Introducing new multilevel datasets: Party systems at the district and national levels

Abstract

For decades, datasets on national-level elections have contributed to knowledge on what shapes national party systems. More recently, datasets on elections at the district level have advanced research on subnational party competition. Yet, to our knowledge, no publicly accessible dataset with observations of the party system at both national and district levels exists, limiting the ease with which cross-level comparisons can be made. To fill this gap, we release two corresponding datasets, the National Level Party Systems dataset and the District Level Party Systems dataset, where the unit of analysis is the party system within either the national or district jurisdiction. More than 50 elections in the two datasets are overlapping, meaning they include observations for a single election at both the district and national levels. In addition to conventional measures such as the effective number of parties, we also include underutilized variables, such as the size of the largest party, list type, and the vote shares for presidential candidates in corresponding elections.

Introduction

The comparative study of electoral and party systems has advanced to a stage where analyzing only national-level datasets is insufficient (Kollman, 2018). Many works that seek to understand party systems employ only national-level data (e.g., Taagepera and Shugart, 1989; Amorim Neto and Cox, 1997; Clark and Golder, 2006), notwithstanding arguments that such questions should be studied from the bottom up, that is, in terms of “coordination” in districts (Cox, 1997). Recently, Shugart and Taagepera (2017) re-asserted the primacy of national-level electoral institutions on district-level party-system outcomes. To facilitate further research bridging national and district levels, we are releasing two new datasets that can be used in conjunction, the District Level Party System Dataset (DLPS) and the National Level Party System Dataset (NLPS).

Among existing party system datasets, we have yet to see one that includes both national- and district-level variables. Mackie and Rose (2016) and Bormann and Golder (2013) capture each country’s overall electoral landscape, but not cross-district variation. The Constituency-Level Elections Archive (CLEA) by Kollman et al. (2017) captures the election information of each party in each district across hundreds of elections, but to date, has not provided a suite of variables that aggregate across parties within districts. Moreover, none of these data efforts provide both national- and district-level data in one place.

Maintaining internally consistent coding criteria, we organize two datasets that describe democratic party systems at each level, whose overlapping cases can be merged to perform cross-level analysis with new variables included in the NLPS described below. These cases will allow scholars to ask a variety of questions on how national party systems affect those at the district level and vice versa. For example, which countries have more cross-district variation of parties, and how does this variation affect the national party system? And under what conditions does fragmentation at the national level correspond to within-district fragmentation as opposed to cross-district fragmentation?

During previous research, we concluded that many relevant election variables were overlooked in other datasets. Therefore, in addition to creating corresponding national- and district-level datasets, we include a number of variables in the NLPS, such as the size of the largest party, the total number of legislative parties in both levels, list type, the names of presidential parties, presidential runner-up parties, and largest legislative parties. Further, the NLPS is the first cross-national dataset we are aware of to include the individual vote shares of presidential candidates in the first round and both candidates in the second round.

Case selection, format, and sources

Because the NLPS covers more elections than the DLPS, we keep the datasets as separate files. Scholars can merge the files together using country, year, and second_election. ¹ Both the NLPS and DLPS are available in tab delimited and Stata formats; they are stored at the Research & Politics Dataverse (https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/ME2W6U) with their codebooks. The list of overlapping country elections included in the NLPS and DLPS are documented in Table 1 of the Online Appendix.

The NLPS is structured so that each row is a single country-election year. Within the dataset, we offer a more explicit set of case-selection criteria than the Bormann and Golder (2013) dataset. Bormann and Golder (2013) include certain countries with only one or two elections without power alternation, such as Burundi, Bhutan, Myanmar (1951–1960), and Cuba (1948), seemingly suggesting a lenient definition of democracy, but exclude current democracies with dominant parties such as South Africa and Namibia. To achieve a more internally comparable sample, we only include countries with at least three consecutive elections and one power alternation. Older national election data are retrieved primarily from Nohlen and Stover (2010), Nohlen et al. (2001), Nohlen et al. (1999), and Nohlen (2005). Data on more recent elections are primarily from Adam Carr’s (N.D.) election archive, Psephos.

In the DLPS, the raw vote and seat share data are sourced primarily from the CLEA (Kollman et al., 2007) Releases 1–7, but with additions from Carr (N.D.), Election Resources on the Internet (N.D.), and national election offices. Because of complications in defining what a “district” is in some multi-tier systems, the DLPS includes only one-tier systems. We also exclude elections where missing vote and seat shares for individual parties make it impossible to calculate district-level aggregate variables. After gathering raw data, correcting errors, and summing vote shares in districts where parties run multiple candidates, we transform individual party observations into district-level observations and calculate a range of district-level variables, as described in the following section.

Unique features of the datasets

Variables that are not included in CLEA, Bormann and Golder, or other significant national and district-level datasets are listed in Table 2 of the Online Appendix. All NLPS variables are available for the DLPS country elections. We describe the key unique features and variables of the NLPS and DLPS datasets below.

The size of the largest party and related measures

In addition to the standard measure for party system fragmentation, the “effective number of parties” (Laakso and Taagepera, 1979), we record the largest parties’ seat shares (s₁) and vote shares (v₁), two crucial but often overlooked variables, for both the national and district levels. The largest party’s seat share conveys whether a single party exercises potentially unilateral power over various functions (e.g., passing legislation, amending the constitution, forming a cabinet in a parliamentary system). The largest party vote share allows researchers to assess the nature of a country’s main social cleavage (Li, 2017).

We also record the largest legislative party’s name (pty_n_s1), which can be convenient for studying power alternation and inter-branch relationships. Occasionally, the party with the most votes may not have the most seats (known as plurality reversal) due to disproportionality; the NLPS includes a variable (plu_reversal) to readily identify such cases.

Party-system variables in systems with alliances

Our datasets include several countries where two or more parties present joint candidacies or lists. In some cases, it is analytically advantageous to use the alliance rather than the party as the unit of analysis for variables such as the effective number of “parties” and largest “party”. In the NLPS and DLPS, we treat alliances as the observational unit only when two criteria are satisfied: (a) the electoral system directly allocates seats to alliances in districts, and (b) alliances themselves have a nationwide scope. A few examples illustrate how and why these criteria are used.

Chile’s pre-2017 system, open-list proportional representation with two-seat districts, satisfies the criteria at the national and district levels. Alliances presented two-candidate lists with candidates from different parties, with alliance composition consistent across districts. Thus, we treat the alliances as if they were parties in both the NLPS and DLPS. Likewise, Brazil and Finland allocate seats to lists often presented by alliances. In these cases, we follow the same procedure as for Chile in the DLPS, presenting the alliance vote and seat totals that the electoral formula directly operates on. However, at the national level, measures must be calculated on the basis of the individual parties because the alliances are not nationwide (our second condition). Any given party may be in alliance with different parties in different districts, rendering it impossible to aggregate alliances as electoral and seat-winning agents at the national level. ²

Other cases to note are Australia and Germany. We count alliance partners such as the Liberal and National parties (Australia) or the Christian Democratic Union and Christian Social Union (Germany) individually. These alliances are the exact opposite of what is observed in Brazil and Finland, in that the parties are effectively in a permanent alliance at the national level but run as distinct organizations at the district level (although not typically competing against one another in the same district).

Presidential election variables

In the NLPS, we record presidential election variables in the same row as the following or concurrent assembly election. They include many previously unavailable variables such as the vote share of the top six presidential candidates that are larger than 1%, top two presidential candidates’ party names (pty_n_pres, pty_n_runner_up_pres), whether the first round runner up wins the second round (runoff_comeback), and whether the president’s party has the most seats (pres_largest). To our knowledge, these variables are not available in other national-level election datasets. Such data can help scholars measure the competitiveness, fragmentation, and volatility of presidential elections, as well as their relationships with their countries’ assembly elections.

Variables for studying small parties

We record an unweighted total number of seat-winning parties in addition to the effective number of parties for both national and district levels (N_{s 0} and N_{s0_dist}). At the district level, we also count the total number of vote-winning parties (N_{v 0} _dist). These variables contain highly valuable information, because changes are often associated with the emergence and disappearance of small parties. These variables may help scholars understand the persistence of niche groups and detect institutional and social changes that are less visible in studies using N_s1 or s₁.

List type

In the NLPS, we also include a variable for the type of list for PR systems or the list tier of mixed systems, including open, closed, or the intermediate type known as flexible or semi-open (listtype). This variable is crucial for analyzing the often neglected intra-party dimension of representation (Shugart, 2008; Colomer, 2011), including candidates’ incentives to cultivate a personal vote (Carey and Shugart, 1995).

Other district-level variables

In the DLPS, we include a collection of district-level summary variables, such as the effective number of seat-winning and vote-earning parties (Ns_dist, v_dist), the five largest parties’ sizes and names in order of rank, and the smallest and largest district magnitudes by country election (mag_min, mag_max). The NLPS variables can be easily merged into the DLPS to carry out cross-level analysis. For instance, one might consider whether third-party vote share has a relationship with district magnitude or with fragmentation at the national level. Party names also allow researchers to explore questions such as whether regionally focused parties run candidates outside their core region.

Conclusion

The NLPS and DLPS provide data that link the national and district levels of many elections with new variables that should further institutional studies. We hope comparative politics scholars will explore new research opportunities with either or both of the datasets and expand them in the future.

Supplemental Material

online_appendix – Supplemental material for Introducing new multilevel datasets: Party systems at the district and national levels

Supplemental material, online_appendix for Introducing new multilevel datasets: Party systems at the district and national levels by Cory L. Struthers, Yuhui Li and Matthew S. Shugart in Research & Politics