party minority government or multi‐party minority government). Like the effective number of parties, these institutional differences might affect a party's incentive to make commitments as well as constrain its ability to pursue its preferred policies after an election (e.g., a minority government needs support from other parties to implement its policies). These two variables are from the Comparative Political Data Set (Armingeon et al., Reference Armingeon, Engler, Leemann and Weisstanner2022).

In the issue‐level model, I included issue dummies to control for unique issue‐specific characteristics. Furthermore, it is important to control for the current saliency of an issue, since higher saliency generally makes it more demanding to achieve a high share of commitments irrespective of how much the party changed the saliency since the previous election (e.g., if it has written 10 quasi‐sentences on an issue, it only has to make commitments in five of them to achieve a 50 per cent share of commitments, while it has to commit in 50 quasi‐sentences if it has written 100 quasi‐sentences on the issue to achieve the same share).Footnote ⁶ Similarly, I controlled for overall issue engagement at the previous election (i.e., whether the party talked about the issue). Other potential issue‐level controls would be the inclusion of voter agendas as well as problem indicators since these could affect parties’ incentives to change their issue focus (as argued in the Theory section). However, it is difficult to get data on these measures that can be easily matched with the issues from the Comparative Manifesto Project. Furthermore, it is not easy to specify the model with these controls, as parties’ saliency change might both be the result of voters’ agendas and different problem indicators and the cause of these.

In Table 1, I give an overview of the descriptive statistics of the dependent variable as well as the four main independent variables in the analysis. The statistics are based on Model II. In Online Appendix G, I show the descriptive statistics of all variables and from each of the two models.

Time trend hypothesis

As shown in Table 2, there is a positive and significant time trend (illustrated by an incidence rate above 1), suggesting that political parties make a higher share of commitments in recent elections compared to previously. The incidence rate of 1.02 in both models means that a party has increased its share of commitments by a factor of 1.02 per year.

As illustrated in Figure 1, this is a substantial effect. Based on the predicted values, a party doubled the share of commitments within manifestoes from the start until the end of the period under study in this paper. While policy commitments were present in around 1 out of 10 quasi‐sentences in 1984, this was the case in around 1 out of 5 in 2019. Thus, the time trend hypothesis finds support in the data.Footnote ⁷

Figure 1. The use of policy commitments in election manifestoes over time.

Note: Negative binomial regression; see Table 2 (Model I). Confidence intervals are at the 95 %‐level.

System‐level hypothesis

Looking at the second primary independent variable, the coefficients in both Models I and II are significant and below 1 (the incidence rate ratio is 0.72 and 0.70, respectively). This suggests, contrary to the expectation, that the share of commitments parties make decreases as the effective number of parties increases.

Figure 2 shows that the effect is remarkable. In two‐party systems, parties make a commitment in almost every third quasi‐sentence, which drops to less than every 10th in multi‐party systems with more than six parties (measured as the effective number). Since I do not find support for the system‐level hypothesis, I will discuss this result in more detail in the concluding parts of the paper.

Figure 2. The use of policy commitments as a response to the effective number of parties.

Note: Negative binomial regression; see Table 2 (Model I). Confidence intervals are at the 95 %‐level.

Party‐level hypothesis

The third hypothesis focuses on party differences, specifically the difference between mainstream and niche parties. The expectation that mainstream parties will make a higher share of commitments than niche parties is supported by the coefficients in Table 2. The incidence rate ratios are 1.39/1.36 in the two models, meaning that the share of commitments increases by a factor of 1.39/1.36 when we move from niche parties to mainstream parties. As the predicted share of commitments is 13.39 per cent for niche parties, the share for mainstream parties is 1.39/1.36 times higher (i.e., a predicted share of commitments around 18,67 per cent; see Figure 3). Thus, I find support for the party‐level hypothesis.

Figure 3. The use of policy commitments by mainstream and niche parties.

Note: Negative binomial regression; see Table 2 (Model I). Confidence intervals are at the 95 %‐level.

Issue‐level hypothesis

The final hypothesis is at the issue level. To reiterate, I expect that parties will make a higher share of commitments when they have increased the saliency on an issue. As shown in Model II of Table 2, I find support for the hypothesis. The independent variable goes from a large decrease in saliency to a large increase in saliency and changing from lower levels to higher levels on this variable increases the share of commitments. The incidence rate ratio is 1.25 and significant, meaning that when a party increases the logged saliency (or decreases the logged saliency to a lesser extent) by one scale point (the variable ranges from −5.84 to +4.99), the party increases the share of commitments by a factor of 1.25.

Figure 4 illustrates the main effect with predicted values of the dependent variable, specifically the predicted share of commitments across different levels of my independent variable (i.e., saliency change). From a very low value of saliency change (observation at the 10th percentile), that is, a high decrease in saliency, to a very high value of saliency change (observation at the 90th percentile), that is, a high increase in saliency, the share of commitments increases from around 11 per cent to more than 20 per cent. This reveals the substantial effect of saliency change on the use of the policy‐committing strategy.

Figure 4. The use of policy commitments after changing the saliency of an issue.

Note: Negative binomial regression; see Table 2 (Model II). Confidence intervals are at the 95 %‐level. ‘High decrease in saliency’ is −1.32 (10th percentile). ‘High increase in saliency’ is 1.30 (90th percentile).