national elections
Presented to Dr. Afsoun Afsahi
and Dr. Floris Vermeulen
by Boris van Puffelen 10782834
June 2019
Master thesis Political Science
Specialisation: Political Theory
Department of Political Science
University of Amsterdam
Contents
1 Introduction 1
2 Theoretical framework 5
2.1 Parties and candidate selection procedures . . . 5
2.1.1 Features of candidate selection procedures . . . 5
2.1.2 Centralisation . . . 9
2.1.3 Inclusiveness . . . 11
2.2 The impact on candidates . . . 12
2.3 Comparing Countries and Political Families . . . 13
3 Methodology and data 15 3.1 Data . . . 16
3.1.1 Integrated Party Organization Dataset . . . 16
3.1.2 Country, Party Family and Centrality . . . 17
3.1.3 Female Candidates and the Country discussion . . . 19
3.2 Regression and Logit function . . . 25
4 Conclusions 26 4.1 Centralisation . . . 26
4.2 Regression Analysis . . . 28
4.2.1 Limitations . . . 30
5 Appendices 36 5.1 Appendix A: Table of Included Cases . . . 36
1
Introduction
Politicians and political parties are very much the main actors in the political show, with voters showing up only every few years and lobby groups and civil servants often acting as stagehands behind the screens. This has led many a political scientist to dedicate much of their time and effort to the study of their acts, characteristics etc. It is therefore all the more surprising that candidate selection procedures, one of the defining features of political parties and one of the main ways that political parties and (new) politicians interact, have received little attention from political scientists (Cross 2008, 597;Katz 2001, 278). I contribute to closing this gap in this thesis.
Specifically, I will focus on the impact of candidate selection procedures on the representation of women. There is some concern that specific characteristics of candi-date selection procedures can be barriers to the representation of women in political bodies (Rahat 2007; Vandeleene 2017). The reason for focusing on the role of can-didate selection on the representation of female cancan-didates is that parties function gatekeepers to the political arena. Though parties often started out as loose asso-ciations of legislators over time, and with the continued extension of the suffrage, these loose associations developed further into stable features of many a political system and the role of parties also started to adapt (Mair 1990, 1-17). As parties became longer older and stronger institutions the need of selecting candidates to replace those representatives who retired from politics developed. Because parties can select any candidate they like they effectively started functioning as gatekeepers of the political arena, excluding those that they did not select as candidates (Cross 2008, 607; Rahat 2007, 158). As not all parties present gender balanced candidate groups (in this study it ranges from 12% for the Irish Fianna Fail to 49% for the Finnish National Coalition) it could be expected that this is part of the reason for the gender unbalanced representation of women in parliaments. This thesis will therefore concern itself with the question:
What is the impact of candidate selection procedures on the
representa-tion of women as party candidates?
Though there seems to be no systematic study of the history of candidate selec-tion procedures it seems to have at least been considered a significant feature of the party system as far back as 1942 by E. E. Schattsneider and other scholars have later reaffirmed this (Cross 2008, 597; Katz 2001, 278). Yet study of candidate selec-tion procedures seems to have hidden itself from scholarly attenselec-tion as few scholars engage with the subject (Cross 2008; Rahat 2007). Though I have little evidence for the reason why this is one possibility springs to mind. Information concerning candidate selection procedures is difficult to collect as the process primarily takes place within political parties and have received both less public attention (outside of leadership contests or US primaries) and are not always publicly accessible to outsiders. This does put barriers in place for studying these processes.
This lack of interest seems to have turned a little at the start of the 21st cen-tury, with papers focusing on questions concerning candidate selection procedures. Shomer (2014, 533) identifies three distinct questions concerning candidate selection procedures.
• What is the effect of candidate selection procedures on the behaviour of rep-resentatives?
• What are the determinants of candidate selection procedures?
• What is the effect of candidate selection procedures on the representativeness of the party list?
It is the third question that this paper is concerned with and which seems to have attracted the least interest from scholars. The gatekeeper function of political parties may be expected to significantly impact what kind of candidates they produce.
This is after all the purpose of candidate selection procedures: “how legislators gain permission to use the party’s banner” (Shomer 2014, 533). Of course, these selection decisions can be guided by a variety of different considerations. The first one that may spring to mind is the ideological position, their views on a (or a variety of) policy area(s) or the political skills of a candidate. Though considering that women are both under represented as political office holders means that it could be possible that gender may be another consideration in the selection of candidates to be put forward by political parties. These considerations are expected to interplay with the way that the candidate selection procedure is organised.
Rahat (2007, 166) argues that there are specific strategic incentives for selecting diverse candidate groups in centralised procedures (organised on a national level) whilst this is less true for decentralised (organised locally or regionally) procedures. Strategically it is beneficial for a party to select a diverse candidate group because then they run candidates that address specific constituencies. In centralised pro-cedures the group that selects candidates (the selectorate) has the overview to in-troduce this kind of diversity strategy, but that in decentralised procedures this overview and coordination is lacking and thus this strategy can’t be adopted.
What should be mentioned is the way that interested scholars have come to clas-sify candidate selection procedures as there has been considerable discussion about how to do this. Two groups seem to have developed. One that argues that these procedures should be classified according to their inclusiveness (what are the back-ground characteristics of the selectorate) and centralisation (whether the decision is taken by the local, regional or national organisation)(Cross 2008, 598-600). (See Vandeleene (2017) for a study including both these concepts.) The other group that these two concepts are indistinguishable and should therefore be taken as one concept, often centralisation (see for instance Bille (2001, 365) who argue that decen-tralised procedures can also be considered more inclusive as allows greater numbers to participate). This seems to have led to a lack of classifying parties by how
inclu-sive their selection procedures are, at least in available datasets. It is therefore that I cannot include inclusiveness as a concept, though I would’ve preferred it.
The reason for studying specifically the representation of women is partially practical but also it does fit with other research intro candidate selection procedures and can therefore directly build on pre-existing work (citealt[]vandeleene2017a). The practical consideration is that most data concerning candidate selection procedures that is publicly available was collected from the 1980s to the early 2000s, but that I am not aware of any dataset that registers any other characteristics of candidates from this period (Giger & Schumacher 2015).
To come to an answer to the main question a regression analysis will be performed using the measure of centralisation of candidate selection procedures of 28 parties in six countries. The regression will include a measure of centralisation from the Integrated Party Organization Dataset (IPOD) developed by Giger & Schumacher (2015) out of a variety of different sources. The three that will be relevant for this thesis will be the Parliaments and Governments Database (Parlgov) for general data on parties and specifically the Party Family they belong to and in which Countries they operate, the combination of the work by Bille (2001) and Lundell (2004) about the degree of centralization of party procedures and self-collected data on the pro-portion of female candidates for the parties. This data will be used to develop an argument about the impact of candidate selection procedures on the representation of women and answer the main question.
The summary conclusion is that the degree of Centralisation in this party does not seem to have any impact on the representation of women among candidates. Partially this may be because of limitations for this thesis in particular (which are address in depth in the conclusion), but also because the effect of differences between Countries and Party Families may be much greater than the effect of how they organise their candidate selection procedures.
2
Theoretical framework
This paper focusses on the effect of candidate selection procedures on the repre-sentation of women. To develop the theoretical framework necessary to study this effect we will focus on two sources of theory. Firstly, work concerning the practices of candidate selection procedures in general and, Secondly, theory on the effects of these procedures on the representation of women. These two sections will conclude in the formulation of a hypothesis that will be tested further in the paper.
2.1 Parties and candidate selection procedures
In democratic systems parties field candidates for elections (Hazan & Voerman 2006, 148). Though there are some exceptions where individual independents run, most candidates are nominated by parties. To decide which candidates to field parties employ different candidate selection procedures. These methods vary considerably (Hazan & Voerman 2006, 147).
The function of parties in candidate selection make them a gatekeeper for many democratic political systems (Cross 2008, 607; Rahat 2007, 158). It is this gatekeep-ing function I suspect that will have an effect on the representation of women. Yet there are few studies of the effects of candidate selection procedures in general or concerning this female representation in particular (Cross 2008, 596).
2.1.1 Features of candidate selection procedures
To understand the differences between candidate selection procedures it is important to look at different concepts and questions related to these selection procedures. Rahat & Hazan (2001, 298-299) identify four such themes in their work:
• Who can be selected? For any candidate procedure it is important to contemplate who can become a candidate for that party. Limiting factors here can be that only members may stand as candidates for many parties, but
national laws also have a significant impact on who can be selected in that they give some inhabitants the right to stand as a candidate in elections. Parties work within a national framework of laws that effect their candidate selection procedures.
• Who selects candidates? For any selection procedure there are the people doing the selecting. There are parties where all members or just national leaders select the candidates. This of course has an impact because most parties are not perfect representations of the electorate and will have their own hang-ups and logic, but also different selectorates will have their own strategic incentives for selection procedures. It is this question that generally in meant when scholars in this field speak about inclusiveness (Bille 2001; Cross 2008; Vandeleene 2017).
• Where are candidates selected? Are candidates selected nationally or sub-nationally? And if it is sub-nationally then is it regionally or locally. this theme is captured in the concept of centralisation (Bille 2001; Vandeleene 2017).
• How are candidates nominated? There are procedures in which there is a public vote, a nomination by a board or where the nominations by the board are almost always followed. This relates of course to the selectorate for question 2.
List taken from Hazan & Voerman (2006, 147).
All these questions are important, but difficult to answer for any group of parties. The finesses of any selection procedure can be complex (Rahat & Hazan 2001, 299-300). To overcome some scholars have tried to develop dimensions on which parties can be classified. These dimensions classify parties on relevant aspects of candidate nomination procedures. Though this may not do justice to every peculiarity they
are considered to at least give a good enough understanding of candidate selection procedures so as to be able to compare parties (Cross 2008, 598-600).
Rahat & Hazan (2001, 300) suggest several dimensions by which these themes can be further worked out:
• Inclusiveness of candidacy. This dimension runs from allowing any citizen with passive suffrage to stand for candidacy to highly exclusive systems where there are many conditions for be allowed to run (Rahat & Hazan 2001, 300-301). Examples of this would be having to be a member for a long time, having to have provided services to the party (campaigning) or having to be part of a group of significant members. Even more radically are parties that limit candidacy to specific groups of people like the Dutch Staatskundig Gereformeerde Partij (SGP) which excluded women from being candidates until 2012 (Willems 2012).
• Inclusiveness of selectorate. The selectorate is the group that effectively selects the candidate. This runs from allowing the entire electorate (or regis-tered supporters) to select the candidates to just allowing a leader to select the candidates (Rahat & Hazan 2001, 301-303). Significantly also is that for parties that allow their members to select candidates it should also be kept in mind who gets to be a party member (Rahat & Hazan 2001, 302). Besides this there are also parties where nomination comities select candidates, but here remains of course the question who mandates the selectorate, are they mandated by the leadership or by the members (Rahat & Hazan 2001, 302).
• Centralisation Rahat & Hazan (2001, 304) argue that there are two forms of decentralisation. Firstly territorial, where the candidate selection procedures are localised nationally, regionally or locally. Secondly corporate, where it con-cerns functional representation of relevant constituencies. Are women, ethnic minorities or trade unions (for instance) included in the selection process in a
way that is distinct form general incorporation into the selectorate. (An exam-ple would be that specific groups get to select a certain number of candidates.) These two different forms of centralisation provide a field in which parties can operate.
• Nomination systems. There are (often) more aspirant candidates then can-didacy positions. This means that there must be some way in place to select from this pool. Rahat & Hazan (2001, 206) identify two extremes in this di-mension: Appointment and voting. But these can take different forms. The selectorate can be allowed to vote for a proposed model list or individual candi-dates. There is also the question of how voting by the selectorate is conducted: are there rounds or a single vote, majoritarian or proportional. The reason for different systems can be complex, but it may be expected that smaller selec-torates allow for voting procedures with more rounds (Rahat & Hazan 2001, 309).
List taken from Rahat & Hazan (2001, 300).
To engage with the difficulties of understanding selection procedures scholars have tried to capture them in two distinct concepts: centralisation and inclusiveness (Cross 2008; Lundell 2004; Vandeleene 2017). It is noteworthy that the questions concerning the systems and candidate inclusiveness have received less attention. This was surprising as some authors have assumed that it is inclusiveness which would most strongly impact the types of candidates that are selected (Rahat in Martin et al. 2014, 6; Vandeleene 2017, 8-9). But this does not necessarily appear to be true as Vandeleene (2017, 16) finds, in her study of that it is centrality of the process has an impact on the on the representation of women among the candidates, but inclusiveness doesn’t have a significant impact.
2.1.2 Centralisation
The location of candidate selection can have a significant impact on the way it is organised. There can be made a distinction between placing the locus of candidate selection with the national leadership or local members (Bille 2001, 367-369; Cross 2008, 598-599). The reason for the localisation seems to be complex, but mainly rely on characteristics of parties and not on the electoral system (Lundell 2004, 39-41). The only relation between political system and the candidate selection procedures of active political parties Lundell (2004, 27) finds is the degree of federalism.
Bille (2001, 367) has developed a scale on which to rate parties by how centralised their selection procedure is:
1. The national party organs completely control the selection of candidates.
2. The subnational party organs propose candidates, but the national party or-gans make the final decision.
3. The national party organs provide a list of names from which the subnational party organs can select the final list.
4. The subnational party organs decide, subject to the approval of the national party organs, including the right to add or delete names according to a variety of stipulated qualifications.
5. The subnational party organs completely control the process and make the final decision.
6. Finally, it is of interest to see whether a membership ballot is introduced, thus making the process more inclusive.
(List taken from (Bille 2001, 367))
As stated before Bille (2001, 365) argues that centralisation should be taken as the most important way to classify candidate selection procedures. The sixth
category does mention inclusiveness (and is numbered six in Bille’s article) but is used as a separate categorical variable. Lundell (2004, 31) adapts the list into a new ordinal categorisation:
1. Selection at local party meetings, by local selection committees or by primaries open for all party members.
2. Selection at the district level by a selection committee, by the executive district organ or at a convention (congress, conference) by delegates from the local parties.
3. The same as 1 or 2 but regional or national organs exercise influence over the selection process, e.g. add names to the lists or have veto power. The decision, however, is taken at the district level. Formal approval by regional or national organs without actual involvement in the process belongs to the second category.
4. The same as 5, but local, district or regional organs exercise influence over the selection process, e.g. party members, the local parties or committees at the constituency or the regional level propose candidates. The decision, however, is taken at the national level.
5. Selection by the party leader, by the national executive organ, by a national selection committee, or by primaries at the national level.
(list taken from Lundell 2004, 31)
Lundell (2004, 31) uses this categorisation as a dependant variable to measure how centralised the selection process is for different parties. this because she consid-ers the concept of centralisation a more straightforward way of classifying candidate selection procedures (Lundell 2004, 30).
This paper will use a combination of the Bille 2001 and Lundell 2004 categori-sation from the Integrated Party Organization Dataset (Giger & Schumacher 2015)
as it extends later than the date by Bille or Lundell individually. The new ordinal categorisation has been based on the original scale by Bille and has reworked the data collected by Lundell so as to be able to combine them.
2.1.3 Inclusiveness
Inclusiveness on the other hand has received less attention ten centralisation. Van-deleene (2017) states that this has led to it being not having an established method of categorisation. Rahat (2009, 3) does provide a one but it is limited to three groups:
1. Party members, who directly select candidates in party primaries, represent the most inclusive method. The purest type of party primaries is where the voting of party members alone decides the composition and rank of the candidate list or the candidacy in each Single Member District (SMD).
2. Candidate selection by a selected agency of the party, composed of delegates who were selected by party members, represents a medium level of inclusive-ness.
3. The highly exclusive kind of selectorate is the nomination committee, a small group that is usually composed of a few leaders or their aficionados.
(list taken from Rahat 2009, 3)
Rahat in (Martin et al. 2014, 6) argues that inclusiveness is the most important factor for the study of candidate selection procedures, but this is not expressed in the amount of research into the subject (for an exception to this see Vandeleene et al. (2013)). This leads to the problem that there are few studies and no international datasets that classify parties by the inclusiveness of their selection procedures. This has made inclusiveness impossible to use as an independent variable for the scope of this paper.
2.2 The impact on candidates
There is surprisingly little work done on the effect of candidate selection procedures on the candidates that come out of the procedures. Mainly the focus has been on research into the reasons why candidate selection committees are organised the way they are (see Lundell (2004)), on the effect of candidate selection procedures on party discipline (see Depauw & Martin (2009)) or on the effect of candidate selection procedures on the ideological congruence between the masses and the elites (see Vandeleene et al. (2013)). In this paper I hope to expand on this this field by looking at the effect of these selection procedures on the characteristics of the candidates it produces. To do so it is necessary to develop expectation about the way that differing forms of candidate selection procedures impact the representation of women.
Vandeleene (2017) is one of the few researchers that has focussed on the rela-tion between the centralisarela-tion and inclusiveness of candidate selecrela-tion procedures and the representations of women on candidate lists. She finds that the variable centralisation does have an effect on the presence of female candidates and that a more decentralised organisation means a better balance between male and female candidates (Vandeleene 2017, 16). At the same time, she finds that this does not hold true for the variable inclusiveness. She also argues that gender balanced selec-torates lead to more gender-balanced candidate lists and finds that this holds true for the cases she researches (Vandeleene 2017, 16). But van she does only study parties in Belgium, a country known for producing relatively gender balanced lists (Vandeleene 2017, 15-17).
Vandeleene’s (2017) findings contrast with the theoretical expectations where the expectation would be that the increased overview over the totality of the candidates would allow for the lead to a more focussed management of the gender balance. Ra-hat (2007, 166) for instance argues tRa-hat highly inclusive methods and decentralised candidate selection methods do not allow selectors to effectively coordinate with one
and other and therefore may be expected to produce less representative results. He therefore argues that if a representative outcome is desired it may be necessary to have methods in place that restrict selectorate choice (for instance strong quota’s) or restrict the size of the selectorate. Especially as there may be strategic incentives for parties to present a diverse candidate list so as to address differing parts of the electorate whilst in broader selectorates the lack of overview allows for less strategic considerations (Vandeleene 2017, 3). That being said the question remains whether Vandeleene’s (2017) findings hold for a broader group of countries. As she herself mentions Belgium is a country that that has been considered a leader when it comes to the political gender-balance. It may be that this relation does not hold for coun-tries where ideas about gender and leadership are less inclined towards a balance. This would be expected as evidence has been found for the idea that ideology, in particular specific beliefs about the role of women in society, does have an impact on the representation of women in parliament (Paxton & Kunovich 2003).
This leads me to formulate my hypothesis in line with the existing theory con-cerning the impact of candidate selection procedures.
H1: The degree of centralisation of candidate selection procedures positively
corre-lates to the representation of women among candidates for a party.
H0: The degree of candidate does not correlate to the representation of women
among candidates for a party.
2.3 Comparing Countries and Political Families
In any analysis of multiple countries there is a need to keep in mind that it can’t be expected that all countries have similar levels of female representation of women among candidates. This because ideas on the roles of women in politics or leadership positions or subconscious associations with women (Paxton & Kunovich 2003). Sim-ilarly, not all political movements hold similar views on women as political leaders
which means that for depending on the party family it may be expected that some parties are more or less prone to select women as candidates. Though to study this can be difficult as classifying countries by values is limited by the amount of data on this subject I’ve decided to control for country and party family.
Lundell (2004, 32; 35) boarder categories than party family and country. She groups the party families into three different groups named Left, Right and Centre, and she groups countries by region. Her argument for grouping parties by these three ideological categories is that parties on the extremes may be expected to be more ‘authoritarian’. This argument does not entirely translate to this thesis as how ‘authoritarian’ or extreme a party is cannot be expected to translate directly to an expected impact on the representation of women. That being said there is research that shows that ideological factors do have an effect on the representation of women in parliament. Paxton & Kunovich (2003) show that ideas that exist within a country about women’s participation (both in politics and in other parts of society like education, the workforce and housework) correlate to the representation of women in their parliaments. Though they do not directly relate these findings to parties it could well be that it is relation holds true for parties as well. They argue against using broad ideological categories as it has failed in the past to capture this effect, but their methodology is not available to this paper as the World Values Survey in the years that this paper is concerned with did not include the countries that are included in this paper (Paxton & Kunovich 2003, 102-105). I’ve therefore decided to include the less precise but available Parlgov data on party families that the IPOD dataset includes to classify parties (Giger & Schumacher 2015).
Lundell (2004, 35) argues that there may be regional factors that impact the candidate selection procedures itself. She groups countries into four regions, namely Northern, Central, Southern Europe1 and Westminster. She argues that these groups can be used as countries within them share political culture (and history
1As no cases are available in this thesis Southern Europe won’t be discussed further. For Lundell
in some cases) and try to imitate each other. The northern European region con-sists of the broad interpretation of Scandinavia (including Denmark and Finland2). The central group consists of Austria, Belgium, France, Ireland, Luxembourg, The Netherlands, Switzerland and the United Kingdom. And the Westminster system is an adaptation around the United Kingdom and its former Colonies which also includes Australia, Canada, Ireland and New Zealand. As the political culture and electoral system is expected to have an effect on the representation of women among party lists I will control for regions in the regression analysis.
A second concern for Country and Party Family is one of statistics. Peduzzi et al. (1996) showed in their paper that including variables that only pertain to few cases in a logistic regression analysis can lead to problems with the outcomes (they argue that less then 10 cases per variable would lead to concerning problems). Though Austin & Steyerberg (2015) find that this is not a problem for OLS regressions so there is no direct need to include regional variables (as all countries have more than two cases related to them) I will still run two distinct models for both broader categories and individual countries to maintain both the granularity of using variables more closely related to the individual included parties and to address parties in line with other research in this field and allow for better comparison to other studies.
Considering this I have decided to run two models for the regression. One with broad categories that combine families and countries into broader groups and one where the narrower original data is used as control variables for the effect of cen-tralisation on the representation of women as candidates for political parties.
3
Methodology and data
In this section I will address where the data for this paper comes from and how it will be used to develop the conclusions. This will include a discussion of the
2Lundell (2004, 35) excludes Finland because it has laws concerning the centralisation of
candi-date selection procedures. I don’t need to as this is not my dependant variable and so how candicandi-date selection procedures come about is not relevant to this thesis.
dataset and the collected data for the different countries. What should be noted is that this will be a party level analysis and that, even though I expect that local circumstances (Country) or ideological circumstances (Party Family) may be signif-icant factors, any data on these should be interpreted with caution as they are quite all-encompassing categories.
3.1 Data
This paper will only make use of data collected by others. The necessary data will be supplied by one dataset (Centralisation; Party Family and Country) and collection by the author from other sources. The dataset is the Integrated Party Organization Dataset (IPOD) developed by Giger & Schumacher (2015).
3.1.1 Integrated Party Organization Dataset
This dataset was developed by combing a variety of different datasets related to parties. For this paper in particular the data on centralization (called cent in the codebook and the dataset) will be relevant. It is a combination of both the work by Bille and Lundell on classifying candidate selection procedures. Though it is of course desirable to take a measurement at the same time, this is impossible as the parties only field candidates during elections. To have multiple cases on which to base the regression I will therefore look at elections in the dataset spanning five years (2000-2005). I’ve chosen for this limited timeframe so as to not allow for too great a variation on the development of values concerning women’s participation.
The data used in from this dataset comes from two sets of others. Firstly, the Parlgov data concerning party family, country and party name. Secondly, a reworking by the developers of the dataset, Giger & Schumacher (2015) of Lundell’s and Bille’s classification. This new variable was created to extend available data concerning the centralisation of selection procedures. The data was changed firstly by flipping the scale by Lundell as it ran in the opposite direction of the scale
by Bille. Secondly Lundell’s middle category was slightly adapted to more closely approximate the scale by Bille 3The reworking of the scale in this way means that it runs from 1: most centralised to 5: most decentralised. This effectively makes it a scale of decentralisation. I’ve therefore reworked it by multiplying by negative 1 and adding 5 so as to invert it.
General descriptions of the data In this timeframe there are elections in
four-teen countries (see appendix A for the complete list of parties). The complete data includes a total of 73 parties for which a classification is available, but not all parties are included in the final analysis. the reasons for excluding specific countries will be discussed further on in in the country by country section.
3.1.2 Country, Party Family and Centrality
Table 1: Number of parties by country
Country Number of Parties
Canada 3 Denmark 6 Finland 6 Great Britain 3 Ireland 5 Netherlands 5 Total 28
As can be seen here there is some regional diversity included in in the data, but also a distinct lack of southern and eastern European countries. Though this is a limiting factor in this analysis there is little that can be done about this as the data is not included in the dataset. This should be kept in mind with the interpretation of the regression analysis and how it applies to countries outside of northern Europe.
3
It should be noted that the codebook for this dataset does not contain an explanation of how the combination of these two scales was achieved. Through contact with one of the authors I’ve been able to establish that this is the method used.
Table 2: Number of parties by ideological group and Party Family
Ideological Group Party Family Number of Parties
Left Communist/Socialist 1
Green/Ecologist 3 Social Democratic 6
Centre Agrarian 1
Liberal 8
Right Christian Democratic 4
Conservative 5
Total 28
There is also a diverse grouping of different political families as can be seen in Table 2. This is important because, as expressed earlier, this metric will be used as a control for ideological position. Though there are no parties included that expressly exclude women from candidacy, it may be expected that parties from certain families are less open to women as candidates. The country and party family variables will be used as a way to check for a group of environmental concerns. These categorical variables can, of course not be included directly in the regression as no ranking can be assigned to them. This means that there is a need to recode these into dummy variables. As there is a need to exclude one dummy variable from (to function as a sort of baseline) for countries I’ve excluded the central European region and the Netherlands as these would provide the greatest consistency between the two models and for the party families I’ve excluded the Liberal and Centre categories for the same reason.
Table 3: Number of parties by centrality
Degree of Centrality Number of parties
0 15
1 9
3 4
Total 28
there are no parties that are classified as a one or a three.
3.1.3 Female Candidates and the Country discussion
The data for the candidates running in the elections does not come from a single dataset as no one was available for the 2000-2005 period. It is therefore that the data has been collected from a variety of different sources, one for each country. Though it would be preferable to have a single dataset as it would mean that collection could be assured to be consistent it is presumed that this won’t be a fundamental problem as the formality of the candidacy process would imply a clear distinction between who is and isn’t a candidate and therefore reasonable consistency of collection. The corresponding country by country data can be seen in tables 4 to 9 and an overview of all included parties can be found in Appendix 11.
Canada For Canada the number of female candidates is taken from a report by
Elections Canada (The Office of the Chief Electoral Officer) which reports to par-liament on Canadian elections. It splits the number of female candidates by party (available by province or territory) (Elections Canada 2018).
Great Britain The data on Great Britain was taken from a report by Morgan
(2001, 9) on the election results of the 2001 election for the Library of the House of Commons.
Ireland For Ireland the data was taken from a paper by McElroy & Marsh (2010,
4).
Denmark The data on the Danish parties comes from a report by the Danish
Ministry for Interior and Health (Indenrigs- og Sundhedsministeriet 2003).
Finland Unlike the other countries in this paper there is no data on the numbers
resource available reporting on the percentages of female candidates by party. I’ve reported the data for the parties in table 8. Though it reports on the major parties one of the smaller parties that is included in the IPOD dataset, The Finns Party (formerly often translated as the True Finns), is lumped under other parties so they have been excluded from this analysis.
Netherlands The Netherlands reports the gender of candidates on the ballot
pa-per. The data has therefore been collected by counting the total and female candi-dates directly from the ballot paper. Though the Netherlands does not have electoral districts it does make use of ‘election circles’4, areas between which party lists can vary slightly for the final few candidates. The reported candidate numbers include just candidates that appear on the lists in all ‘electoral circles’ for the included parties.
4Candidates that stand in few or one ‘electoral circle’ are often local politicians or locally
T able 4: Canada New Demo cratic P art y Lib eral P art y of Canada Progressiv e Conserv ativ e P art y Cen tralit y 0 1 0 F emale candidates 88 65 39 T otal Candidates 298 301 291 Prop ortion 0,2953 0,2159 0,1340 T able 4: (Elections Canada 2018) T able 5: Great Britain Conserv ativ e P art y Lab our P art y Lib eral Demo crati c P art y Cen tralit y 1 0 1 F emale candidates 92 149 139 T otal Candidates 643 640 639 Prop ortion 0,1431 0, 2328 0,2175 T able 5: (Morgan 2001, 9)
T able 6: Ireland
Cen
tralit
y
F
emale
Candidates
T
otal
Candidates
Prop
ortion
Fianna
F
ail
3
13
106
0,1226
Lab
our
P
art
y
1
12
47
0,2553
Progressiv
e
De
mo
crats
3
6
20
0,3
Fine
Gael
3
14
85
0,1647
Green
P
ar
ty
1
9
31
0,2903
T able 6: (McElro y & Marsh 2010, 4) T able 7: DenmarkCen
tralit
y
F
emale
Candidates
T
otal
Candidates
Prop
ortion
Radical
P
ar
ty
0
34
98
0,3469
Conserv
ativ
e
P
eople’s
P
art
y
0
27
102
0,2647
Christian
P
eople’s
P
art
y
0
21
84
0,25
Lib
eral
P
art
y
0
28
96
0,2917
So
cial
Demo
cratic
P
art
y
1
38
105
0,3619
So
cialist
P
eople’s
P
art
y
1
33
103
0,3204
T able 7: (Indenrigs-og Sundhedsminis te riet 2003, 4)T able 8: Finland
Cen
tre
P
art
y
0
0,4150
So
cial
Demo
cratic
P
art
y
0
0,4820
Sw
edish
P
eople’s
P
ar
ty
0
0,4150
National
Coalition
0
0,4910
Left-Wing
Alliance
0
0,4460
Christian
Demo
crats
0
0,4120
T able 8: (Myrskylä 2003, 4)T able 9: Netherlands
Cen
tralit
y
F
emale
Candidates
T
otal
Candidates
Prop
ortion
Christian
Demo
cratic
App
eal
1
21
57
0,3684
Demo
crats
66
0
9
25
0,3600
Lab
our
P
art
y
1
30
75
0,4000
Green
Left
0
11
24
0,4583
P
eople’s
P
art
for
F
reedom
and
Demo
c
racy
3
18
75
0,2400
T able 9: (Cen traal Stem bureau 2002)Excluded Cases Several countries that the IPOD database does contain data on have been excluded from the analysis in this thesis. Australia, France, Germany, New Zealand, Norway and Sweden have been excluded because no data could be found on the proportion of female candidates for the elections in the researched period. Austria and Belgium have been excluded as there were legal quota rules in place for parties in selecting their candidates. Though Vandeleene (2017; 2014) did develop a method for dealing with this issue, namely taking the proportion of female candidates on a party list placed on a position on the list or nominated where they have a reasonable chance of winning, this would’ve required establishing what constituencies and what positions on the party list would’ve had such a chance this would have taken too long for the timeframe of this thesis. I’ve therefore decided to exclude them for the purposes of this paper.
3.2 Regression and Logit function
The core of this thesis is a regression analysis based on the collected data. This is not wholly without problems. The main problem is the dependant variable, as it is bound within a narrow band between zero and 1 (in theory as no party actually scores either of these). This is a problem as a regression analysis assumes that the dependant variable is unbounded. This problem was similarly faced by Vandeleene (2017, 14) and I will make use of the same solution. Namely to remap the variable form the finite space (0;1) to an infinite space (-infinity; infinity) through the use of the natural log.
ln(ProportionFemale)=ln((Proportionfemale)/(1-ProportionFemale))
This does mean that the eventual interpretation of the outcomes of the regression will have to be interpreted on the logarithmic scale, but it does allow me to run a linear regression as now the values are unbounded.
Figure 1: Scatterplot of the Proportion of female candidates by degree of centrali-sation
4
Conclusions
4.1 CentralisationThe simplest form of analysis is to have a look at the simple data in a scatterplot. What seems, at least for this simple form of analysis with scatterplot 1, to hold true is what Vandeleene (2017) also found. Namely that with increased centrali-sation the proportion of female candidates decreases. Yet the bottom line remains remarkably stable. Though no statistical inference can be drawn directly from scat-terplot 1 this stable bottom line could indicate that there is in these countries some minimum level of expected female representation among party candidates.
In the scatterplot 2 concerning the country level we see a picture as expected. Namely that these variability between parties in countries are smaller than the variability of all parties. That means that it is likely that expected proportions of female candidates are influenced at least to some degree by the countries that
Figure 2: Scatterplot of the Proportion of female candidates by Country
the parties operate in. One may expect that this is due to a variety of country specific factors, but it may be due to a hidden relationship between countries and the centrality measure (Paxton & Kunovich 2003). This would be in line with the findings of Lundell (2004, 39-41) that the electoral system is not related to the measure of centrality of parties, but region is. Though region and country re not exactly the same, and may hide considerable differences within these regions.
On the other hand, the Party Family, as represented in scatterplot 3 seems to have a far less consistent influence on the representation of female candidates. This may initially seem surprising as one might expect there to be a strong influence of ideology on the representation of women, but do not forget that the country variables are not held stable. To understand what the exact influences are it is necessary to keep these stable, as is done within a regression analysis.
Figure 3: Scatterplot of the Proportion of female candidates by Party Family
4.2 Regression Analysis
In table 10 the coefficients are presented. As can be seen in model 1 effect of centralisation on the proportion of female candidates seems initially promising as the metric is statistically significant. In the second and third model this disappears. It is therefore that we can conclude that H1 can be considered invalid and that
centralisation, at least on the basis of the limited data used in this thesis, has no significant effect on the representation of women.
As we can also see from table 10 with the adjusted R2, that model 1 only explains about 20% of the variance whilst model 2 explains 70%5. This would suggest that there are more in-depth explanations why in some regions the representation of women as candidates for parties is far more common than others. It may in the future be fruitful to rerun this form of analysis with a richer pallet of variables that
5Calling it ‘explains’ in this context may seem odd, as it would mean being a Nordic party is an
apt explanation of have more women on the candidate list. Here it should be taken to mean ‘reason for the variance’
Table 10: Regression Results
Variables Parameter Estimate Standard Error
Model 1 Centrality -0,4858** 0,089 Constant -0,6628 Adjusted R2 0,236 Model 2 Centrality -0,1749 0,07 Westminster -0,6395** 0,1776 Nordic 0,0113 0,184 Left 0,2047 0,1504 Right -0,2112 0,1545 Constant -0,5198 Adjusted R2 0,7059 Model 3 Centrality -0,2596 0,0732 Canada -0,4462** 0,2055 Denmark -0,2175 0,1651 Finland 0,2812 0,1965 Great Britain -0,4661** 0,2001 Ireland -0,3456 0,1872 Social Democracy 0,1533 0,1408 Conservative -0,2585 0,1512 Christian Democracy -0,1270 0,1612 Green/Ecologist 0,1262 0,1864 Agrarian -0,0803 0,2998 Communist/Socialist -0,0361 0,2998 Constant -0,6628 Adjusted R2 0,8747 Table 10: Sign.: * p < 0.05, ** p < 0.01.
Dependent variable: Ln proportion of female candidates Reported is the adjusted Beta.
Model 1 does not use control variables. Model 2 introduces the narrow control variables and model 3 the broad control variables.
could do more justice to the included countries and how this might influence the parties.
The ideological variables in this model aren’t statistically significant, which may mean that the region (or country) in which parties operate are far more important than their cross region/national ideological affiliation. This could mean that political parties, whatever their ideological affiliation, decide their issue positions concerning women more by interaction with domestic competitors and on the basis of the issue positions of their constituencies than on cross regional/national ideological affilia-tions.
Therefore, the answer to the central question of this paper would be that there is no impact of the candidate selection procedure on the representation of women as party candidates. That may be surprising as Vandeleene (2017) does find an effect. This could have three different causes. Firstly, that simply not a large enough group of parties was studied in this thesis. Secondly, that the effect for candidates in national elections is small, but that the effect is stronger in elections that are local, regional and/or European. Thirdly that there is something specific to the Belgian case that she studies that is not true for other cases. The thing that should be noted is that in this model, similar to Vandeleene (2017, 16), as far as there is any influence of centrality on the representation of women that it is negative. That would mean that there may indeed be some reason to revaluate the theoretical expectation, that some scholars have espoused, that centralisation is positively correlated to the representation of women (Rahat 2007).
4.2.1 Limitations
It is important to address the limitations of the conclusions on the basis of the regression analysis. Four limitations can be identified that relate to the availability, spread, party ideology and the gender balance of the selectorate.
Data Recent data concerning candidate selection procedures is limited. This led to a limited number of cases to be included in the regression analysis. At the same time there is only limited data available concerning the representation of women as candidates for political parties. In this paper I have tried to use data from trusted sources, as discussed in the methodology, but this type of work was unavailable for several countries in the IPOD dataset. Though other methods of collecting this data have been developed 6 these were not were not available due to time constraints. In other countries there were indeed reports on the proportion of female candidates but these were not split by party so not useful for the purposes of this thesis. This is the reason that only a limited amount of parties is included in this paper.
Regional spread The Centralisation measure in is regression analysis is not
sta-tistically significant, it should be mentioned that the used data does not extend to southern Europe or other party democracies throughout the world. Therefore, any inference would not have been able to be extended to these regions. This is further complicated by the fact that the there is only limited data available on the candidate selection procedures in southern Europe and other regions 7. This means that for any future research into this subject it would desirable to extend the data to non-and Southern European cases. This especially as Lundell (2004, 39) has argued that candidate selection procedures in Southern Europe may function differently than in other areas . Several of these countries democratised later than countries in other parts of Europe and there is often a large emphasis on charismatic leadership and authority in these countries (Lundell 2004, 40).
Ideology In this thesis I have used party families and regions to try and control
for likely country and ideology related factors that explain the variance of the
de-6Valdini (2012) uses for instance several sources on gendered names to classify entire candidate
lists
7The IPOD dataset (Giger & Schumacher 2015) only has classifications available on Greece, for
pendant variable. But this is certainly no perfect way of doing this. It would have been preferable to include actual data on the ideas that exist concerning female po-litical participation, or proxies for those, for the countries included in the regression. Valdini (2012, 743) for instance uses two questions from the European Values Sur-vey to classify countries by the prevalence of traditional gender norms. Similarly, Paxton & Kunovich (2003) argues that ideology should be understood in a different way. Namely focus more specifically on the values and beliefs in a country about women and their capability to function as political leaders and their role in society. Copying this methodology was not available to this paper as the European Values Survey doesn’t include countries outside of Europe and the World Values Survey doesn’t include all the countries inside of Europe. Though controlling for the region is a way of controlling for region specific factors, the disadvantage is that it gives no way to find out what specifically causes any variation on the dependant variable.
Gender balance of the selectorate Part of the idea behind this paper is that
the representation of women in politics is important. Yet in one significant way it fails to ‘practice what it preaches’. I would have preferred to include data on the characteristics of the selectorate. In particular a measure of the gender balance in selectorates would have contributed to the regression analysis. Vandeleene (2017, 13) includes the presence and absence of a gender quota for the selectorate as a control and finds that it is statistically significant. Though it would not have mattered for the statistical significance of the degree of centrality, it may in itself have said something about the candidate selection procedure. Though some older data was available on gender quotas it was collected in the 1980s and could not be expected to still give an accurate representation of the situation in this period and therefore was not included.
References
Austin, P. C. & Steyerberg, E. W. (2015), ‘The number of subjects per variable required in linear regression analyses’, Journal of clinical epidemiology 68(6), 627– 636.
Bille, L. (2001), ‘Democratizing a democratic procedure: Myth or reality? candidate selection in western european parties, 1960-1990’, Party politics 7(3), 363–380.
Cross, W. (2008), ‘Democratic norms and party candidate selection: Taking contex-tual factors into account’, Party Politics 14(5), 596–619.
Depauw, S. & Martin, S. (2009), ‘Legislative party discipline and cohesion in com-parative perspective’, Intra-party politics and coalition governments pp. 103–120.
Distribution of confirmed candidates, by political affiliation and sex 37th general election (2018), Technical report, Elections Canada.
URL: https://www.elections.ca/ele/pas/can/distribution.pdf
Folketingsvalget den 20. november 2001 (2003), Technical report, Indenrigs- og Sund-hedsministeriet, Slotsholmsgade 10-12 1216 København K. 17.44.
Giger, N. & Schumacher, G. (2015), Integrated Party Organization Dataset (IPOD).
URL: https://doi.org/10.7910/DVN/PE8TWP
Hazan, R. Y. & Voerman, G. (2006), ‘Electoral systems and candidate selection’, Acta Politica 41(2), 146–162.
Kandidatenlijsten verkiezing leden Tweede Kamer der Staten-Generaal (2002), Tech-nical report, Centraal Stembureau.
URL: https://zoek.officielebekendmakingen.nl/stcrt-2002-75-p22-SC34204.html
Katz, R. S. (2001), ‘The problem of candidate selection and models of party democ-racy’, Party politics 7(3), 277–296.
Lundell, K. (2004), ‘Determinants of candidate selection: The degree of centraliza-tion in comparative perspective’, Party politics 10(1), 25–47.
Mair, P. (1990), The West European Party System, Oxford University Press on Demand.
Martin, S., Saalfeld, T. & Strøm, K. (2014), ‘The oxford handbook of legislative studies’.
McElroy, G. & Marsh, M. (2010), ‘Candidate gender and voter choice: Analy-sis from a multimember preferential voting system’, Political Research Quarterly
63(4), 822–833.
Morgan, B. (2001), General election results, 7 june 2001, Technical report, House of Commons Library, Room 407, 1 Derby Gate, London, SW1A 2DG.
Myrskylä, P. (2003), Parliamentary elections 2003 - background information on the candidates and elected mps parliamentary elections 2003 - background information on the candidates and elected mps, Technical report, Statistics Finland.
URL: https://www.stat.fi/til/evaa/2003/evaa20032004 − 05 − 31kat002en.html
Paxton, P. & Kunovich, S. (2003), ‘Women’s political representation: The impor-tance of ideology’, Social Forces 82(1), 87–113.
Peduzzi, P., Concato, J., Kemper, E., Holford, T. R. & Feinstein, A. R. (1996), ‘A simulation study of the number of events per variable in logistic regression analysis’, Journal of clinical epidemiology 49(12), 1373–1379.
Rahat, G. (2007), ‘Candidate selection: The choice before the choice’, Journal of Democracy 18(1), 157–170.
Rahat, G. (2009), ‘Which candidate selection method is the most democratic? 1’, Government and Opposition 44(1), 68–90.
Rahat, G. & Hazan, R. Y. (2001), ‘Candidate selection methods: An analytical framework’, Party politics 7(3), 297–322.
Shomer, Y. (2014), ‘What affects candidate selection processes? a cross-national examination’, Party Politics 20(4), 533–546.
Valdini, M. E. (2012), ‘A deterrent to diversity: The conditional effect of electoral rules on the nomination of women candidates’, Electoral Studies 31(4), 740–749.
Vandeleene, A. (2014), ‘Gender quotas and ‘women-friendly’candidate selection: Ev-idence from belgium’, Representation 50(3), 337–349.
Vandeleene, A. (2017), Do candidate selection modes matter for the gender diversity within political elites? evidence from the belgian case in 2014, in ‘ECPR Joint Sessions of Workshops’.
Vandeleene, A., De Winter, L., Meulewaeter, C. & Baudewyns, P. (2013), The impact of candidate selection on mass elite ideological congruence: The case of belgium, in ‘7th ECPR General Conference’, pp. 4–7.
Willems, M. (2012), ‘Nu definitief: Sgp mag vrouwen niet langer uitsluiten van kieslijst’, NRC .
URL:
5
App
endices
5.1 App endix A: T able of Included Cases Region P art y Ideology Y ear Coun try F amily name Cen t. Prop. F. W estminster New Demo cratic P art y Left 2000 Canada So cial demo cracy 0 0,2953 Lib eral P art y of Canada Cen ter 2000 Canada Lib eral 1 0,2159 Progressiv e Con ser v ativ e P art y Righ t 2000 Canada Conserv ativ e 0 0,1340 Conserv ativ e P art y Righ t 2001 Great Britain Conserv ativ e 1 0,1431 Lib eral Demo cratic P art y Cen ter 2001 Great Britain Lib eral 0 0,2175 Lab our P art y Left 2001 Great Britain So cial demo cracy 1 0,2328 Fianna F ail Righ t 2002 Ireland Conserv ativ e 3 0,1226 Lab our P art y Left 2002 Ireland So ci al demo cr acy 1 0,2391 Progressiv e Demo crats Cen ter 2002 Ireland Lib eral 3 0,3000 Fine Gael Righ t 2002 Ir eland Christian demo c racy 3 0, 1786 Green P art y Left 2002 Ireland Gre en/Ecologist 1 0, 2903 Nordic Radical P ar ty Cen ter 2001 De n mar k Lib e ral 0 0,3469 Conserv ativ e P eople’s P art y Righ t 2001 Denmark Conserv ativ e 0 0,2647Region P art y Ideology Y ear Coun try F amily name Cen t. Prop. F. Christian P eople’s P art y Righ t 2001 Denmark Christian demo cracy 0 0,2500 Lib eral P art y Cen ter 2001 Denmark Lib eral 0 0,2917 So cial Demo cratic P art y Left 2001 Denmark So cial demo cracy 1 0,3619 So cialist P eople’s P art y Left 2001 Denmark Green/Ecologist 1 0,3204 Cen tre P art y Cen ter 2003 Finland Agrarian 0 0,4150 So cial Demo cratic P art y Left 2003 Finland So cial demo cracy 0 0,4820 Sw edish P eople’s P art y Cen ter 2003 Finland Lib eral 0 0,4150 National Coalition Righ t 2003 Finland Conserv ativ e 0 0,4910 Left-Wing Alliance Left 2003 Finland Comm unist/So cialist 0 0, 4460 Christian Demo crats Righ t 2003 Finland Christian demo cracy 0 0,4120 Cen tral Christian Demo cratic App eal Righ t 2002 Netherlands Christian demo cracy 1 0,3684 Demo crats 66 Cen ter 2002 Netherlands Lib eral 0 0,3600 Lab our Left 2002 Netherlands So cial d e mo cracy 1 0,4000 Green Left Left 2002 Netherlands Green/Ecologist 0 0,4583 PPFD* Cen ter 2002 Netherlands Lib eral 3 0, 2400 T able 11: * P eople’s P ar ty for F reedom and Demo cracy
