Elite Investments in Party Institutionalization in New Democracies: A Two-Dimensional Approach

This article conceptualizes party institutionalization and theorizes the conditions under which party elites invest in institutionalized parties in new democracies. We specify routinization and value infusion as two central dimensions of party institutionalization and theorize conditions relevant for party institutionalization across three central spheres: the party system, the state, and society. Constructing measures for routinization and value infusion based on expert survey data, we test our framework through multivariate regression models across parties in 18 Latin American democracies. As theoretically expected, some conditions (access to executive office, a party’s formative environment, and group ties) significantly relate to both dimensions, while others (party system polarization and fragmentation, permanent state subsidies, and legislative office) relate to one dimension only. This highlights the multidimensionality of party institutionalization as a phenomenon and the complexity of the empirical conditions associated with it.

I n a widely cited review article on political parties in Latin America, Levitsky urged us more than 10 years ago to develop more nuanced conceptual frameworks that differentiate various dimensions of parties as organizations, conceding that such efforts are seriously constrained by a lack of data (2001a, 106-7). Reflecting this difficulty, most of the literature to date has focused either on in-depth qualitative studies of a single or a few cases at the party level or, alternatively, on the cross-national study of party system institutionalization at the country level. In contrast, this article conceptualizes and empirically explores the diversity of party institutionalization, an important party property in both new and established democracies. According to Levitsky (1998) as well as Randall and Svåsand (2002), party institutionalization-a party's development toward consolidation-becomes manifest in two distinct internal properties: routinization, a structural dimension that refers to rule-guided processes between a party and its fol-lowers, and value infusion, an attitudinal dimension that refers to followers' emotional affiliation to their party. While party institutionalization "is not identical with the party's development in purely organisational terms" (Randall and Svåsand 2002, 12), structures and practices party elites decide to invest in can be either conducive or detrimental to institutionalized relationships between followers and their party (Panebianco 1988, 53-65). This is important as routinization and value infusion are notoriously difficult to capture directly. 1 To measure the mechanisms and practices party elites choose to build their own party allows us to capture empirically whether investments in each dimension of party institutionalization have been made or not. Using data from the Democratic Accountability and Linkages Project (DALP; Kitschelt et al. 2009) to construct such measures provides us with nuanced proxies for each dimension. This, in turn, allows us to examine which conditions are conducive to routinization and value infusion and which are not.
Party institutionalization in new democracies constitutes a puzzle as parties are often born "in the state" rather than formed on the basis of strong societal ties as in old democracies (Levitsky, Loxton, and Van Dyck 2016;van Biezen 2005). They operate in contexts characterized by high levels of political uncertainty, which is bound to alter fundamentally the considerations of elites regarding the type of party base to build and might lead to a prioritization of flexibility, in turn conducive to low institutionalization levels Riedl 2013, 1349). It has been argued that in new democracies, creating a permanent anchoring in society is time-consuming, laborintensive, and chosen only if no other option is available to build up support (van Biezen 2005, 155). Yet if this is so, why do we find considerable diversity among parties in, for instance, Latin American democracies, with some parties being highly institutionalized, such as the Workers' Party (PT) in Brazil or the Institutional Revolutionary Party (PRI) in Mexico, while others have remained weakly institutionalized in the long term such as the Ecuadorian Roldosista Party (PRE) and the Radical and Social Democratic Party (PRSD) in Chile? Generally speaking, why should party elites consider costly investments in a routinized and value-infused party base worthwhile in highly uncertain environments that incentivize electoral, "catchall" strategies and flexible, weakly institutionalized parties (Gunther and Hopkin 2009;Kirchheimer 1966;Lupu and Riedl 2013;Tavits 2013)?
Studying party institutionalization is particularly important in new democracies since stable parties can be crucial for the viability and functioning of democracy (see, among others, Svåsand [2013] and Tavits [2013]). Institutionalized parties tend to be more firmly anchored in society, helping parties to overcome collective action problems, allowing for the effective channeling of preferences from citizens to party elites embedded in democratic institutions. Party institutionalization can thereby stabilize patterns of party competition and is thus relevant to party system institutionalization, a central element in assuring politicians' electoral accountability to citizens. 2 While we focus on the party level, our study nonetheless contributes to the refined measurement of party system institutionalization as a multidimensional concept. To measure institutionalization on the party level is relevant to two of the four conceptual dimensions of party system institutionalization (Luna 2014, 406-7): the rootedness of parties in society suggesting stable ties between parties and followers (echoing the concept of value infusion) and the extent to which party infrastructures are well developed and stabilize relationships with followers (echoing the concept of routinization).
Especially in the Latin American context, relatively little is known about the conditions under which party elites invest in institutionalization, which is problematic since the region's much studied "crisis of representation" has been closely linked to the decline of traditional political parties as well as the decay of entire party systems (e.g., Sanchez 2008). To be able to examine the various consequences of different patterns of party institutionalization, the nature of parties as complex organizations (Wills-Otero 2016, 759) needs to be conceptualized and mapped out first, which is one major aim of this article.
After introducing our two-dimensional conceptualization of party institutionalization in the following section, we theorize conditions-related to party system characteristics, the nature of party-state, and party-society relationsthat incentivize party elites in new democracies to invest in either routinization, value infusion, or both. While our hypotheses theorize conditions relevant to party building in new democracies generally, we apply them to parties across 18 Latin American democracies. Estimating multivariate regression models, we show that routinization and value infusion do not always go together, nor are they necessarily more likely to be present under the same conditions. We conclude with a discussion of the broader implications of our findings for the study of political parties in new democracies more generally.

CONCEPTUALIZING PARTY INSTITUTIONALIZATION
Party institutionalization (hereafter PI) captures a party's development toward consolidation. Focusing on the conditions under which elites incentivize the formation of an institutionalized party base, we start out from Panebianco's seminal book (1988). Building on his contribution, conceptual work on parties' internal life has distinguished two dimensions of institutionalization: routinization and value infusion. The routinization of a party's relationship to its followers takes place when processes within it become more rule-guided and regularized. This becomes visible in an increasingly elaborate and stable infrastructure (Panebianco 1988, 49, 53), conducive to organizational rules becoming "perceived as permanent structures" (Levitsky 1998, 81) and to parties as structures being increasingly dense, regularized, and thus able to guide followers' behavior (Janda 1980). 3 Importantly, the presence of permanent structures as a factor 2. Party system institutionalization is defined by four dimensions: the stability of the rules of competition, the legitimacy of the party system, the stability of parties' roots in society, and the strength of party structures. Roots in society have so far been operationalized through indicators as diverse as party age or ideological party-voter linkage; the strength of party organization usually is left aside. See Luna (2014), Mainwaring and Scully (1995), Mainwaring and Torcal (2006), and Sanchez (2008). See Casal Bértoa (2016) on the distinction between party institutionalization and party system institutionalization.
conducive to base-level routinization is not equivalent to the creation of formal party branches in the traditional (western European) sense. It equally can be assured by networks of local intermediaries (Freidenberg and Levitsky 2006;Kitschelt and Kselman 2010). In this sense, permanence does not presuppose one particular "organizational form" but refers to structures (formal or informal) that guide the behavior of followers who interact within them and thereby create continuity between elections. These structures support base-level routinization because formal party branches as well as informal networks incorporate followers into the party (Freidenberg and Levitsky 2006;Kitschelt and Kselman 2010, 13-14). They provide channels for communication between party and grassroots (Levitsky 2001b, 54-56;Tavits 2013), with party officials forming part of these structures who "establish routines and standard operating procedures" (Kitschelt 1994, 222), thereby familiarizing followers with rules and procedures that govern the internal life of a party.
While also characterizing the relationship between party and followers, value infusion is not a structural feature. It shows when followers start caring about the survival of their party as such, rather than seeing it as a mere instrument to achieve a set of goals (Janda 1980, 19;Levitsky 1998, 82). Wilson (1973) as well as Panebianco (1988) have stressed that most organizations will try to combine the provision of different incentives to followers to stabilize voluntary support, which parties continuously depend on, since followers are free to leave. That said, Panebianco associates value infusion especially with nonmaterial collective incentives such as party identification linked to a particular cause or to social and political goals shared by followers (10-11). Similarly, the provision of such incentives can be generated by leaders embodying core party values (Pedahzur and Brichta 2002, 40;Rosenblatt 2013). 4 While selective incentives are important to motivate ambitious elites, nonmaterial incentives provided to followers generate noninstru-mental attachments and thus a diffuse loyalty to the organization independent of the ongoing provision of material benefits or career advancements (Panebianco 1988, 10-11;Randall and Svåsand 2002, 10).
To sum up, routinization and value infusion capture qualitatively distinct phenomena, reflecting Randall and Svåsand's (2002) insightful distinction between structural and attitudinal institutionalization that do not necessarily coincide (see fig. 1).
Reviewing recent studies, purely structural institutionalization is indeed widespread in very young democracies (Svåsand 2013, 265), while new parties formed by individual entrepreneurs in old democracies often routinize internal processes with little interest in value infusion (Bolleyer 2013, 215-17). Similarly, not only Latin American parties but also parties in central eastern Europe have created informal branch structures, without cultivating a committed membership (Freidenberg and Levitsky 2006, 179;van Biezen 2005, 155-56).

THEORIZING ELITE INVESTMENTS IN PARTY INSTITUTIONALIZATION
What are the conditions under which party elites are likely to invest in routinization or value infusion? We distinguish three sets of factors that create incentives toward building an institutionalized relationship with followers, each derived from the party's position in a core arena of party activityparty system, state, and society. All factors theorized below 3. Note that we distinguish base-level routinization (through which followers are socialized into party rules) from constraints on leaders' autonomy through formal mechanisms for rank and file to hold leaders accountable (see Wills-Otero 2009, 132-33). Routinization as defined here does not require vertical accountability mechanisms between followers and leaders. Base-level routinization is likely to shape the behavior of leaders, if leaders themselves are recruited from the party base. The two aspects are assumed to go together in the classical mass party model (e.g., Panebianco 1988). Yet various studies have indicated that outsider recruitment as well as high leadership autonomy or fluidity can coexist with a routinized party base (Kitschelt 1994;Levitsky 1998;Wills-Otero 2009). 4. While Panebianco has stressed the incompatibility between charismatic leadership and institutionalization, he presents strong institutionalization as the second most likely outcome of a charismatic leader's departure, with weak institutionalization being the least likely outcome (1988,. Thus, if a "charismatic party" can outlive its leader, attachments to this past leader can strengthen followers' loyalty and value infusion.  Randall and Svåsand (2002). feature in debates on party development beyond one particular region, assuring our framework's broader applicability.
The nature of party systems and party institutionalization Polarization, the ideological distances between parties, and fragmentation, the number of parties that compete, are two constitutive features of party systems (e.g., Sartori 1976). As each of these two features has repercussions for the nature of competition between parties, we can expect them to influence elite choices whether to invest in PI or not.
Starting with polarization, the broadening of a party's programmatic profile-or in Lupu's (2013) words the dilution of its brand-can facilitate the mobilization of a diverse range of followers. From this perspective, parties' attempts to create stable and permanent connections to clearly demarcated groups through PI seem counterproductive (Gunther and Hopkin 2009, 214;Riedl 2013, 1349). However, considering a polarized party system in which ideological differences between parties are pronounced, citizens are more likely to recognize those differences and, in turn, to form stable party attachments (see LeBas 2011; Lupu 2015, 334-35). In such a context, attempts on behalf of elites to generate stable attachments by appealing to long-term partisan loyalties (value infusion) are less costly and more likely to pay off as compared to systems where ideological differences between parties are smaller and less visible (and "catchall strategies" are thus more beneficial). Consequently, elites' incentives to foster followers' durable emotional affiliation should grow with the polarization of the party system in which they are embedded.
H1 (Polarization Hypothesis). The more polarized a party system is, the more likely party elites invest in value infusion. Moving to party system fragmentation, as the number of political parties in a party system rises, it becomes increasingly costly for voters to gather sufficient information to assess the record of incumbents and the promises of potential challengers (e.g., Blais and Dobrzynska 1998;Schleiter and Voznaya 2014). Under such conditions, parties face higher pressure to try to cultivate distinctive party labels able to underpin stable commitments of clearly defined groups of followers (Hanson 2010;LeBas 2011;Lupu 2015), which, in turn, supports value infusion. More importantly, as presenting distinct policy packages or identities becomes, ceteris paribus, more difficult in increasingly fragmented party systems, invest-ments in an infrastructure stabilizing ties to followers (Kitschelt and Kselman 2010, 13-14), that is, routinization, becomes increasingly valuable.
H2 (Fragmentation Hypothesis). The more fragmented a party system is, the more likely party elites invest in routinization and value infusion.

Party-state relations and patterns of party institutionalization
Problems of resource scarcity are bound to complicate party building in new democracies, where parties-after transition-were suddenly confronted with the pressure to run democratic elections and win over volatile electorates. Consequently, direct and indirect access to state resources should strengthen parties' capacity to build institutionalized parties (see Casas-Zamora 2005; Grzymala-Busse 2002; Molenaar 2012).
Starting with direct state funding, permanent state subsidies received directly by the party organization (rather than electoral subsidies received by individual candidates) should increase the capacity of elites to build a routinized infrastructure including party officials who "establish routines and standard operating procedures" (Kitschelt 1994, 222), socializing followers into party rules. We do not expect a link between permanent subsidies and value infusion as disposing of money or other material resources itself is not enough for parties to incentivize citizens' emotional affiliation to them (Levitsky et al. 2016). The same rationale applies to the link between institutionalization and legislative office providing access to indirect state funding. Parliamentary representatives receive formal support, for example, through access to professional and technical staff (Morgenstern, Negri, and Pérez-Liñán 2008, 174). Although usually reserved for "parliamentary usage," parties still benefit from those resources (Bolleyer and Gauja 2015;Carey 2003, 31-33). As they multiply with a party's size in the legislature, investments in routinization should become easier the stronger the party's position in the legislature.
H3 (Permanent State Subsidies Hypothesis). If a party has access to permanent state subsidies, party elites are more likely to invest in routinization.
H4 (Legislative Office Hypothesis). The bigger a legislative party is, the more likely party elites invest in routinization.
Finally, taking over the core executive office in a regime should increase a party's capacity to build a routinized party. Government patronage, for instance, can be used by parties to establish links to recipients of resources, helping to develop routinized exchanges between these parties and their followers (Kitschelt 1994;Kopecký, Mair, and Spirova 2012;Stokes 2007). 5 We further expect a link between executive office and value infusion. While permanent party subsidies and parliamentary seats tend to be accessed by the large majority of parties including minor ones (Casas-Zamora 2005), parties holding the core executive office such as the presidency tend to be major players in their party systems. Such players need to gather and sustain support across a wide range of relatively diverse constituencies. This positively incentivizes the cultivation of long-term loyalties through the provision of nonmaterial collective incentives rather than the sole reliance on selective incentives such as patronage that are most effective when targeting specific groups (Kitschelt and Wilkinson 2007;Stokes 2007). In conjunction, occupants of core executive offices enjoy superior visibility, which puts them in a particularly suitable position to present themselves as a major identification figure to followers (Samuels 2002;Weisberg 2002, 342), an advantage supportive of value infusion.
H5 (Executive Office Hypothesis). The longer a party holds the core executive office in a regime, the more likely party elites invest in routinization and in value infusion.
While up to now we have theorized party-state relations considering the opportunity structures of parties' current regime, parties often emerged in settings different from the one they currently operate in. The nature of this formative environment is likely to leave its "genetic imprint" on parties' characteristics in the longer term (Panebianco 1988, 49-50;Randall and Svåsand 2002, 17-18;Riedl 2016;Slater and Wong 2013). Parties formed in authoritarian, thus less competitive, settings find it easier to build stable attachments and establish a distinctive brand than parties formed in fully democratized settings facing stiff competition from not only numerous new but also already established parties (Loxton 2015, 160-61;Randall and Svåsand 2002, 18). This suggests that the former find it easier to generate value infusion than parties formed after the transition. Furthermore, parties formed in authoritarian settings that managed to survive transition have been ascribed a competitive advantage thanks to "inherited" resources, such as territorial organizations or informal local networks, that facilitate routinization after transition (Frantz and Geddes 2016;Loxton 2015, 161-62). More specifically, formerly ruling parties and those antiregime parties that used violence under authoritarianism face enhanced pressures to legitimize themselves in newly democratized regimes through forming close ties to society (de Zeeuw 2010;Grzymala-Busse 2002, 69-70;Holland 2016;Tavits 2013). Investments in routinized relationships to followers are one strategy to do so.
H6 (Formative Environment Hypothesis). A party formed in an authoritarian context is more likely to invest in routinization and value infusion.

Party-society relations and patterns of party institutionalization
Despite the importance of parties' relationships with the state, the support of societal groups can be an important foundation for still fluid party organizations to consolidate. Compared to old democracies, in Asia, Latin America, Africa, or central eastern Europe, ties to religious or ethnic groups have had importance similar to class-based ties to unions (Mainwaring and Scully 2003; Randall and Svåsand 2002;Van Cott 2007). Ties to these three types of groups provide a particularly fruitful foundation for PI because these groups often tend to provide widespread local infrastructures and a pool of followers used to operate in organizational, rule-based settings, both supportive of routinization. Simultaneously, these groups pursue a strategy of social incorporation and cultivate strong, encompassing group identities-supportive of value infusion-from which individuals can exit only a very high cost (e.g., LeBas 2011).
H7 (Group Ties Hypothesis). A party with strong ties to a union, ethnic, or religious group is more likely to invest in routinization and value infusion.

DATA, MEASUREMENT, AND MODEL CHOICE
The following analysis draws on a cross-sectional data set that combines expert ratings on political parties' organizational characteristics and linkage strategies in 2009 with electoral statistics and data on chief executives from 1978 to 2008 in 18 Latin American democracies. We consider our approach applicable to democratic regimes that have been stable for a certain period of time and repeatedly experienced competitive elections, as in those regimes parties are exposed to the com-5. Studies show that particularistic exchange relationships between parties and followers tend to be based on rule-guided and regularized behavior. The distributive efficiency of patronage-based parties depends on their capacity to gather information about the responsiveness of potential clients to selective benefits and their commitment to support the party even in the absence of formal sanctions (Stokes 2007, 610-15). To overcome these problems parties invest in dense networks of local intermediaries (brokers) who identify and monitor potential clients over time and build long-term reciprocity-based relationships with them (see, e.g., Schaffer and Schedler 2007). petitive pressure of democratic elections, an assumption that underlies several of the hypotheses in our framework. The Latin American regimes covered in our empirical application are suitable cases because they meet this condition: all countries experienced at least four democratic elections since transition and each has an average (polity2) democracy score of 5 or higher during the 10 years prior to this study (see Mainwaring and Pérez-Liñán 2013;Marshall, Jaggers, and Gurr 2013). Moreover, the Latin American region shares a similar history of economic development compared to other world regions (see, e.g., Bértoa and Ocampo 2012). This relative unit homogeneity is advantageous since it allows us to concentrate on the explanatory variables in our framework, while providing crucial variation in the latter to test our hypotheses. Our analysis includes all legislative parties covered in the DALP data set (Kitschelt et al. 2009), which-considering our purposes-has various advantages over alternative data sources, including a (comparatively) broad coverage of parties, leading to a sample of 88 political parties across 18 countries. 6

Operationalization of party institutionalization
The DALP survey allows us to capture directly whether party elites invest in structures and practices that are closely associated with routinization and value infusion, respectively, which provides us with suitable proxies to measure each dimension empirically. 7 In line with the conceptualization of base-level routinization as structural institutionalization detailed earlier, we combine two items in the DALP survey as a proxy for its presence: the establishment of permanent local party offices (item A1) and the existence of parties' local intermediaries (item A3), which avoids a bias in favor of a formal organization (Kitschelt and Kselman 2010;Levitsky 1998). The routinization index is constructed as follows: first, each indicator is normalized between zero and one, with higher values indicating higher levels of routinization. Second, we calculate mean expert ratings per party for each indicator. Third, we use an additive aggregation rule (i.e., the mean of both proxies), assuming each indicator measures one aspect of the dimension captured.
We measure value infusion through the appeal of party elites to long-term partisan loyalty invoking the party's historical origins or achievements of historical leaders, or ref-erences to party symbols and rituals (item E4). This indicator directly taps into Panebianco's argument that party identification and solidarity between followers (reinforced by shared symbols and rituals) are primary sources of collective incentives central to generate value infusion (1988, 10-11; see also Wilson 1973). It also reflects the importance of past leaders as identification figures who (if a party could outlive them) can serve as important sources of followers' long-term attachment (Rosenblatt 2013). We normalize the item to range from zero to one, with higher values indicating higher levels of value infusion, and calculate the mean expert ratings per party. Figure 2 provides a graphical overview.
Considering the face validity of our measures, among the parties that are strongly institutionalized on both dimensions are the Brazilian Workers Party (PT), the Paraguayan Colorado Party (ANR), and the Institutional Revolutionary Party (PRI) in Mexico, echoing earlier qualitative studies. Interestingly, the Peronist Party (PJ) in Argentina- Levitsky's (1998) case of attitudinal PI-ranks highly on routinization. This goes back to, on the one hand, the inclusion of informal structures into our measurement and, on the other, our focus on baselevel routinization. 8 Levitsky considered the PJ as weak on "formal routinization" with outsider recruitments being common and leaders enjoying wide discretion (1998, 2001b). At the same time, he characterized base-level Peronist activity as informally routinized "to an important degree," with activity in local units being "rooted in widely shared norms" and Peronist practices "widely known and remarkably similar across territorial units" (1998, 87; 2001b). This not only substantiates "our" location of the PJ with respect to base-level routinization but also underlines the appropriateness of our decision to include both formal party structures and informal networks into our measure. It also explains why the category of (relatively) strongly institutionalized parties includes state-centered party machines such as the Colorado Party (ANR), labor-based patronage parties such as the Institutional Revolutionary Party (PRI), and programmatic mass parties such as the Brazilian Workers Party (PT). Our operationalization avoids dismissing any party distinct from the classical "mass party model" as weakly institutionalized.
Returning to the broader picture, most parties are strong or weak on both routinization and value infusion, indicated by a correlation coefficient of .74 (p ! .000) between the two dimensions. Nevertheless, we find some routinized parties that do not foster value infusion in the same way as they foster routinization (i.e., structurally institutionalized parties) such as 6. We provide additional information on the data set and its advantages, question wording, as well as a list of countries and parties in the supplementary material. For more details on the survey, see https://web .duke.edu/democracy. 7. See figs. A1 and A2 (available online) for more details on our measures of the dependent variables.
8. Note that at the time of the survey two factions of the Peronist Party competed against each other, the Federal Peronists (PJ in fig. 2) and the Front for Victory (FPV in fig. 2), of which the FPV is less institutionalized.

Operationalization of the explanatory variables
Since our analyses are based on cross-sectional data, we can examine whether and how the conditions theorized in our hypotheses are related to different patterns of PI but we cannot test the causal relationships between them. We partially account for these data limits through the time structure of our data. Reflecting our theoretical rationale, we lag the measures of our independent variables by using data from before 2009 (the year of the DALP survey) as far as possible. 9 Descriptive statistics are provided in the supplementary material. We measure party system polarization (hypothesis 1) using the Taylor-Herman index (Taylor and Herman 1971) calculated on the basis of item D6 in the DALP data, which captures political parties' left-right placements, and parties' seat shares in the national legislature. Party system fragmentation (hypothesis 2) is measured on the basis of seat shares capturing the effective number of parliamentary parties (Laakso and Taagepera 1979). We calculate the index for each election between 1999 and 2008 and then take the mean for each country, to control for sudden changes in party systems. We measure permanent state subsidies (hypothesis 3) capturing the number of years a political party has received permanent organizational funding prior to the survey based on data provided by Molenaar (2012, 13, 28). We use the logged variable to account for decreasing returns for each year after funding was introduced. To test hypothesis 4 (legislative office) we use parties' parliamentary strength based on the seat share parties held prior to the survey. 10 We measure the time period during which a party held executive office (hypothesis 5) by coding the number of years the party held the presidency since 1980 or the latest return to democratic rule. 11 We again use the logged variable to account for decreasing returns for each year the party held the presidency. Formation environment (hypothesis 6) captures the regime type in the first five 9. The party system polarization and the group ties measure rely on the DALP data set. Thus these variables could not be lagged.
10. Data for the distribution of seats are provided online by most countries; detailed resources are available on request from the authors.
11. Data on presidents and their parties stem from Ruth (2016).
years of a party's life span using the mean score for that period of the polity2 item in the Polity IV data set. The item captures the regime type of a country ranging from 210 (strongly autocratic) to 110 (strongly democratic; Marshall et al. 2013). We inverse and normalize the variable to run from zero (strongly democratic) to one (strongly autocratic) to mirror the direction of our hypothesis. Finally, we capture a party's group ties (hypothesis 7) based on DALP item A8, which asks experts to indicate the most relevant civil society organization a political party is strongly linked to. We use the percentage of experts positively identifying either unions, ethnic, or religious groups as the most relevant civil society organization the party is linked to. The variable, hence, ranges from zero (no group ties) to one (strong group ties). Finally, we include logged party age as a control variable covering the years from the party's foundation up to 2008.

Model choice
To analyze the relationship between our explanatory variables and the two dimensions of PI, we estimate multivariate ordinary least squares (OLS) regression models with standard errors clustered by country, to account for the nestedness of political parties within countries and the heteroscedasticity in the error term (see Gelman and Hill 2007). 12 All coefficients have the theoretically expected signs. Three variables-executive office, formation environment, and group ties-are significantly associated with both dependent variables. However, three-party system fragmentation, permanent state subsidies, and legislative office-significantly relate only to routinization. Party system polarization relates only to value infusion. All effects reported below are significant at the 95% confidence level or higher and hold controlling for party age. In the following sections, we discuss these findings grouped along the three sets of factors theorized as relevant to party building. 13

Elite investments in institutionalized political parties
Considering conditions linked to a party's embeddedness in its party system, polarization (hypothesis 1) is positively related only to value infusion, while a party's operation in a fragmented party system (hypothesis 2) is associated with routinization (see fig. 3). In increasingly polarized systems, ideological differences between parties become more pronounced, and citizens are more likely to recognize interparty differences and, in turn, are more likely to form stable party attachments (Lupu 2015, 334-35). Parties are therefore incentivized not only to invest in the distinctiveness of the offer within their party system but also to increase the potential identification of their voters with the party's ideological position or the group identity they address, thereby enhancing value infusion (Bartolini 2000;Lupu 2013). The coefficient is highly significant at the 99.9% confidence level. Figure 3A shows that value infusion is more than 0.10 points lower for parties in a party system with very low polarization (e.g., Honduras) compared to parties in a party system with high polarization (e.g., El Salvador).
Party system fragmentation (hypothesis 2), in contrast, is positively related to party routinization (p ! .001). One rationale for the differentiated impact of fragmentation is provided by Epstein (2009), who argues-referring to the Brazilian multiparty system-that high fragmentation complicates the distinguishability of programmatic party platforms, especially if the party system is also prone to clientelistic behavior, as many Latin American party systems are. In such contexts, the building of long-term loyalties is very demanding, which might incentivize elites to invest in party routinization instead. As shown in figure 3B, routinization increases from 0.54 in party systems with an effective number of parties of two to 0.75 in a party system with eight "effective parties." Moving to party-state relations, as theoretically expected, permanent party subsidies (hypothesis 3), access to legislative office (hypothesis 4), and executive office (hypothesis 5) are significantly related to routinization. 14 The longer political parties benefit from permanent state subsidies (hypothesis 3), the more they tend to be routinized. Similarly, the access of a 12. We estimate linear hierarchical models as an alternative. The basic findings stay the same with only slight changes in the level of significance of some variables (see table A4 in the supplementary material). 13. To check the robustness of our findings we ran additional model specifications with several country-level control variables; none of these variables had significant effects on our two dependent variables (for details see table A8 in the supplementary material).
14. We ran three additional model specifications to examine the robustness of our arguments regarding access to state resources (hypotheses 3-5). First, we added a control variable capturing the overall dependency of parties on state resources. Second, we added a variable controlling for participation in government coalitions. Finally, we controlled for single-party dominance (Greene 2007). All results indicate that our results are very reliable. None of our main findings lose significance, and their effects remain remarkably robust in both size and direction. Details are provided in table A5 in the supplementary material. political party to legislative office has a positive and comparatively large effect on routinization (see the standardized coefficients reported in table 1). Routinization increases from 0.55 for a party with a legislative seat share of 10% to 0.78 for a party with 40% of the legislative seats (see fig. 4A). While both the availability of permanent subsidies and legislative office facilitate routinization, they do not support the generation of noninstrumental attachments of followers; that is, we find no significant relationship with value infusion.
In contrast, parties with access to the presidency as the core executive office are characterized by higher levels of both routinization and value infusion, in line with our executive office hypothesis. The presidency provides parties with a comparative resource advantage, facilitating the capital-intensive formation of permanent party structures supporting routinization. One unit increase in executive office access (logged years) leads to a 0.05 point (p ! .001) increase in party routinization. Regarding value infusion, presidential parties-as major players-face incentives to maintain partisan loyalties across a broad range of supporters through the provision of nonmaterial collective incentives conducive to value infusion, rather than to rely solely on selective incentives such as patronage that are most effective when targeting specific groups (Kitschelt and Wilkinson 2007;Stokes 2007). Simultaneously, because of the superior visibility of the presidency as the core executive office, these parties can effectively increase the value of their party label for rank-and-file members (Hicken and Stoll 2011;Samuels 2002;Weisberg 2002). One unit increase in executive office access (logged years) leads to a 0.07 point (p ! .05) increase in value infusion (see fig. 4B).
On the basis of our cross-sectional analyses, it is difficult to rule out reverse causation. While in new democracies access to public office often helps to overcome problems of resource scarcity in the process of party building (Levitsky et al. 2016), PI might also help a party to access presidential office and lead to a stronger representation in the legislature. To examine whether success in executive and legislative elections is affected by either one or both of our two dimensions of PI, we performed additional robustness checks by regressing a party's degree of routinization and value infusion (at t 1 ) on its success in the next legislative and the next presidential election after 2008 (at t 2 ), respectively. The results suggest that reverse causation does not drive the relationship between PI and winning executive or legislative office. We do not find any significant association between routinization (at t 1 ) or value infusion (at t 1 ) and the share of seats a political party is able to win in the next legislative election (at t 2 ) or the probability of winning the next presidential election (at t 2 ). 15 These findings echo earlier work that highlights that access to presidential office does not require presidential candidates to rely on a strong party organization, as the frequent election of political outsiders (Levitsky and Loxton 2013) and the dominance of highly personalized and media-based electoral campaigning illustrate (Mainwaring and Zoco 2007).
Having considered the implications of opportunity structures available to parties in their current regime, we now move to the implications of their formation environment (hypothesis 6), which is positively associated with routinization and value infusion alike. The link to routinization echoes studies stressing the importance of "inherited" resources in terms of territorial organization or informal local networks from which formerly ruling successors as well as their repressed counterparts profit (de Zeeuw 2010;Frantz and Geddes 2016;Loxton 2015, 161-62). Routinization increases from 0.57 for parties formed under a strongly democratic regime to 0.72 for parties formed under a strongly autocratic regime. Moving to value infusion, our results indicate that value infusion increases from 0.57 for parties formed under a strongly democratic regime to 0.75 for parties formed under a strongly autocratic regime. As authoritarian settings are less competitive, parties find it easier to build stable attachments and establish a distinctive brand than parties formed in fully democratized settings facing stiff competition (Loxton 2015, 160-61;Randall and Svåsand 2002, 18).
Concluding with the implications of a party's societal relations, ties to unions, ethnic, or religious groups (group ties, hypothesis 7) are conducive to both dimensions of PI. Parties with strong group ties are, on average, 0.15 points more routinized compared to parties without such ties (see fig. 5A). To uphold these alliances and secure a long-term electoral support base, party elites face strong incentives to invest in permanent party structures to coordinate their interaction with these groups, which simultaneously bring resources with them 15. Detailed information on these endogeneity checks can be found in tables A6 and A7 in the supplementary material.  (Burgess 1999, 106-11), both of which support routinization. Also, parties with strong ties to unions, ethnic, or religious groups are more likely to invest in value infusion than those without. For the former, value infusion is, on average, 0.15 points higher (see fig. 5B). As Collier and Collier (1991) showed, several political parties in Latin America established "stable, institutionalized alliances" with labor unions. A similar stabilizing role was played by ethnic movements (Madrid 2005;Van Cott 2007) and religious associations (see Mainwaring and Scully 2003). Simultaneously, these groups cultivate strong and encompassing group identities supportive to value infusion from which individuals can exit only at very high costs (e.g., LeBas 2011).
To sum up, our findings highlight that distinguishing between routinization and value infusion is not only conceptually persuasive but empirically important to account for patterns of party development: all factors theorized in our framework played a role, but the majority (four of seven) affected only either routinization or value infusion. Two of the three factors having significant effects on both dimensions fall in the sphere of party-state relations, and the third falls in the sphere of party-society relations. In contrast, the two party system characteristics each associate strongly with only one of the two dimensions: polarizationis associatedwith valueinfusion only, while fragmentation is associated only with routinization. Considering which sphere-party system, state, and society-is particularly relevant for each dimension of PI, routinization is related particularly strongly to the nature of party-state relations: four of six variables that are significant fall in that analytical category; that is, all four of our hypotheses on the link between party-state relations and routinization hold. The picture regarding value infusion is more complex: while again two significant conditions are associated with parties' current or past relationship to the state, the remaining two are linked to parties' relations with society (group ties) and their party system (polarization).

CONCLUSION
Building institutionalized parties is a demanding process, especially in new democracies (Casal Bértoa 2016;Levitsky 1998;Randall and Svåsand 2002;van Biezen 2005). To account for the diversity of PI in new democracies, we theorized conditions incentivizing party elites to invest in institutionalized relationships to followers in contexts that are commonly considered unfavorable to extra-parliamentary party building (e.g., Levitsky et al. 2016;Svåsand 2013;van Biezen 2003). Drawing on literatures on party development from various regions, we identified such conditions across three central spheres of party activity: irrespective of the type of democracy they operate in, parties compete with each other in the electoral market, they access (if they can) state institutions, and (to different degrees and in different ways) they are linked to society. Substantiating our decision against a more parsimonious approach, we found that in new democracies-represented by those in the Latin American region-routinization and value infusion were significantly associated with conditions located in all three spheres. Simultaneously, our findings revealed variegated relationships between explanatory factors located in each sphere and each of the two dimensions of PI. Consequently, the conceptual distinction between qualitatively different dimensions of PI (Levitsky 1998;Randall and Svåsand 2002) is indeed empirically relevant.
Exploring the conditions that incentivize party elites in new democracies to build institutionalized parties is important. The latter provide structures allowing citizens to be involved in politics between elections and thereby form longterm attachments rather than solely engaging in politics for instrumental reasons (e.g., to receive material benefits through clientelistic party networks). Value infusion can help counter antiparty sentiments and alienation from politics that are widespread among citizens in Latin American democracies (see Payne 2006), strengthening the legitimacy of a democratic regime. Simultaneously, routinized relationships with followers provide channels of communication between citizens and elites, a core element to assure the electoral accountability of representatives (e.g., Freidenberg and Levitsky 2006;Randall and Svåsand 2002). By systematically assessing under which conditions parties institutionalize structurally or attitudinally, we thus not only address an important theoretical and empirical gap as large-N studies so far have not distinguished these two dimensions. We also contribute to important debates on the quality of democracy and the contribution the nature of individual parties is likely to make to the consolidation of new democracies (Diamond and Gunther 2001;Diamond and Morlino 2005).
A central challenge for future research will be the assessment of the actual effects of different dimensions of PI, which we disentangled conceptually and empirically, on the democratic process. PI is often considered as beneficial for democracy (Huntington 1968;Mainwaring and Scully 1995), yet it remains unclear whether (too) strongly institutionalized parties might not have negative effects as well (e.g., Casal Bértoa 2016;Hicken and Martinez Kuhonta 2011). Similarly, if routinization and value infusion vary in strength within the same party, we face the challenge to theorize separately the effects of each dimension on party system institutionalization and, more broadly, on democracy. Research has stressed that a permanent local organization can be an important tool for clientelistic parties to allocate selective incentives (Levitsky 2001a;Stokes 2007). And, as indicated by our mapping of institutionalization patterns ( fig. 2), some parties routinize without cultivating an emotional attachment of followers to the party and, with it, possibly to the regime that the party represents. Functional linkages underpinned by routinized party structures might support the persistence of parties, which, in turn, might stabilize party competition. Yet the effects of routinization are not equivalent to effects that can be expected from value infusion in terms of generating citizens' support and attachment to political parties that is not instrumentally driven. This, in turn, suggests that once we move toward understanding the contexts in which different modes of party formation and development are likely to occur, we might be able to tackle an even bigger challenge-to understand the latter's consequences for democracy.

ACKNOWLEDGMENTS
Many thanks go to Fernando Casal Bértoa, André Kaiser, Gabriel Katz, Juan Pablo Luna, Thomas Morton, Ingo Rohlfing, Ulrich Sieberer, Valeria Smirnova, Marco Steenbergen, as well as the three JOP referees and the JOP editor for their detailed and helpful feedback on the article. We also thank the staff of the Cologne Center for Comparative Politics at the University of Cologne and the Centre for Elections, Media and Participation at the University of Exeter for their invaluable input, who read and commented on early versions of the article. The same goes for colleagues attending the European Consortium for Political Research General Conference in Bordeaux, where we presented this research at an early stage. All remaining errors are solely ours.