Party crashers? Modeling genuinely new party development paths in Western Europe

As discussed above, focusing on the number of party breakthroughs and their support in the initial election only paints a partial picture of GNP performance and their longer term impact on party systems. In this article, we go beyond breakthroughs and study the evolution of party support following their entry. We rely on class-based LGM (Jones and Nagin, 2012) that allows us to classify parties so that within each class they follow a similar trajectory over time. For example, some parties (e.g. ethnic parties) may be expected to enter with a moderate support and maintain it over time. In a linear model, an intercept (initial vote share) would suffice to describe this trajectory. However, parties with similar initial support can vary – some disappearing quickly (negative slope), others increasing support and challenging established parties (positive slope). A simple linear model clearly offers limited value as it only allows for unchanged support or change at a constant rate (Figure 5(a)). While parties can maintain its initial support, a constant rate of increase is less realistic as support can peak and certainly cannot continue eternally. A decrease at a constant rate is more problematic – even for parties with high initial support, this is likely to predict diving to negative vote shares.

Acknowledging these limitations, Mustillo (2009) models party trajectories using linear and quadratic functions. Still, quadratic terms provide only a partial cure (Figure 5(b)). They allow for curvilinear trajectories and flipping party fortunes but impose a perfect symmetry around the peak or bottom support. Upon reaching a peak, party support must decrease following a trajectory that exactly mirrors its rise. This is unrealistic and can again take parties into the impossible zone. Even more problematic are U-shaped quadratic trajectories that may predict that after a demise (bottoming at zero support), parties will raise like a phoenix from the ashes, mirroring the path of its downfall.

Therefore, we have used LGM with a beta link function implemented in the lcmm (latent class mixed models) package in R (Proust-Lima et al., 2017).⁹ lcmm and the beta link function allow for flexible and meaningful modelling of party trajectories. Flexible because lcmm does not require a specification of the order of the functional form (constant, linear, quadratic etc.) before the analysis as required by the traj module in Stata (used by Mustillo). Meaningful because the beta link function imposes lower and upper boundaries on party vote shares; it assumes that trajectories are caught between zero and the maximum vote (for all parties in the model). Thereby, it becomes impossible for parties to fall below zero or grow perpetually. Moreover, lcmm allows for a more efficient estimation with fewer parameters compared to the quadratic model.

The LGM estimates ‘shape parameters’ – coefficients for time variable(s) and for the link function – for a number of latent trajectories set by the investigator. Population prevalence – how many parties fall into each class – is estimated for each of the trajectories. We compared models with a different number of latent trajectories and alternative specifications.¹⁰ For choosing the optimal model, we followed statistical criteria, principally the Bayesian information criterion (BIC).

Having settled for ‘the best’ model, we found posterior class membership probabilities for each of the parties. These indicate the probability that a given party belongs to a particular class and allows to predict the future performance of recent GNPs – some of our most remarkable cases. It may be too early to assign these parties to any classes because we do not know its subsequent performance. For example, Emmanuel Macron’s LaREM has only contested one election and we do not dare to prophesize whether in future it will increase or decrease its support or even disappear. The model provides a posterior probability based on the information available – for LaREM, its vote share in 2017. Class membership can be determined more accurately for GNPs that have contested more elections. While uncertainty related to incomplete trajectories is unavoidable, great uncertainty for parties with complete trajectories would suggest a poorly fitting model. Conversely, a model with a good fit is one that classifies most parties with high class membership probabilities (Mustillo, 2009: 321).

We tracked the electoral performances of GNPs in the first five elections, starting from the one where they won at least 1% of the vote for the first time. We choose five elections following Mustillo’s argument that ‘the early years of a party’s life are developmentally distinctive from its mature years. The farther in time that we go from birth, the more the result will incorporate elements of a party’s life that are unrelated to early developmental characteristics’ (2009: 323). In other words, five elections provide sufficient time to observe the development of a new party into maturity, without incorporating later developments that have little to do with its initial path.¹¹

Our data set contains 385 party–election dyads; an average GNP contested 3.03 elections.¹² We distinguish between parties that cease to compete or merge with other parties and recently emerged parties that suffer from right-censoring. Otherwise, a disappearing party would look healthier than it actually is (Mustillo, 2009), Therefore, for exiting parties, we recorded a zero vote share after their last election; however, lcmm allows for the inclusion of recent parties (i.e. right-censored cases) in the modelling of trajectories and, thereby, also allows for predictions about the potential further development. We estimated models with one to six latent classes (i.e. party trajectory classes) to find a model with optimal fit based on BIC. Moving from one to five classes lowered BIC but adding a sixth class resulted in a poorer BIC. Hence, we found the five-class model to be optimal.¹³

We labelled the five classes – mostly echoing Mustillo’s terminology – as ‘explosive’, ‘meteoric’, ‘contender’, ‘flat’ and ‘flop’. Their prevalence is fairly balanced: unsurprisingly, ‘flop’ parties constitute a majority (53.3%), the other half have experienced different starting levels and trajectories, going from the ‘flat’ one (13.4%) to the ‘explosive’ one (11%). Figure 6 (and Table B3 in Supplemental Material B) provides an overview of the optimal five-class LGM.¹⁴

The first class are explosive parties: their entry causes an electoral earthquake that leads to party system restructuring. ‘Explosive’ parties are strong from the beginning (mean support 12.7% in first elections), and their performance further improves in subsequent elections. As expected, given the bias towards stability, such parties have been rare in Western Europe. We find only 14 parties (11% of GNPs) that have followed the ‘explosive’ trajectory (e.g. the French Gaullist party, GAP,¹⁵ and Forza Italia, FI) or have a potential to do so in future (e.g. Italian M5S and the Spanish Podemos; see Table 2). Both M5S and Podemos are classified as ‘explosive’ given their high support in the first election and improvement in the second election (M5S: 25.6% and 32.7%; Podemos: 20.8% and 21.3%). While this coincides with the swift institutionalization of these parties, the real test of the classification lies in their future performance. Interestingly, many parties in this class resemble ‘anti-establishment reform parties’ that combine ‘mainstream ideology on economic and socio-cultural issues with fierce anti-establishment rhetoric and demands for political reform, transparency and new ways of “doing politics”’. (Hanley and Sikk, 2016: 522). Parties like LaREM, M5S and Ciudadanos certainly fit this definition (Bordignon and Ceccarini, 2013; Lavezzolo and Ramiro, 2018). Likewise, GAP and FI, surfaced with strong anti-establishment rhetoric, becoming mainstream conservative parties only afterwards.

The second class of meteoric parties – borrowing a term from Taagepera (2006) – contains only two parties – the Dutch Pim Fortuyn List (LPF) and the Portuguese Democratic Renewal Party (PRD). They are characterized by an even stronger electoral performance in their first election than the ‘explosive’ class (17.7% on average). After a remarkable entry, ‘meteoric’ parties dramatically decline in the following elections and then disappear. Notably, LPF and PRD both nose-dived after the parties supported, but did not lead, the government after their maiden election; conversely, of the 14 explosive parties, only 1 (the Danish Progress Party, Z) provided the support for the governing coalition after their first election while two (Forza Italia and En Marche) led the government themselves. Our analysis suggests that parties entering with a high vote share (above 13%) are typically ‘explosive’ and likely to become part of the respective party system. ‘Meteoric’ parties diverge from ‘explosive’ ones after the first election as the confidence intervals for the trajectories only overlap in the first election.

Contenders enter with a meagre vote share (2.3% on average in first election) but subsequently increase their support and challenge established parties. Roughly one-fifth of all GNPs, this class is home to many ‘populist radical right parties’ (Mudde, 2014) such as the Belgian People’s Union (VU), the True Finns, the Norwegian Progress Party, the Sweden Democrats and the UKIP. It also includes the post-materialist Dutch D66, the German radical left PDS-Linke, and the Dutch Socialist Party (SP). SP shows the prototypical trajectory of ‘contenders’: it won 1.3% of votes in 1994 and continuously increased support, eventually reaching 16.6% in its fifth election (2006). ‘Contenders’ also include 10 particularly successful green parties (in Austria, Germany and Sweden) clearly distinguished from their ideological bedfellows elsewhere that have mostly followed ‘flat’ and ‘flop’ trajectories.¹⁶

The most populated class is that of flop parties that contains more than half of the GNPs (53.5%). It consists of parties entering with a low vote share (2.4% on average) and rapidly falling into obscurity, usually after the second or the third election. In other words, these parties (and their ‘midwife voters’) probably overestimated their potential in the first place. This corroborates the idea that political entrepreneurs operate with bounded rationality (Simon, 1957). Many new parties should not be formed at all under full information on institutional constraints, electoral potential, competitor strength and so on (Hug, 2001).

While the flat trajectory shows the lowest initial performance (1.8% on average in the first election), unlike ‘flopped’ parties, their support does not fade away and stays fairly constant over time. This class contains 17 GNPs that are small but retain their support over time. Typical members of the class are the Green parties but also other niche parties (Meguid, 2005), such as the Dutch Party for the Animals, or ethno-regionalist parties (De Winter and Türsan, 1998), such as the Republican Left of Catalonia (ERC) and the Scottish National Party (SNP).

The finding that ‘flop’ parties is the most populated class is meaningful. It shows, on the one hand, that our results are consistent with the ‘bias toward stability’ argument (first and foremost, Bartolini and Mair, 1990).¹⁷ On the other hand, it tells us that notwithstanding the recent wave of electoral instability (Chiaramonte and Emanuele, 2017; Hobolt and Tilley, 2016; Hérnandez and Kriesi, 2016), stability has been the fundamental feature of West European politics since 1945. More than two-thirds of GNPs in Western Europe did not significantly alter their countries’ party systems since they either rapidly disappeared (‘flop’) or maintained only niche support (‘flat’).

The posterior probabilities confirm the good fit and internal validity of the model. After excluding the 26 parties that have not contested five elections yet – for which a high level of uncertainty is expected – only 11 parties have a lower than 90% probability of membership in their class.¹⁸

Our classes are somewhat different from those discovered by Mustillo (2009). Not only did the trajectories fail to mirror exactly the Latin American ones but the prevalence of classes also differed.¹⁹ Among the 299 Latin American new parties and 758 party-year vote percentage observations, almost 90% of the cases belong to the ‘flop’ category, while the remainder are divided between ‘flat’, ‘contender’, ‘explosive’ and ‘flash’.²⁰ Differences were to be expected given the different nature of party systems in the two regions, particularly the level of party system institutionalization (Chiaramonte and Emanuele, 2017; Mainwaring and Scully, 1995; Sanchez, 2009). Overall, by comparing our findings with Mustillo’s ones, it clearly emerges that the birth of new parties in Western Europe is far less common than in Latin America, but once created, European new parties are more likely to persist over time.

While different from Latin America, Western Europe is not homogeneous in terms of new party development. Only eight of the 20 countries witnessed ‘explosive’ competitors (Table B6 in Supplemental Material B), with France and Italy at the forefront of party system transformation with three ‘explosive’ shocks each (Emanuele and Chiaramonte, 2018: 483). Conversely, ‘contenders’ are particularly common in Belgium (five, no other country has more than two) and ‘flat’ parties in Switzerland (four). These findings are understandable considering historical context. On the one hand, the traditionally consensual politics of Belgian elites have been prone to include new competitors into governing coalitions (Casal Bértoa and Enyedi, 2016; Lijphart, 1999). Accommodating elite behaviour clearly favours institutionalization and endurance of recently emerged parties. On the other hand, the highly decentralized Switzerland with its multiple cross-cutting cleavages is a fertile ground for small parties that maintain their niche support (Caramani, 2004; Lipset and Rokkan, 1967). Finally, the two smallest countries in our data set – Iceland and Luxembourg – interestingly account for nearly a quarter of ‘flopped’ parties. This could be due to the combination of the relative ease of setting up new parties and mobilizing some electoral support in small countries, and not particularly permissive electoral systems.²¹ More generally, the role of electoral systems is limited. GNP trajectories in terms of seats are extremely similar to electoral ones (see Figure A1 in Supplemental Material A), except for the predictable parliamentary over-representation of stronger ‘explosive’ parties and an under-representation of all others. As expected, a majoritarian rule favours ‘viable’ parties that are strong enough to win a considerable number of votes in single-member districts (SMDs) (typically ‘explosive’) and small parties with a territorially concentrated support (e.g. SNP in the United Kingdom). Conversely, PR increase the chance of small party survival that may grow over time and become ‘contenders’ that is difficult in majoritarian systems, where ‘contenders’ must become very strong to challenge established parties in SMDs.

So far, we have focused on GNPs trajectories in the first five elections. But what happens to GNPs following the distinct trajectories afterwards? Do five elections provide a good guide to party’s future development, as we initially hypothesized? Figure 7 plots the trajectories of each class beyond the first five elections. On average, the support for parties that lasted five elections remains remarkably stable afterwards, although there is significant variation between individual parties. All ‘meteoric’ and ‘flop’ parties disappeared before the fifth election, the average support for other parties did not fall up to their 10th election and beyond. If anything, they grow stronger past the fifth election. Given that most ‘explosive’ parties are recent, the long-term growth reflects two pairs of parties that now represent different faces of conservatism in France and Italy: the Gaullist Party and the National Front; and Forza Italia and the Northern League.²²

Of the 26 ‘contender’ parties only 3 have by now exited; the rest are still competing elections with an average vote share of almost 10%. The result is interesting given the meagre support at the time of their entry and the class clearly deviating from the typical pattern for parties with low initial support. Indeed, most parties with 2–2.5% of votes in their initial election are doomed to ‘flop’ or maintain ‘flat’ support. Yet, even ‘flat’ parties tend to increase support after the fifth election, approaching the ‘contender’ class by election 15 (when confidence intervals overlap). This increase is mainly due to a class of parties that have been able to reach out beyond their niche support, like the Swedish Christian Democrats (11.8% by their 12th election in 1998), the French Greens and the Scottish National Party.

Overall, the fact that the classes that persist after the fifth election remain stable over time suggests that five elections provide a good guide to party’s future development. In other words, just like in a path dependent process (Pierson, 2000), the early stage of electoral development of GNPs is a critical juncture that shapes its future life path. This also hints that latent trajectories have some predictive capability. Given that parties in different classes tend to follow a trajectory over time, we can formulate future predictions based on first five elections. Some fundamental estimates may be already based on the vote share in the first election. First, a new party debuting with more than 12% of the votes is three times more likely to become an ‘explosion’ than a ‘meteor’. Second, a party emerging with a modest vote share (e.g. between 1% and 3%) has a 60% probability to fail shortly, a 17% chance of maintaining its niche, and a 23% chance of becoming a ‘contender’ in the respective party system. Finally, for a party with 3% to 12% in the first election uncertainty looms as virtually all scenarios are open: a quick failure is the most likely outcome (48%), but it holds almost the same chance of becoming a ‘contender’ (17%) or even an ‘explosive’ actor (28%).