- Top of page
- Theoretical framework
- A new approach to manifesto coding
Saliency theory is among the most influential accounts of party competition, not least in providing the theoretical framework for the Comparative Manifesto Project – one of the most widely used data collections in comparative politics. Despite its prominence, not all empirical implications of the saliency theory of party competition have yet been systematically tested. This article addresses five predictions of saliency theory, the central claim of which is that parties compete by selective issue emphasis rather than by direct confrontation. Since a fair test of the theory's assumptions needs to rely on data that measures party issue saliency and party positions independently, this article draws on new manifesto data from the Austrian National Election Study (AUTNES). Analysing all manifestos issued for the 2002, 2006 and 2008 general elections, it shows that saliency theory correctly identifies some features of party competition. For instance, parties disproportionally emphasise issues they ‘own’. Yet, the core assumption of saliency theory that parties compete via selective issue emphasis rather than direct confrontation over the same issues fails to materialise in the majority of cases.
- Top of page
- Theoretical framework
- A new approach to manifesto coding
Party competition is the core feature of modern representative democracy. Theories of party competition, therefore, have been a central concern for scholars of comparative politics and especially for research on elections. One of the most prominent theoretical approaches to explain the mechanics of party competition is saliency theory (Budge 1982; Budge & Farlie 1983a, 1983b). This theory is not only relevant as a genuine account of partisan contestation; it also provides the theoretical framework for the Comparative Manifestos Project (CMP) which has produced one of the most widely used datasets in comparative politics (Budge et al. 2001; Klingemann et al. 2006).
Given its enormous impact on the discipline, it is striking that saliency theory and the underlying assumptions have not yet been submitted to a comprehensive test. The CMP has received severe criticisms on methodological grounds (i.e., inter-coder reliability), because of its time-bound and by now rather outdated system of issue categories and for the selection of inappropriate party documents (e.g., Benoit et al. 2009; Laver & Garry 2000; Pelizzo 2003; Benoit & Laver 2007; Hansen 2008; Curini 2010). In addition, several alternative methods for deriving party left–right and other broad positional scores from CMP data have been put forward, all of which claim to improve on the CMP's original formulas based on percentage differences of aggregated categories (e.g., Gabel & Huber 2000; Franzmann & Kaiser 2006; Dinas & Gemenis 2011; Lowe et al. 2011). While the bulk of the critique has aimed at coding and scaling procedures promoted by the CMP, only a few studies have addressed individual implications of saliency theory itself (Budge 2001a: 83; McDonald & Mendes 2001; Sigelman & Buell 2004; Sides 2006; Druckman et al. 2012). Notwithstanding these important contributions, this article presents the first systematic test of all of saliency theory's core assumptions in a uniform research design. To that end, we use the same basic data source that saliency theorists originally examined: party manifestos.
Our aim, therefore, is not to address the possible limitations of the CMP's technique, but to empirically evaluate the very foundations that constitute the theoretical basis for its approach to content analysis and consequentially the data used by hundreds of political scientists. The question is whether parties indeed compete in elections the way saliency theory expects them to.
To assess the explanatory power of saliency theory we first provide a concise discussion of its core assertions and theoretical bases. We then test five hypotheses generated from saliency theory on data based on a new hand-coding scheme for election manifestos developed for the Austrian National Election Study (AUTNES) that allows for independent measurements of party issue saliencies and party issue positions. We argue that a fair test of saliency theory needs to rely on data other than the CMP's; otherwise one would subscribe to a circular logic since CMP data are heavily influenced by this theory's assumptions. By using the same type of data source that the original proponents of saliency theory employed we avoid biasing our empirical approach against the theory. Other sources revealing the competitive strategies of parties (e.g., party press releases or statements of party representatives in television debates) by their very nature tend to be more confrontational than election manifestos (e.g., Sigelman & Buell 2004; Damore 2005; Kaplan et al. 2006; Sides 2006).
Our analyses show that the data conform to some of saliency theory's assumptions. In particular, we attest that parties put more emphasis on issues they ‘own’. Also, there is some indication that parties de-emphasise issues on which they exhibit an electorally disadvantageous position. Yet our data do not conform to other expectations of saliency theory. Specifically, parties do not adopt similar positions on most issues. Although some issues (e.g., welfare, environment) display near-unanimity among the parties, others show them spread across the policy space (e.g., Europe, multiculturalism, internal security). The overall picture suggests that in party competition ‘direct confrontation’ is more important than ‘selective emphasis’.
A new approach to manifesto coding
- Top of page
- Theoretical framework
- A new approach to manifesto coding
A fair test of saliency theory requires independent measurements of issue positions and issue saliencies. For that reason, we cannot use CMP data as they derive party issue positions from issue saliencies in party documents. The CMP coding scheme does include some categories that directly capture positive versus negative stances (e.g., ‘108 EU: positive’ versus ‘110 EU: negative’) and therefore also allow for the calculation of positions. However, for the cases we study here these ‘double’ categories cover only a third of all manifesto content. Most of the categories capture only one direction of preference and lack a specific counterpart. A comprehensive matching of categories is therefore impossible. As a consequence, references to the very same issue may result in the coding of entirely different CMP categories, depending on whether a party supports or rejects a specific policy. For instance, in the absence of a CMP category that captures support for higher taxes, coders have to resort to the ‘Social Justice’ category in the following example from the New Zealand National Party in 1972:1
Of course those with other income will receive the benefit too, but they will pay more tax on their bigger incomes.
Because of these problems, we draw on new data which was produced as an integral part of AUTNES. To the best of our knowledge it is the only manifesto-based data that provides separate measures of issue positions and issue attention to date. Although primarily generated for other research purposes emanating from the needs of an election study, the AUTNES manifesto data provide an excellent source for testing the assumptions of saliency theory. Our analysis is based on the three most recent general elections: those of 2002, 2006 and 2008.
The theoretical background of the AUTNES manifesto coding is the relational method of content analysis developed by Kleinnijenhuis and his collaborators (e.g., Kleinnijenhuis & Pennings 2001), which was already applied by Kriesi et al. (2006, 2008). Both groups of researchers used this method to content-analyse mass media reports. The AUTNES coding scheme basically follows their approach, but makes several adaptations and extensions to accommodate the requirements of an election study and the specifics of election manifestos. We now present a short introduction to the coding scheme.2
Unitising and coding
Based on Chomsky's (1957) phrase-structure model, the unitising procedure of our content analysis deconstructs each natural sentence of a manifesto into the smallest possible full grammatical sentence. Chomsky (1957: 45) calls these minimal sentences ‘kernel sentences’. In our terminology they are named ‘statements’ and constitute our unit of analysis. A natural sentence can consist of more than one statement. On average, we extract 2.4 statements from each sentence. In contrast to the CMP approach, our unitising procedure is exclusively based on unambiguous grammatical rules and does not take into account thematic aspects such as a change of the argument that is the basis for the CMP's ‘quasi-sentences’ (Budge et al. 2001: 217–218).
Each statement consists of a ‘subject’ (a political actor, usually the party authoring the manifesto), an ‘object’ (an issue or another political actor) and a ‘predicate’ that numerically records the direction of the relationship between subject and object. In case of approval, the predicate is coded as +1; in case of rejection as −1. In the (rare) case that statements towards issues or other actors do not take a clear position the predicate takes the value of 0.
After a training process of several weeks and equipped with a detailed codebook, two coders independently extract statements from the natural sentences of each manifesto. We test inter-coder reliability with Krippendorff's Alpha, which has become a generally accepted yardstick to assess the quality of content analyses (Krippendorff 2004). The reliability scores for the three most recent Austrian elections (16 manifestos) varied between 0.77 and 0.93, and the mean score was 0.88. These results differ substantially from what Däubler et al. (2012) report from their experimental tests of unitising manifestos according to the CMP rules. Yet given the strict rules of unitising that exclusively rely on grammatical rules and the far more intense coder training, it is hardly surprising that the AUTNES approach produces more reliable results. The two versions of extracted statements are finally merged into one authoritative set by AUTNES researchers.
With respect to the issue content of party statements, our coding scheme is structured into three hierarchically nested levels, which we term ‘issue categories’ (level 1), ‘issue sub-categories’ (level 2) and ‘issues’ (level 3). Coders are trained to allocate each statement into one of over 650 issues on level 3. These fall into one of the 99 issue sub-categories which again belong to one of the 15 issue categories. Two coders independently code the statements obtained in the first stage. At the most aggregate level we obtain reliability scores between 0.81 and 1.00 (Krippendorff's Alpha), on the intermediate level between 0.77 and 1.00, and on the third level with over 650 different categories reliability ranges between 0.63 and 0.74 for the 16 manifestos used in the present analysis. The AUTNES researchers eventually merged two versions of each manifesto coding into one authoritative dataset, thereby resolving inter-coder disagreements.
As the AUTNES coding of party policy positions and saliency is very fine-grained we can aggregate the issues into various policy dimensions. These dimensions were generated in a two-step process. First, each level 3 issue was sorted into exactly one of a number of broad but theoretically meaningful units. Here, it was attempted to capture the core substance of party competition in Austria and, indeed, most West European countries. Thus, the policy dimensions were chosen to closely resemble the dimensions used in previous studies extracting party policy positions from a variety of sources (Laver & Hunt 1992; Benoit & Laver 2006; Kriesi et al. 2006, 2008; Hooghe et al. 2010).
As no generally accepted concept or definition of ‘issues’ exists in the literature, the degree of abstraction remains a major challenge for any examination of saliency theory. Apart from defining contemporary policy areas that are used in the literature on party competition we also constructed policy areas at a level of abstraction comparable to what was originally proposed by the authors of saliency theory (Budge & Farlie 1977, 1983a, 1983b). We do so in order to avoid biasing our test against the theory. These are the categories we use in this article.
In a second step, each issue within such a policy area was allocated to either pole on a pre-defined scale (see Table 1). We thus arrive at policy dimensions on which parties could be placed by averaging the predicate's values from all relevant statements (i.e., statements with issues but not actors as objects). As an example, assume that a party's manifesto contains ten statements on ‘European Integration’, three of which express pro-European (+1) views while the seven remaining are eurosceptic (−1). The party's position on the European integration scale is calculated as follows: (3 * 1 + 7 * −1) / 10 = −0.4.
Table 1. Classification of policy areas
|Policy area||Left pole||Right pole||Overall saliency (%)|
|Taxes||Raise taxes||Cut taxes||4.9||3.1||4.6|
|Welfare||Welfare state expansion||Welfare state retrenchment||18.3||17.2||14.6|
|Regulation||State intervention||Free market||4.5||3.7||4.2|
|Enterprise||Anti-enterprise, anti-corporations||Pro-enterprise, pro-corporations||3.1||2.1||3.9|
|Environment||Less environmental protection||More environmental protection||5.3||8.0||9.4|
|Education||Inclusive forms of education||Selective forms of education||13.0||12.9||12.6|
|Social values||Liberal social values||Conservative social values||7.6||9.4||8.8|
|Security||Civil liberties||Law & order||6.0||4.8||6.1|
|Defence||Weak (militia) military||Strong (professional) military||2.7||1.4||0.8|
|Constitutional issues||Concentration of power||Diffusion of power||5.0||3.5||3.5|
|Europe||Euroscepticism||Support for European integration||4.0||3.9||4.6|
The party-specific saliency of a policy area (or policy dimension) is calculated as its relative share of all statements with issue content. Table 1 reports the content of the 14 selected policy areas, the positional scales that were used to create the respective policy dimensions and their overall saliency in the three elections observed. The first five areas refer to different aspects of the socioeconomic domain: the struggle about the welfare state and its financial basis (i.e., taxes), the (de-)regulation of the economy and issues related to labour or enterprise. As noted in the theoretical part of this article, saliency theory assumes that voters often see no connection between issues that tap into the same domain. This is why we refrain from aggregating economic issues into more widely used categories such as ‘taxes versus spending’ (e.g., Benoit & Laver 2006).
The sixth category, ‘urban–rural’, refers to a further classic aspect of political competition and primarily includes stances on the agricultural sector. ‘Environment’ captures support or rejection of environmental protection measures. ‘Education’ comprises all references to schools, universities and other institutions of the educational system. In order to construct a policy scale that sensibly differentiates the parties' positions, we contrast egalitarian (or universalistic) versus selective programmes on educational matters.
The next three dimensions pertain to the sociocultural realm of politics – more specifically to questions of liberal versus conservative social values, attitudes towards immigrants and ethnic minorities, and the subject of law and order versus the protection of civil liberties. The ‘defence’ category includes all statements in favour or against a strong military. To capture the various aspects of institutional reforms and the parties' statements on the constitution and the political system in general, we refer to ‘constitutional issues’ (cf. Gerring & Thacker 2008). ‘European integration’ is the final category.
Together, these 14 policy areas cover about 86 per cent of the manifestos analysed (2002: 87.2 per cent; 2006: 84.2 per cent; 2008: 85.6 per cent). Remaining parts of the documents refer to subjects not relevant for the present purpose (e.g., statements on the preferred composition of the government, or biographic details of top candidates).3
Apart from our new manifesto data, we will also use survey data in the subsequent analyses. To evaluate H4, the relation of issue ownership and issue saliency, we will work with seven opinion polls conducted by GfK, a polling firm, between 2002 and 2007. In these surveys voters were asked which party is the most competent with respect to various policy areas.4
- Top of page
- Theoretical framework
- A new approach to manifesto coding
In order to examine whether parties, rather than engaging in dialogue, talk past each other (H1), we employ the measure of issue convergence proposed by Sigelman and Buell (2004) for two-party systems (or races with two candidates). Table 2 reports the average degree of issue convergence for each party-pair across the three elections between 2002 and 2008. The overall average value of 0.74 is almost identical to the results reported by Sigelman and Buell (2004: 656) and suggests that about three-quarters of the issue agenda are shared across the six parties. While the figures are somewhat lower for the smaller parties (FPÖ, Greens, BZÖ, LIF) than for the traditional major ones (SPÖ, ÖVP), this is still a remarkable degree of ‘issue dialogue’, especially given that multiparty systems may be much more conducive to single-issue or ‘niche’ strategies that are based on differentiation via selective issue saliency (Meguid 2005, 2008). We therefore conclude that H1 is not confirmed by our data.
Table 2. Issue convergence between party-pairs
|ÖVP||0.82|| || || || |
|FPÖ||0.77||0.78|| || || |
|Greens||0.75||0.74||0.73|| || |
As an important qualification of this result, it should be noted that a more orthodox interpretation of ‘talking past each other’ would be to examine whether parties directly address each other in their manifestos (Budge & Farlie 1983b: 274). This is hardly the case. On average, only 4 per cent of all statements in a manifesto refer to other parties. While there are some remarkable outliers (such as the Greens in 2008 at 20 per cent, or the SPÖ and FPÖ in 2002 at over 10 per cent), saliency theory is right to assume that references to political opponents are rare in election manifestos.
A test of H2 and H3 requires a more systematic look at policy positions and saliencies across issue areas and parties. To get a first visual impression Figure 1 plots the position and saliency scores from the 16 manifestos in this study across the 14 policy dimensions. The horizontal axis represents party positions ranging from −1 to +1. The vertical axis indicates the saliencies that parties put on the issues, ranging from 0 per cent (no statement at all) to (theoretically) 100 per cent, when the entire manifesto is dedicated exclusively to one policy area. As no single category's share is larger than 23 per cent, Figure 1 covers the first quartile on this axis.
Figure 1. Issue positions and saliences of Austrian parties across 14 issue areas, 2002–2008.
Notes: Dashed lines are linear regression slopes. The dotted curves plot the kernel density of the positional spread, thus indicating the extent to which party positions are concentrated at some point in the policy space or spread out across the whole scale.
Download figure to PowerPoint
H2 proposes that parties do not differ in their issue positions within individual policy areas (Budge 2001a: 82). A cursory look at the graphs reveals, however, that parties take diverging positions with respect to many policy dimensions. Otherwise, there would be no spread along the x-axis and the kernel density curves would always exhibit a clear single peak and steep tails. We find support for the saliency logic to be present only for a limited number of issues. Specifically, the pattern predicted by saliency theory is prevalent with regard to enterprise, environment, labour, welfare, urban-rural and, with some qualifications, constitutional issues. Together these six issue areas amount to roughly a third of the party statements in altogether 14 issue areas (see Table 1). Note, however, that such selective emphasis may partly be attributed to country-specific factors such as the non-controversial nature of environmental politics in Austria and the country's longstanding tradition of consociationalism (Luther 1999).
To obtain a more systematic test of H2 we calculate the standard deviations of the parties' policy positions, averaged across three elections, as a rough but straightforward measure of their differences in each policy area. The values range from 0.07 (enterprise) to 0.56 (Europe), with an average of 0.33. Interestingly, the majority of the socioeconomic scales display a low degree of positional variation (values from 0.07 to 0.31), with only regulation exhibiting considerable spread (0.41). In contrast, there is a greater degree of policy conflict when it comes to ‘new cultural issues’ (Kriesi et al. 2006, 2008) such as social values, multiculturalism, security and European integration (values between 0.39 and 0.56). Thus, the overall picture suggests that Austrian parties do differ substantially in their programmatic supply, albeit with considerable variation across issues. Consequently, we refute H2.
The related H3 assumes that parties differ from each other with respect to issue saliency rather than in terms of policy positions. Analytically, this requires a direct comparison between variation in the saliency scores and the positional scales. Yet it is difficult to directly compare these two theoretical concepts of party competition. First, we have to normalise both measures to the range from 0 to 1. However, each normalisation procedure must necessarily make some arbitrary judgements as to the exact specification of the upper and lower extremes of the original scale. While this is straightforward for the positional scores since the empirical distribution covers the whole theoretically possible range from −1 to 1, it is empirically implausible to assert that the range for saliency values is from 0 to 100 per cent as this would require that a party devotes its entire manifesto to one single issue. In order to provide an empirically informed transformation of the saliency scores to the range from 0 to 1, we therefore set 25 per cent as the upper limit, meaning that a saliency score of 25 per cent translates into a value of 1 on the normalised scale.5 While the equivalence of scales measuring different phenomena may always be questioned by referring to differences in their very nature, we go to some length to make our test a fair one. For this purpose, our normalisation procedure is based on the actual distribution of saliency scores shown in Figure 1. As such it clearly – and deliberately – biases our test in favour of saliency theory since the normalisation thus artificially increases the variation in the saliency values.
Even so, Table 3 indicates that only one issue (environment) has a clear ‘saliency’ character: the mean difference between the three pairs of standard deviations is negative (−0.12), indicating that, on average, parties differ much more in terms of saliency than with respect to their policy position. For the majority of issues, however, the mean difference between position and salience is close to 0 or greater. Again, the ‘positional logic’ dominates especially with respect to sociocultural issues, whereas the values for the socioeconomic issues (except regulation) are not clearly in favour of either logic. Thus, the overall conclusion with regard to H3 is that party competition is hardly ever confined to variation in issue saliency. It is dominated by a ‘positional logic’ in at least half of the instances examined here. Except for ‘environment’, the other half of the issues exhibits roughly similar levels of variation in saliency and position. H3 must therefore be rejected.
Table 3. Differences between the parties regarding position and saliency (standard deviations)
| ||2002||2006||2008||Mean difference (Position–Saliency)|
We now turn to testing H4 which postulates that issue ownership explains issue saliency. To be sure, there are many other factors that can explain why parties emphasise an issue at the expense of another. Some issues may be of general importance in a campaign and can only be avoided at a high cost, especially in highly competitive elections (Kaplan et al. 2006). External shocks such as natural disasters or acts of terrorism may have a sudden impact on the campaign agenda. Also, parties may tailor their issue profile to fit a specific constituency. At times, a party may have accumulated a track record on some issue that is worth emphasising (Sides 2006). Conversely, a party may want to direct attention to an incumbent's dismal performance in office.
While we recognise these potential influences, we are only interested in the relationship between issue ownership and saliency. To that end, Figure 2A plots the party-specific saliencies of issues against the parties' ownership of these issues for each election. Each dot therefore represents one of the 116 party–issue pairs that emerge from the data. In calculating these ownership values we assume that voters holding a specific party preference are likely to consider their respective party most competent with regard to most, if not all, issues. We thus have to find a more realistic assessment of the variation in issue ownership (see Müller 2000: 36–41). The horizontal axis in Figure 2A, therefore, shows the percentage of respondents ascribing the highest issue competence to a party minus the percentage of respondents expressing a voting intention for that party in the same survey. Hence, values greater than 0 indicate a positive net (or relative) issue ownership. The vertical axis displays parties' deviation from the median issue saliency for each category. Positive values thus designate above-median saliency when compared with the other parties at a given election. Figure 2B plots the regression lines per issue.
The data plotted in Figure 2A correlate with r = 0.54 (p-value < 0.000), indicating a highly significant positive correlation of medium strength. Clearly, this result is heavily influenced by four party–issue pairs for which the relative ownership percentages exceed 30 per cent.6 Yet even after removing these four pairs from the analysis, the correlation coefficient is still at 0.36 (p-value < 0.000). Figure 2B shows that while there is substantial variation across issues, the relationship between saliency and issue ownership is positive for all policy areas except one (social values). All things considered, our data thus support H4. Parties indeed stress their ‘owned’ issue areas in their election manifestos.
Finally, we test H5 by examining the relationship between the absolute distance from the median party position7 and the deviation from the median saliency. Each dot in the left-hand graph in Figure 3A represents a party–issue year. We use the relative deviation from the median saliency (as the saliency values are not independent from each other and sum up to 100) and the absolute deviation from the median issue position (since the direction of the deviation is irrelevant to the hypothesis). At r = −0.12 (p-value = 0.09), the correlation is very small, thus indicating, at best, a weak association. Figure 3B plots the regression slopes for the relationship between position and saliency by issue. It illustrates that there is no consistent link, with some negative and positive slopes.
However, closer inspection of the graph reveals that there is some support for H5. The fact that the top right quadrant of the graph is not populated at all signals that while parties do not necessarily emphasise all issues on which they take an electorally beneficial position, there is hardly any case where a party significantly deviates from all others in terms of policy position while at the same time putting disproportionate emphasis on the issue. A generous interpretation of the graph may therefore lead to the conclusion that the weak correlation coefficient understates the relationship actually present in the data. Still, we believe that the evidence is too sketchy to confirm H5.
- Top of page
- Theoretical framework
- A new approach to manifesto coding
Applications of saliency theory have widely contributed to the understanding of contemporary party competition. Especially the CMP, which can be regarded as a direct implementation of this specific theoretical approach, has produced an enormous amount of data that has become extremely important in party research. In this article we have made explicit the core ideas and underlying assumptions of saliency theory. These, in turn, have exerted a strong influence on the nature of the CMP's data. A fair test of saliency theory therefore cannot rely on them. We therefore turn to data that have been generated within the AUTNES framework for manifesto analysis and allow for independent measurements of party issue saliencies and party issue positions. Our analysis covers one decade of party competition in a single country. It analyses positions and saliencies on 14 issue dimensions based on 16 manifestos issued by six parties of diverse ideological character and varying government/opposition status over the course of three elections.
Although idiosyncrasies can never be ruled out with certainty on the basis of single-country studies, we assume that the Austrian case is representative of party competition in many multiparty systems in Western Europe and thus allows for some generalisation. Saliency theory was initially conceptualised in the context of two-party systems (the United States and the United Kingdom), suggesting that a multiparty system may not be the most appropriate testing case. Yet, the original authors of the theory claim that it is valid for all party system types (Budge & Farlie 1983b: 281).
According to our results, some of saliency theory's core assumptions do not conform to the parties' actual behaviour. Let us summarise our findings with respect to the five hypotheses that contain the core ideas of saliency theory. Our test of H1 shows high degrees of issue convergence, especially for larger parties. By this measure there is a considerable degree of confrontation over the same issues. In contrast, the more orthodox reading of saliency theory – that parties do not mention their opponents in their manifestos (Budge & Farlie 1983b: 274) – can be confirmed. On average, only 4 per cent of all statements refer to some other party or party actor. The following implications of saliency theory could be refuted without much reservation: It is simply not the case that most parties gravitate towards a similar position in each policy area (H2) and differ more with respect to the saliencies they put on the issues than in terms of issue positions (H3). Note that we deliberately based our test on manifesto data. Other sources revealing party strategy, such as party press releases, television debates, or news media reports, appear a priori less plausible to show that parties use ‘selective emphasis’ rather than ‘direct confrontation’. Theories of issue ownership, on which saliency theory builds, explain which issues parties accentuate in their manifestos (H4). Here, we found a clear correlation between voter perception of parties' issue competence and the degree to which a party emphasises or downplays a certain issue. We thus confirm one central claim of saliency theory. Finally, the test of H5 showed no unambiguous support for the notion that issue saliencies diminish as a party's distance from the median position increases.
In sum, our examination of the empirical fit of saliency theory suggests that it correctly identifies some features of party competition. Such partial success is no mean achievement for a theory in the social sciences. Yet, its core assertion – that parties compete via selective issue emphasis rather than direct confrontation over the same issues – fails to materialise in the majority of cases. This is all the more remarkable given that the empirical focus on party manifestos is likely to push the theory's predictive power. Even so, direct confrontation over policy appears to be the dominant pattern of party competition. Given that saliency theory was originally developed as a fundamental alternative rather than a partial complement of confrontational theories, it thus falls short of the authors' original ambitions and claims. However, our results suggest that some policy issues are more prone to the saliency logic of party competition than others. Future analyses of party competition should therefore attempt at identifying the specific conditions under which parties are more likely to adopt a strategy based on selective emphasis rather than direct confrontation.