Party and Party System Institutionalization

In the analyses of party politics, PI and PSI are often collapsed into a combined fuzzy syndrome (e.g., Kreuzer and Pettai Reference Kreuzer and Pettai2003; Lewis Reference Lewis2006; Meleshevich Reference Meleshevich2007; Moser Reference Moser1999; Rose and Mackie Reference Rose, Mackie, Lawson and Merkl1988; Shabad and Słomczyński Reference Shabad and Słomczyński1999). This is, on the one hand, understandable. Both individual parties and the patterns of party competition can become predictable, anchored, robust, impersonal, and routinized, and there are strong reasons to expect the institutionalization of individual parties and party systems to be related (Lane and Ersson Reference Lane and Ersson2007; Rose and Munro Reference Rose and Munro2009; Sartori Reference Sartori and Mair1990). On the other hand, we will never learn about their similarities and differences unless we rely on separate empirical indicators.

Even when the party level and the systemic level are treated separately, the features of the units (i.e., parties), primarily their organizational strength and stability, are used to characterize the whole system (e.g., Mainwaring Reference Mainwaring1998; Mainwaring and Scully Reference Mainwaring and Scully1995; Mainwaring and Torcal Reference Mainwaring, Torcal, Katz and Crotty2006; Kuenzi and Lambright Reference Kuenzi and Lambright2001; Tan Reference Tan2006). In other words, there is a mismatch between theory and empirical strategy.

The confusion of the two levels can be minimized if one conceptualizes the institutionalization of parties primarily through organizational continuity and longevity (Levitsky Reference Levitsky1998; Bolleyer and Ruth Reference Bolleyer and Ruth2018; Scarrow, Wright, and Gauja Reference Scarrow, Wright and Gauja2023), while thinking of systemic institutionalization in terms of the stability of the structure of inter-party competition (Mair Reference Mair, Karvonen and Kuhnle2001). Since there is no deterministic mechanical link between these factors, we allow for the possibility that the patterns of partisan interactions may solidify despite the discontinuity of individual parties or that a substantive change in the dynamics of the system may occur even without a drastic change in the support of parties and without the appearance of new players on the scene. At the opposite end of the spectrum, relationships among parties may remain unpredictable and unstable despite the persistence of party labels.

For measuring institutionalization at the systemic level, we choose the party-system closure index (Mair Reference Mair, Karvonen and Kuhnle2001). Among the available empirical proxies for PSI (e.g., fragmentation, electoral volatility, programmatic stability, etc.) this is the one that focuses most unambiguously on the interactions among parties. Party-system closure is considered high if the combinations of parties constituting governments rarely change, if new parties have a marginal presence in governments, and if governments are not based on the combination of hitherto opposition and government parties. While the closure scores are calculated using information obtained from the governmental arena, they have been shown to reflect the logic of party relations in general (Casal Bértoa and Enyedi Reference Casal Bértoa and Enyedi2021). The reliance on the composition of governments permits us to extend our investigation to a particularly extensive large timespan.

As far as the party level is concerned, the same objective can be reached by focusing on the age of parties and their continuous electoral support. Other possible indicators of PI (e.g., indices of party identification, membership, party switching, party replacement, party nationalization, bureaucratization, financial resources, etc.), are usually available for a limited number of cases over time. Even more importantly, age and electoral support are relevant indicators of PI because they depend directly on the organizational stability and social rootedness of the parties.

This operationalization strategy allows for the examination of the differences and similarities between party and systemic institutionalization across time, and for establishing the causal sequence between the two. In general, based on the literature and on the arguments to be elaborated further later, one should expect a positive relationship between these phenomena. But this relationship is likely to be less tight than assumed by the studies that mix information from the two levels.

Party and Party System Institutionalization: What Kind of Relationship?

Despite the importance of both discussed notions, few studies have probed into the causal relationship between them. Within this relatively small literature, PI is mostly assumed to be a cause of PSI (Roberts and Wibbels Reference Roberts and Wibbels1999; Toole Reference Toole2000), but some studies treat it as a consequence of the latter (Tavits Reference Tavits2008).Footnote ³ Stable party interactions are unlikely when political parties come and go and frequent changes in party relations may hinder the ability of parties to develop electoral roots and stable organizations. Thus, intuitively, both models are conceivable.

The model according to which continuity in party relationships anchors and stabilizes individual parties is definitely plausible. This logic assumes that when parties do not change partners, new party initiatives are discouraged, and voters are reluctant to change partisan preferences (Mair Reference Mair1997). In contrast, if the electorate sees parties colligating in an ad hoc and pragmatic manner, making alliances out of convenience rather than based on long-standing common interests and ideological proximity, the loyalty of voters and politicians to their party will diminish, making both electoral shifts and organizational splits more likely (Tavits Reference Tavits2008). In other words, uncertain party relations may weaken political parties themselves.

The opposite model considers the institutionalization of parties to be the more profound and more consequential phenomenon and expects the systemic level to stabilize only after political parties achieved considerable organizational continuity and gained a critical mass of loyal supporters. We find the arguments in favor of the second model to be more powerful, and therefore we expect PI to precede PSI. The first argument is quasi-mechanic: predictable relationships between parties require some continuity in the basic building blocks of the system (Moser Reference Moser2001, 36; Rose and Munro Reference Rose and Munro2009). In theory, a center-left party can be replaced by another center-left party that will have the very same systemic role as its predecessor, but it is highly unlikely that this logic will characterize every single example of party discontinuity.

More substantively, PI implies routinized internal processes for recruitment, deliberation, conflict resolution, and decision-making (Harmel, Svåsand, and Mjelde Reference Harmel, Svasand and Mjelde2019). The establishment of conflict-solving mechanisms reduces the room for intra-party factionalism and dissent (Mader and Steiner Reference Mader and Steiner2019) and increases the chances for internal cohesion to develop (Harmel and Svåsand Reference Harmel and Svåsand2019; Basedau and Stroth Reference Basedau and Stroh2008). In a political system dominated by such parties, there is less reason for voters to abandon parties and to consider newcomers (Gherghina Reference Gherghina2015; Tavits Reference Tavits2013). The smaller magnitude of “homeless” and “wandering” voters is helpful for systemic-level stability: abrupt shifts in electoral balance from one election to another can easily undermine the consolidation of inter-party competition (Ponce and Scarrow Reference Ponce and Scarrow2022).

Furthermore, strong and recognizable party brands are typically associated with relatively stable ideological platforms (Lupu Reference Lupu2016, Rosenblatt Reference Rosenblatt2018), promoting recurrent alliances and coalitions (Mainwaring and Zoco Reference Mainwaring and Zoco2007). In those systems where parties lack a secure core of supporters, stable partisan relationships will be more difficult to achieve, as parties will be more inclined to change their ideological stances, confounding both voters and politicians as to what they stand for.

The routinization of coalition preferences is also helped by the fact that established parties are likely to be present across the whole nation and to have professional staff (Szczerbiak Reference Szczerbiak2001). Professional politicians who are in party office across longer time periods are more apt at striking cross-party deals and at holding each other accountable.

To conclude, while the causal arrow running from the systemic to the party-level also appears to be plausible, is likely to be weaker and more indirect than the one running from the party level to the systemic level. A party system can be expected to institutionalize primarily when it has a continuous core of parties. When that is not the case, we can expect an open arena of government-building, and constant coalitional innovations. Therefore, we hypothesize that in most cases the institutionalization of political parties will lead to the institutionalization of the party systems, and not the other way round.

Data and Operationalization

We make use of a new dataset comprising all European countries that have had a meaningful experience with democracy since 1848. A country is considered to be democratic when 1) it has a score of ≥6 in the Polity IV index, 2) universal (male) suffrage elections have been held at least once, and 3) governments are formed with (and rely on) parliamentary support rather than on the exclusive will of the head of state (Casal Bértoa and Enyedi Reference Casal Bértoa and Enyedi2022). The full dataset spans 171 years (from 1850 to 2021) and 60 different historical party systems across 45 countries (refer to table 1 for details). The number of electoral cycles varies between 1 (e.g., Greece’s post-World War II Kingdom or Poland’s First Republic) and 35 (Switzerland). Partial and constitutional assembly elections are not taken into consideration except when explicitly stated otherwise. The number of cabinets per country varies between 2 (post-World War II Turkey and inter-war San Marino) and 61 (France’s Third Republic). The full data set thus covers 729 elections, 878 parties, and 1,475 instances of government formation. All in all, this constitutes almost the entire universe of cases for European democratic party systems.Footnote ⁴

Table 1

Party systems of democratic periods included in the dataset

By default, the attributes of party relations of specific (democratic) countries are used to characterize the party system of that country. In this sense, countries and party systems could be used as synonymous labels for our units of analysis. But some countries had a ruptured history, with democratic periods being divided by—often lengthy—non-democratic spells, leading to a profound transformation of both party-political actors and their relations. In such instances we talk about different “party systems” nested in the history of a particular country. The temporal boundaries of such party systems are provided by the continuity of democratic arrangements, and the scores refer to the (relational) features of party politics within the uninterrupted democratic spells. In other words, our actual units of analysis are democratic periods. Typically, the entire democratic history of a country is one single spell, but this is not always the case. This is why, for example, we distinguish between four separate French systems in table 1.

Operationalization of Main Variables

As mentioned earlier, PSI is measured by the closure index. The empirical basis of the index is provided by data on the party composition of governments and on the partisanship of the ministers. The index has three components: alternation, formula, and access. The alternation sub-index shows a low score if partial alternations are the rule, meaning that some of the ruling parties continue in office while others do not, and cabinets are a mix of parties from within the previous government and from without. The formula sub-index has low scores if there are frequent innovations in the coalition formulae. Finally, the access sub-index has low scores if new parties have easy access to government. Low scores are taken as indicators of open, under-institutionalized party systems. In contrast, we speak of highly institutionalized (closed) systems if the incumbent parties tend to leave office together or if governments tend to remain in office across long time periods, if governments are typically based on fixed alliances of parties and if they consist of parties that have already governed in the past.

To indicate the degree of closure for a particular year, we use a weighted average measure for each particular year that takes the year in question as well as all previous years of the party system into account. The weighted average closure for year n of the party system takes all years from 1 to n into account. Weights for each year are the year number divided by the triangular number of n. This results in a linearly decreasing weighting scheme, where the more recent years are given more weight and the weights would reach 0 beyond the beginning of the party system.Footnote ⁵

To capture the level of PI, we introduce two variables. The first is the average age of parties, widely used in the literature (see Dix Reference Dix1992; Huntington Reference Huntington1968; Kuenzi and Lambright Reference Kuenzi and Lambright2001; Mainwaring and Scully Reference Mainwaring and Scully1995; Mainwaring Reference Mainwaring1998; Roberts and Wibbels Reference Roberts and Wibbels1999; Tan Reference Tan2006). In order to calculate this indicator, we have collected data on the year of foundation of political parties from multiple sources, including party statutes and official websites. While party mergers are considered cases of organizational rupture (e.g., Liberal Democrats in the UK, Christian Democratic Appeal in the Netherlands, Freedom Union in Poland), simple name changes (e.g., People’s Party in Portugal and Spain) are not. In the case of splits, the age of the splinters starts from the moment of the split, while any successor party will not have its age re-started. We consider all those parties that obtained at least 3% of the national vote at least once in their lifetime.Footnote ⁶

Capturing organizational continuity exclusively through average age gives an advantage to older democracies. Additionally, a simple average is not sensitive to the fact that some parties are more important than others. In order to make up for these deficiencies, the employed PI measure also includes the Established Party Domination (EPD) index (Casal Bértoa and Enyedi Reference Casal Bértoa and Enyedi2021, 150-151). The index has a high score if voters tend to cast their vote for parties that were established at the beginning of the party system. Each party contributes to the index with its vote share plus 5% Footnote ⁷ for each year since the party first participated in elections. The index calculates a reference score and an actual score for each election. The reference score for an election assumes that all parties have endured since the founding elections,Footnote ⁸ while the actual score takes their actual durations into account. The EPD index is the ratio of the two.

Considering the vote share of a party implies a contrast between the analyzed party and the total vote. This may sound like bringing in the systemic level, but the goal is simply to capture the relative weight of the party in the electorate and not to tap the relations among parties. Because EPD is calculated differently than party age, it allows for tapping a different layer of PI. It can be high even in a situation where the average party age is relatively low, for example in those newly democratized party systems in which the very same parties are returned to the legislature election after election.

We use the mean of the standardized values of both components, average party age and established party domination as an overall indicator of PI.

The descriptive statistics for the indices and their components across all the years and countries in our full data set (n=2400) (Casal Bértoa, Enyedi and Mölder, Reference Casal Bértoa, Enyedi and Mölder2023) are provided in table 2 and their correlations are shown on figure 1. The correlation between PI and PSI is 0.42. This indicates that the overlap between the two phenomena may be smaller than often assumed, at least using the previously introduced indices. If we look at the components, we see that the relationship is driven by average party age being related to the formula- and access-component of party system closure. Established party domination and alternation play a lesser role in the overall association between party and systemic institutionalization.

Table 2

Descriptive statistics

Figure 1

Correlations between PSI and PI and their components

Differences across Time and across Systems

The temporal extent of our data set allows us to trace the co-evolution of party and systemic institutionalization in European democracies over more than a century. Figure 2 looks at the two levels of institutionalization aggregated by 5-year periods from 1900 (the earlier nineteenth-century data are too sparse for identifying trends) to 2021. The trajectory of PI across this period has an inverted-U shape with a steep increase followed by an equally steep decline. These data demonstrate, dramatically, the advent and the wane of the mass party era. Political parties were central institutions in the class-dominated politics of the 1950s and 1960s, but the subsequent social transformations led to the appearance of new forces, mergers, and splits of the established actors, and a general appetite for experimentation on behalf of voters.

Figure 2

Average party and party system institutionalization over time (1900-2021).

The process of de-institutionalization is not an exact mirror-image of institutionalization. The latter process was part of social modernization and democratization. De-institutionalization does not require a rejection of the norms of modernity and liberal democracy. On some aspects these norms may even be supported by the decline of bureaucratic mass organizations.

De-institutionalization is primarily related to phenomena like the personalization of party politics, the professionalization of party elites, the rise of the “party in public office” at the expense of the “party in central office” and of the “party on the ground,” and the extension of public funding of parties (Katz and Mair Reference Katz and Mair2018). These phenomena amplified instability and sped up the process of the voters’ withdrawal from the electoral arena (Mair Reference Mair2013; Ignazi Reference Ignazi2017; Scarrow, Webb, and Poguntke Reference Scarrow, Webb and Poguntke2018). Socially rooted parties were replaced by entrepreneurial parties (Bolleyer Reference Bolleyer2013). The rise of populism also contributed to the spread of fluid and personalistic organizational style (consider Podemos, Five Stars Movement, Brexit Party, or the Czech ANO). Together, these processes led to a sharp decline of PI, especially after the 2008 Great Recession.

Party-system closure shows less dramatic changes. Between the 1920s and the 1980s the trend was towards (moderate) growth. The turbulence of the 1990s affected closure negatively, but afterwards systemic institutionalization proved to be more resistant against the forces of destructuration than PI. Having said that, the last half century produced less predictable party systems than the previous one. This trend applies whether the new post-communist party systems are excluded or not from the analysis.

As it follows from table 3, stability in parties and in party relations need not go hand in hand. About one quarter of the cases are characterized either by relatively institutionalized parties but not systems, or the other way round. The configuration of chaotic inter-party relations despite the existence of socially rooted and organizationally continuous parties is well exemplified by Finland. During the interwar period virtually all combinations among the Socialists, the Agrarians, the Communists, the Liberals, the Conservatives, and of the party of the Swedish minority, were tried. Most governments relied on the right-wing and centrist parties, but the Socialists were also able to form a government. Politics changed significantly after World War II, but the logic of complex re-combinations of well institutionalized parties continued. The same applies to the French Third Republic, where the individual parties achieved a high level of institutionalization, but the system remained unstable. With the exception of the Communists all major political forces participated in coalition-making. Out of the 61 governments only 15 had a familiar party combination.

Table 3

Party and party system institutionalization across 60 political systems

At the opposite end of the spectrum, in Fifth Republic France, Hungary, the Netherlands or post- World War II San Marino, to name a few, the splits and disappearance of individual parties did not undermine the development of meaningful systemic level patterns, providing these cases with relatively high closure scores and low PI scores.

The relatively weak correlations, the occasionally diverging temporal trends, and the existence of cases that excel on one dimension but lag behind on the other one indicate that PI and PSI, while related, are still different phenomena. However, can one of them be considered to be the cause of the other? This is a question we address in the following section.

Parties or Party Systems: The Chicken or the Egg?

In general, we test for what is known as Granger causality (Granger Reference Granger1969)—whether the past values of one time series are able to predict the future values of another time series. This approach has its weaknesses (e.g., Freeman Reference Freeman1983) as it is only able to establish mechanical temporal precedence that should be interpreted through theoretical expectations. Temporal precedence does not prove causality, but it constitutes a major and empirically verifiable precondition of it.

We first test for Granger causality in panel data using the method suggested by Dumitrescu and Hurlin (Reference Dumitrescu and Hurlin2012) and implemented in the “plm” (Croissant and Millo Reference Croissant and Millo2008) package in R (R Core Team 2022). The method implements a series of Granger tests per unit (country) and provides a hypothesis test for whether there is evidence for Granger causality in the panel as a whole. This can mean that there is evidence for causality for some of the units, but not necessarily all. We run the test for the aggregated indices using lags from 1 to 5 years (which limits the party systems that we can include in this particular analysis to those that have more than 20 observations—i.e., 36 party systems).

We then test for the causal sequences across years within party systems using a fixed-effects panel model with lagged dependent variables to account for serial correlation and lagged predictors. That is, we test if the variation of PI within party systems is associated with subsequent variation of PSI or vice versa. We use one-, two-, and five-year lags for the predictors in our models to allow for the potential causal effect to play out over various periods of time.

In general, we follow the logic suggested by Beck and Katz (Reference Beck and Katz2011) for the analysis of time-series cross-section data. As the units of analysis are countries that constitute the whole population and as we are not aiming to draw direct conclusions beyond our population, a fixed-effects model is preferable over a random-effects model (Hsiao Reference Hsiao2014, 48-49). Our dependent variables are correlated over time and this introduces serial correlation into the model, which can be addressed by including lagged dependent variables. We are aware that in a fixed-effects panel model this introduces bias into the estimates, but also that this bias is mitigated as the number of over-time observations per unit (democratic periods of party system) is considerable (Beck and Katz Reference Beck and Katz2011, 342). We fit the models using the “plm” package (Croissant and Millo Reference Croissant and Millo2008) in R (R Core Team 2022), using party system as well as year of democracy fixed effects to take into account the fact that levels of PI and PSI might differ for different party systems and that there might be a pattern of variation that depends on the time that a party system has been democratic (e.g., PSI tends to be lower in the earlier years of the system). A fixed-effects model like this analyzes associations with respect to the time or panel unit means, i.e., the variables are effectively centered around the unit means. Variation between countries or regions is thus taken out of the picture. Finally, we use robust standard errors as suggested by Beck and Katz (Reference Beck and Katz1995).

In order to illustrate the full effect of our variables of interest in the panel models with lagged dependent variables, we show the long-term effects of the predictors that take into account how the effect carries on through the lag of the dependent variable (Beck and Katz Reference Beck and Katz2011, 336) as well as the confidence intervals for these long-term effects calculated using the delta method as implemented in the “msm” package (Jackson Reference Jackson2011) in R (R Core Team 2022).

Finally, as robustness checks we also present the results of models that focus on the components of the indices, as well as models containing additional covariates that could help to explain variation in both party and systemic institutionalization. We also run the primary analysis using an alternative approach for serially correlated data—the Prais-Winsten estimator for AR(1) serial correlation (Prais and Winsten Reference Prais and Winsten1954). This method estimates the serial correlation (ρ), transforms the data accordingly, and fits an OLS model to the data.

The overall Granger causality tests in the panel between PSI and PI indicate that there is evidence of causality moving from PI to PSI (Z-tilde = 9.22, p-value < 0.001), rather than from PSI to PI (Z-tilde = 1.36, p-value = 0.17). This suggests that for at least some of the party systems in the data set there is indication for a temporal sequence in the data. Changes in PI help to predict subsequent changes in PSI and not the other way around. Malta, Germany, and Italy illustrate well the causal pattern. In Malta the predictable two-party system could build on the well-entrenched political organizations of Labour and of the Nationalists, with very strong social roots dating back, respectively, to “the Riformisti, who favored British policy, and the anti-Riformisti, who upheld Italian language and culture,” political movements in the late nineteenth century (Carammia and Pace Reference Carammia, Pace, Casal Bértoa and Dumont2022, 113). In Germany, party relations became routinized after the Christian Democrats achieved their final shape through mergers and absorptions. In Italy, the seven main political parties (i.e., Communists, Socialists, Social Democrats, Republicans, Liberals, Christian Democratic, and neo-Fascists) were socially rooted and organizationally stable well before Sartori’s (Reference Sartori1976) polarized pluralist pattern of inter-party competition, with the Christian Democracy making governing coalitions with all major parties except the Communists and the neo-Fascists, consolidated in the early 1960s. Italy illustrates well the typical logic of de-institutionalization too: it was the dissolution of six of those seven political parties in the mid-1990s that led to the collapse of the party system and the appearance of socially less anchored and organizationally weaker political forces.

The data suggest that in only three cases—Andorra, Denmark, and North Macedonia—was the institutionalization of parties preceded by the consolidation of the party system. Out of the three, Andorra is the most extreme case: the current-day protagonists of the party system (the Democrats for Andorra and the Social Democratic Party) were established only at the beginning of the twenty-first century, following a long period of predictable party politics structured around the Liberal Party, a party of “notables” (Minoves Reference Minoves, Casal Bértoa and Dumont2022, 80).

Table 4, showing results of panel data regressions, provides further evidence for PI preceding PSI. Both one-, two-, and five-year lags of PI indicate a positive association to PSI at time t, while for PSI there is no association for the one- and two-year lagged values. We do see a positive association with PSI at t-5 and PI at t, but compared to the effects of PI on PSI, this association is with much more uncertainty.

Table 4

Two-way fixed effects panel models with lagged dependent variables (DV) and lagged predictors (standardized coefficients)

Note: *p<0.05; **p<0.01; ***p<0.001

The corresponding standardized long-term effects that consider how a change carries forward in time through the lagged DV are shown in figure 3. Note that even though the instantaneous effect of PSI (t-5) on PI (t) is smaller than the effects of PSI on PI, the fact that the serial correlation for PI is much stronger means that the long-term effects are comparable. But even so, there is clearly more evidence for potential causality running from PI to PSI than vice versa.

Figure 3

Long-term effects between PI and PSI

In order to test the robustness of these findings, we repeated the analysis substituting our indicators for other suggested measures of party and systemic institutionalization. We used electoral volatility and parliamentary fragmentation as substitutes for closure and the Varieties of Democracy (VDEM) (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Alizada, Altman, Bernhard, Agnes Cornell, Gastaldi, Gjerløw, Glynn, Hicken, Hindle, Ilchenko, Krusell, Lührmann, Maerz, Marquardt, McMann, Mechkova, Medzihorsky, Paxton, Pemstein, Pernes, von Römer, Seim, Sigman, Skaaning, Staton, Sundström, Tzelgov, Wang, Wig, Wilson and Ziblatt2021) party institutionalization index (v2xps_party) as well as two of its components, the party organization index (v2psorgs) and the party linkages index (v2psprlnks), as substitutes for our PI index. The party organizations index captures the extent to which parties have permanent organizations and the party linkages index measures the extent parties have programmatic, as opposed to clientelistic, linkages to constituents.

In our yearly data set, closure has a correlation of -0.3 with parliamentary fragmentation and -0.43 correlation with electoral volatility. The negative correlations reflect that for both of these alternative measures of PSI, higher values should indicate less institutionalization. The VDEM overall party institutionalization index has a correlation of 0.32 with our indicator of PI, the party organizations index has a correlation of 0.26 (0.44 with party age) and the party linkages index has a correlation of 0.22. In other words, the associations are weaker, but they are all positive and significant.

Substituting fragmentation for closure (figure A4 in the online appendix) we see that PI at time t-1 is related to fragmentation at time t in the expected direction, but the coefficient was not significant with the other two time lags. Substituting volatility for closure (figure A5 in the online appendix), we found that PI predicted electoral volatility in all three-time lag specifications. Replacing our PI measure with the VDEM indicators (figures A6–A8 in the online appendix), we received significant results only in one instance: VDEM’s party organizations index at time t-5 helped to predict closure at time t.

To summarize the robustness checks, they largely confirmed the pattern found with our measures, but the results were typically less strong and less clear (refer to the online appendix for more detail).

We ran the same analyses that are reported in table 4 also on the components of the two indices separately and the long-term effects from these models are shown in figures 4 and 5 (refer to tables A1-A5 in the online appendix). These results give us further details about the structure of potential causality between party and systemic institutionalization. We see that both established party domination and the average age of parties are positively related to subsequent PSI on the dimensions of formula and access. This is equally true for 1-, 2-, and 5-year lags for the predictors. Alternation, however, seems to capture something different as alternation at time t is not related to the components of PI across all time lags.

Figure 4

Long-term effects of components of PI on PSI

Figure 5

Long-term effects of components of PSI on PI

If we look in the other direction, as shown in figure 5, we see that the positive association between PSI and PI is driven by established party domination at time t being related to formula and access at time t-5. We see a positive association also for alternation at t-5, but the confidence interval for that association narrowly crosses 0. There is no association between average party age at time t and the components of PSI at times t-1, t-2 and t-5.

Following the existing literature on the institutionalization of party politics (see especially Bolleyer and Ruth Reference Bolleyer and Ruth2018; Lupu Reference Lupu2016; Tavits Reference Tavits2005, Reference Tavits2008; Reference Tavits2013, Mainwaring and Bizzarro Reference Mainwaring, Bizzarro and Mainwaring2018; Casal Bértoa and Enyedi Reference Casal Bértoa and Enyedi2021), we modeled our main variables (PI & PSI and their components) and the potential causal associations between them also through models that included various institutional, economic, and political variables traditionally used to represent different aspects of the environment of party politics as contemporaneous predictors, namely:

1. a) the electoral system, whose impact is operationalized through Gallagher’s (Reference Gallagher1991) the least-squares index of disproportionality;

2. b) economic development, measured using the logarithm of GDP per capita of the election years (Gapminder 2021);

3. c) fragmentation, calculated through the Laakso and Taagepera (Reference Laakso and Taagepera1979) effective number of parliamentary parties index; and

4. d) polarization, captured by the percentage of votes for anti-political-establishment partiesFootnote ⁹ (cf. Karvonen and Quenter Reference Karvonen, Quenter, Berg-Schlosser and Mitchell2002; Powell Reference Powell1981).

The detailed results of these analyses are reported in tables A6 and A7 in the online appendix. In none of the models did the results reported here change substantially (only the uncertainty related to the estimates increased, as expected).

Finally, we also tested for the potential causal sequence using the Prais-Winsten estimator, which is reported in table 5. The results for this model echo what was shown earlier—that past increases in PI help to predict future increases in PSI and not vice versa. One major notable difference here is that the model shows a negative association between PSI at t-2 and PI at time t. But this only strengthens the overall conclusion that increased PI is more likely to eventually lead to increased PSI and not the other way around.

Table 5

Results from Prais-Winsten AR(1) model (standardized coefficients, panel corrected standard errors)

Note: *p<0.05; **p<0.01; ***p<0.001

Standardized coefficients, panel corrected standard errors