Abstract
Mainly investigating the share of women in national parliaments, the vast cross-national literature on women’s descriptive (numerical) representation frequently overlooks women’s local representation. Yet, local councils are important political arenas. To what extent are women underrepresented there? What are the determinants of the variation of women’s local representation within and across countries? We investigate these questions through a subnational-level study, covering 272 regions in 29 European countries. Using multilevel modeling, we find that regions with high female labor force participation support for leftist rather than radical right parties and high degrees of urbanization tend to elect more women. Our results also indicate that high women’s representation levels at the national level trickle down to the local level.
Keywords
Introduction
While the share of women in legislative bodies around the world has been rising over the past decades, women are still underrepresented in parliaments (Inter-Parliamentary Union, 2013). Driven by the concern that female politicians are best suited to advance women’s substantive representation (Phillips, 1995), a large literature investigates the determinants for this underrepresentation across countries (Paxton et al., 2007; Wängnerud, 2009). Yet, national parliaments are not the only tier in the political system that matters because decisions taken at the local level also affect citizens. In fact, local councils, defined here as local deliberative assemblies constituted by councilors elected by direct universal suffrage, constitute ‘a crucial element in local representative democracy, linking ordinary citizens to local decision makers’ (Egner et al., 2013: 12). In this article, we build on the growing field of studies examining the gendered dimensions of such councils. Adopting a fine-grained regional analysis, we make two contributions to the literature on women in politics. (1) Our research is one of the pioneer studies in determining women’s representation at the local level across a large number of regions and countries. (2) It also allows us to determine the influence of structural/cultural factors on women’s representation, such as population density or support patterns for leftist and radical right-wing parties, which are commonly left out of national-level studies because the level of analysis is too aggregated.
To tackle these two research goals, we developed an original data set that includes information on the share of women in local councils across Europe, as well as relevant regional-level and national-level covariates across 272 regions from 29 countries. Drawing on the general women’s representation literature and, more specifically, the case-study-based literature on gendered patterns in local councils, we test the influence of five regional-level covariates (female labor force participation, population density, electoral strength of leftist parties, electoral strength of radical right-wing parties, and development) as well as six national covariates (voluntary party quotas, legislative quotas at the subnational level, a proportional representation (PR) electoral system, the years of female suffrage, federalism, and women’s representation on the national level). Using multilevel modeling, we mainly find that the three regional-level variables, women’s labor force participation, population density, and voting patterns for left- and radical right-wing parties, explain variations in the regionally aggregated share of female local councilors. Pertaining to national-level factors, our results also indicate that higher levels of women’s representation at the national level tend to trickle down to the local level.
The article is structured as follows. The next section reviews the literature on women’s representation at the local level, followed by the hypotheses and variables that inform our study. In the subsequent section, detailing design and methods, we explain our comparative approach and analytical strategy to test these postulations. We then report the results and situate them within the literature before concluding and discussing important themes for future comparative research.
Literature
There is a large body of research on women’s political representation at the national level. These studies propose numerous explanations for the variation in the share of elected women across countries, including aspects of parties’ organization (Fox and Lawless, 2010; Kenny, 2013; Lovenduski and Norris, 1993), institutional features such as quotas or the electoral system type (Dahlerup, 2006; Kittilson, 2006; Krook, 2009), and social economic variables such as economic development or the participation of women in the workforce (Inglehart and Norris, 2003; Iversen and Rosenbluth, 2008; Kenworthy and Malami, 1999). By studying variation in the percentage of women elected as local councilors, we switch the analysis from the national to the subnational level. This switch of analysis has two advantages, one empirical and one theoretical/methodological.
Empirically, the local level is an important level of analysis. Not only do local councils often decide upon infrastructure projects, allocations of resources to social institutions such as kindergartens and schools, and municipal taxes, but local-level decisions are also relevant for policy areas related to women’s collective interests (cf Rigon and Tanzi, 2012; Smith, 2014). For instance, focusing on Norwegian municipalities, Bratton and Ray (2002) find that local councils with a larger share of female members tend to be better at providing childcare. Similarly, Chattopadhyay and Duflo (2004) report that villages in India that were randomly assigned a quota for a reserved share of women in leadership have better provisions of public goods such as health and sanitation services than councils without quotas. However, despite the importance of local councils and the fact that the majority of elected politicians in the world are active at the city or municipality level, women’s local representation in a comparative perspective across countries receives relatively little attention.
In fact, the literature on women’s local-level representation largely exhibits single-country studies. The vast majority of these studies focus on the United States. Early works in this vein of research explain women’s representation at the local level by the desirability and importance of office (Karnig and Welch, 1979), the competition for these posts (Welch and Karnig, 1979), the role of media coverage (MacManus, 1992), and the size of councils and cities (Bullock and MacManus, 1991). More recently, Trounstine and Valdini (2008) find that a plurality system increases the presence of white female councilors but not the proportion of minority women, whereas Smith et al. (2012) indicate that the ideological climate of a city is a strong predictor of the share of elected women.
Other case studies, which are mainly descriptive in nature, cover Australia (Ryan et al., 2005), Canada (Tolley, 2011), Chile (Hinojosa, 2009), France (Bird, 2003), India (Bhavnani, 2009), Poland (Rincker, 2009), Sweden (Kokkonen and Wängnerud, 2015), and the United Kingdom (e.g. Bochel and Bochel, 2008; Borisyuk et al., 2007; Rao, 2005; Yule, 2000). These single-country studies mainly focus on the dynamics of party recruitment (Bhavnani, 2009; Bird, 2003; Hinojosa, 2009; Yule, 2000), changes in the share of elected women across time (Borisyuk et al., 2007; Ryan et al., 2005), and the social interplay and tensions between female and male councilors (Bochel and Bochel, 2008; Kokkonen and Wängnerud, 2015; Rao, 2005; Rincker, 2009).
There are also a handful of comparative studies on women’s local representation. For instance, Welch and Studlar (1990) as well as Briggs (2000) discuss the personal motivations, expectations, and obstacles of women entering local political posts in North America and the United Kingdom. Focusing on South East Asia, Drage (2001) suggests that campaigns by non-governmental organizations (NGOs) and the presence of international conferences may provide momentum for women to gain local influence. Similarly, Rai et al. (2006) confirm that quota provisions or reserved seats for women are important features for increasing the local presence of women in India, Pakistan, and Bangladesh. Finally, Pini and McDonald (2011) examine women’s representation in councils in, among others, Australia, England, and Germany, illustrating several obstacles women running for offices face, such as the persistence of ‘old boys’ networks’. Pini and McDonald’s volume also points to factors that may increase the local presence of women, such as the use of gender quotas, and the positive influence of norm change that can be initiated both from ‘below’ through grassroots campaigns and ‘above’, through the work of state feminism and having role models in higher leadership positions.
In sum, these case studies provide illustrative insights into a range of factors, which might influence women’s presence at the local level, but none of these analyses systematically compares the determinants of women’s local representation across a larger number of countries. Vengroff et al. (2003) come closest to this goal. However, the authors focus on state and provincial assemblies rather than locally elected bodies. Also, the authors average the share of elected women (to meso-level assemblies) for the country as a whole. Therefore, a country has only one mean figure for women’s representation in the data set. In doing so, the study (Vengroff et al., 2003) cannot account for much of the intra-national variance in the share of elected women. Our study aims to explain this subnational variation in women elected to local councils across Europe.
Highlighting the (subnational) factors that explain variation in the percentage of local councilors also contributes to our understanding of women’s representation, more generally. According to Smith et al. (2012), ‘gauging the macro-level determinants of women’s descriptive representation at the local level, therefore, may reveal a great deal about the gendered electoral dynamics that continue to limit the numbers of women in public office at all levels’ (p. 325). Following this call, we believe a study of the subnational level across a range of countries has much to offer. We see three advantages in this approach: First, our data show that regional variation better illustrates women’s local representation than between-country differences (i.e. the standard deviation around the mean is larger for the variable women’s representation at the regional level as compared to the national level). In addition, there are vast differences in women’s representation within countries in several states. For example, women’s local representation in the regions of Greece fluctuates between 10 and 24 percent. For some independent variables, including population density and left- and radical right-wing support, regional differences are even more pronounced.
Second, it would be difficult to examine some of the variables we investigate here at the national level. For instance, countries as the unit of analysis are too aggregated to test whether more urban areas (which are supposed to be more cosmopolitan and more encouraging for women to engage in a political career) have more elected women than rural areas (which are supposedly more traditional and patriarchal). Similarly, many regions (e.g. Carinthia in Austria, Bavaria in Germany, or Lombadia in Italy) have distinct support patterns, be it for parties to the extreme right or the left of the political spectrum. These support patterns, which are not representative of the national average for any of these parties, might have a distinct influence on women’s representation. Relatedly, even for variables that differ less within regions (e.g. the gross domestic product (GDP) per capita), a regional-level analysis is still more fine grained than a national-level analysis because it can capture more of the variation in independent variables. The third argument is methodological. Adopting a regional focus not only increases the variation in the data, it also tremendously increases the degrees of freedom. Instead of comparing women’s local representation across 29 countries, we can evaluate women’s representation across more than 270 regions.
So why has such a subnational approach not been conducted? The main reason is data unavailability. Dolan (2008) pinpoints this dilemma when she writes that while the most variance is in the numerical representation of minorities at the subnational level, this is where research currently lacks in data availability (p. 112). We overcome this data problem by compiling a unique data set on the percentage of women councilors for 272 regions in 29 countries (Sundström, 2013) and relevant covariates.
Determinants of women’s local representation: theoretical expectations
Subnational-level covariates
Hypothesis 1. As more women participate in the labor force of a geographical region, the share of locally elected women will increase.
The literature on women in politics advances the argument that women’s role in the labor market is a good indication of gender relations in society (e.g. Iversen and Rosenbluth, 2008; Paxton, 1997). In the words of Matland (1998), the more women advance in the workforce, the more traditional gender roles, which see women as caregivers and housewives, become obsolete. This should have direct influence on women’s representation in elected positions; it should increase both demand for stronger inclusion of women in positions of political power and the supply of capable candidates who are willing to run (see Kenworthy and Malami, 1999).
Besides a number of national-level studies that find this expected link empirically (e.g. Rosenbluth et al., 2006; Stockemer and Byrne, 2012), the few subnational-level studies that use such indicators appear to even find a stronger relationship between high levels of women in the workforce and women’s presence in elected office. For example, Studlar and Matland (1994) identify women’s labor force participation in a district as an important factor that affects female representation in Canada. Similarly, Sanbonmatsu (2002) finds for the United States that high women’s workforce participation at the state level increases the number of both democratic and republican female candidates, as well as women’s representation in both parties (see also Nelson, 1991). Following these subnational-level studies, we hypothesize that variation in women’s labor force participation could explain differences in the representation of women at the local level:
Hypothesis 2. The share of locally elected women will increase with the degree of urbanization in geographical units.
Modernization theory (see Norris and Inglehart, 2001) identifies two stages through which modernization happens: First, industrialization includes women in the workforce, raises their levels of education, and brings female enfranchisement and initial participation into the parliamentary arena. Second, postindustrialization, or a switch toward service sector jobs, promotes egalitarian, participatory, and self-expression values (Inglehart, 1990; Sugarman and Straus, 1988). We expect this postmaterialist shift to occur more in urban areas than rural ones because postindustrial features and postmaterialist values tend to develop in regions that are more cosmopolitan (e.g. Kaldor, 1970). The case study–based literature on women in local politics lends support for this reasoning. For example, Conroy (2011) illustrates that urban areas in Australia had about twice the share of women in 2008 as compared to rural ones. Borisyuk et al. (2007), Holli (2011), and Tolley (2011) find similar patterns for the United Kingdom, Finland, and Canada, respectively. Since city dwellers in our sample live in a multicultural environment, have access to a multitude of information sources, and confront new and modern forms of living, we also expect them to be more likely than citizens in rural areas to hold attitudes promoting equality between the sexes (Matland, 1993):
Hypothesis 3. The share of locally elected women will be higher, the greater the economic development per region.
As the third regional-level covariate, we include economic development into our analysis. The literature concurs that as societies become richer, they create favorable conditions for a higher presence of elected women (e.g. Inglehart and Welzel, 2005; Paxton, 1997). For example, Burns et al. (2001) affirm that increased material affluence leads to a more educated citizenry, which, in turn, should favor the emancipation of women. In another study, Norris and Inglehart (2001) suggest that improved economic development boosts women’s educational and professional status, which then increase the size of the eligible pool of candidates for public office. The one subnational-level study by Dubrow (2006) on Romania suggests that this relationship may also be present at the local level. We therefore believe that differences in development across regions, which can sometimes be substantial (e.g. in Austria the GDP per capita in the Vienna region is double the GDP per capita in the Burgenland region), could also explain variation in the share of locally elected women:
Hypothesis 4. The share of locally elected women will be higher in regions where the electoral support for leftist parties is high.
The literature conveys that areas with a strong presence of leftist parties will elect a larger share of women than regions where leftist parties are weak. Several considerations support this postulation: (1) Socialist, social democratic, and green parties are generally receptive to groups traditionally excluded from the circles of power, including women (Htun, 2005; Matland and Studlar, 1996). (2) Parties to the left show higher representational responsiveness (involving a higher share of recruited women to seats of power) and policy responsiveness (which entails giving attention to issues of concern to women) (Caul, 1999; Krook and Childs, 2010). (3) Such parties frequently have strong internal women’s organizations (Jenson, 1995; Kittilson, 2006). The case study literature on women in local politics confirms that a leftist party ideology might influence gendered patterns at the local level. For instance, Yule’s (2000) study on women in British local councils concludes that the Labour Party is more supportive toward women’s upward mobility than conservative counterparts, where ‘hostility to feminism predominates’ (p. 52). 1 Also Rincker’s (2009) study of Polish local institutions supports this assertion:
Hypothesis 5. The share of locally elected women will be lower, the higher the electoral support for the radical right.
We also believe that strong local support of radical right-wing parties should impede the presence of women in local councils. Not only are many features that characterize the radical right (e.g. xenophobia, populism, nationalism) rather reactionary, but some of these parties still believe that men and women are inherently different and should occupy different social positions (Fennema, 1997). To highlight, some radical right-wing parties implicitly or explicitly advance the argument that men are natural leaders and belong in the public sphere (Hainsworth, 2008). Based on these considerations, we maintain that the electoral strength of the radical right might be a good proxy variable for a very traditional political culture. Given that the strength of the radical right varies tremendously between regions within the same country (e.g. in Belgium or Italy, where it fluctuates by several hundred percent), we believe that regional variation in locally elected women may partly stem from these patterns.
National-level covariates
Hypothesis 6. The share of locally elected women will be higher in regions belonging to a country with a proportional representation electoral system.
Formal institutions, or the ‘rules of the game’, should be principal explanations for the variance in the share of locally elected women. The first institutional factor we include in our analysis is the electoral system type (Kittilson and Schwindt-Bayer, 2012). The literature strongly supports the premise that PR systems tend to generate higher percentages of elected women than plurality/majority systems (see overview in Clark and Caro, 2013). Under PR, parties have an incentive to diversify electoral lists by including women for electable positions. In contrast, because of the zero-sum game under plurality, parties in most single-member districts tend to nominate highly educated middle-aged to senior men of the dominant ethnicity as candidates (Henig and Henig, 2001). These theoretical considerations are empirically supported by Vengroff et al.’s (2003) comparative study of subnational variation in the share of elected women at the meso-level, which finds that ‘proportional systems generally provide greater representation of women’ (p. 171):
Hypothesis 7a. The share of locally elected women will be higher in regions belonging to a country with quota provisions at the subnational level.
Hypothesis 7b. The share of locally elected women will be higher in regions belonging to a country with party quota provisions.
The second institutional factor, quotas, which guarantee women a certain share of elected seats through legislation or on party candidate lists through voluntary rules, are generally considered an efficient means to boost women’s representation, regardless of the level they are implemented in (Dahlerup, 2006; Krook, 2009). The first type of quotas, legislative quotas, should boost women’s representation, as they force parties to nominate a certain share of women as candidates (Drage, 2001). The second type, party quotas, ought to not only increase the presence of women for the party in question, but they might also trigger a reaction by other party – that is, in systems where one party adopts quotas, other parties might feel pressure to also nominate more women on lists, which in turn may positively impact the share of female candidates/deputies overall (Kittilson, 2006).
There are numerous examples for the efficiency of quota rules at the local level. For example, the Equality Law in Spain guarantees women 40 percent representation in local bodies (Verge, 2012). Similarly, the share of women in municipal councils in Finland rose from 35 to 47 percent after the implementation of legislative quotas in the 1990s (Holli, 2011). Finally, focusing on another part of the world, South East Asia, Rai et al. (2006) report that quota reservations have been an efficient means to increase the local presence of women in the region:
Hypothesis 8. The share of locally elected women will be higher in regions belonging to a federal state.
The third national-level factor is federalism. We hypothesize that federations elect higher shares of women to local offices than central states. First, federations provide several entry points for women to access positions of power. Second, a candidacy with decisional power at a lower level can provide women a fruitful training ground for a run at a higher level. Third, federalism makes it easier for social movements (including the women’s movement) to lobby at various levels of government: In other words, there are multiple avenues for promoting their interests (Vickers, 2011, 2012). Fourth, federal states are more prone than centralist states to advance women’s issues (Stockemer and Tremblay, 2015). For the purpose of this study, we hypothesize that the same relationship should apply at the local level:
Hypothesis 9. The share of locally elected women will be higher, the larger the proportion of women at the national level.
We expect that a high presence of women at the national level may benefit the local entrance of women into the political sphere. Indirectly, a large presence of elected women in influential positions may constitute a role model for women in junior positions to follow (Campbell and Wolbrecht, 2006; Fox and Lawless, 2010; Pini and McDonald, 2011; Wolbrecht and Campbell, 2007). More directly, a high presence of women at the national level may also reduce gender bias and discrimination in recruitment for locally elected seats (Bjarnegård, 2013). A recent study by Barnes and Burchard (2013), focusing on sub-Saharan Africa, empirically supports this reasoning. Their analysis reveals that the incorporation of women into positions of political power at a higher level also encourages the political engagement of women at a lower level over time:
Hypothesis 10. The longer period since women gained suffrage, the higher the share of locally elected women.
The representation of women in a country may be affected by the number of years that men and women have experienced equal political representation, as such countries have had ‘more time to incorporate demands of women’s inclusion in parliament’ (Ruedin, 2012: 98). Many studies include this reasoning in their models attempting to explain the variance in women’s descriptive representation cross-nationally, generally measured in the number of years since the introduction of women’s suffrage (Fallon et al., 2012; Kenworthy and Malami, 1999). We assume that the same reason ought to apply sub-nationally and therefore hypothesize that the years of universal suffrage also hold at the local level.
Research design
We test our 10 hypotheses using data that focus on the share of female local councilors, aggregated by region. We define a region according to the system of the Nomenclature of Territorial Units for Statistics (NUTS) (see Eurostat, 2014) and analyze regions that range from NUTS 1 to NUTS 2. 2 In more detail, we employ the first official geographical unit below the national level. 3 While we acknowledge that the subnational units by which we have aggregated the data differ in size, we do not see this as a potential problem. Analyzing societal features of both NUTS 1 and NUTS 2 regions has a long tradition in the social sciences, which we support and follow in this article (Bottazzi and Giovanni, 2003; Brulhart and Traeger, 2005; Charron and Lapuente, 2013; Marks et al., 1996). Despite the fact that the geographical units are still large, we also think that the comparison of local governments across regions provides more on-the-ground results than the respective national-level analysis would (Lidström, 1999; Snyder, 2001).
Our dependent variable is the average percentage of female local councilors per region. Following the definition advanced in the introduction, a local councilor is an individual elected to local deliberative assemblies constituted of councilors elected by direct universal suffrage. We have collected this mean share of female councilors for 272 regions across 29 European countries and constructed the data set from a range of sources (see Table 1, Appendix 1, and Sundström, 2013). 4 For all but two of the 29 countries analyzed, we could aggregate the share of local female councilors by region. However, finding comparable data for Romania and Slovakia was more difficult. For Romania, data on local councilors are not available to researchers. Consequently, we use figures on their intermediate-level councils. In Slovakia, national authorities do not collect data on the gender composition of local councils. Instead of excluding the country from the analysis, we include figures on the sex ratio of mayors.
Local councils in 29 countries.
Some countries do not hold simultaneous local elections across all regions. Therefore, the table reports data across several years for these countries. In Austria, local elections occur at different times in the Bundesländer. In Finland, Åland has a special electoral cycle. This is also the case for Scotland in the United Kingdom.
Table 1 reports details about the local councils of each country. Due to the fact that the number of municipalities varies greatly in the 29 countries under investigation, there is variance in the number of local councils that exist in a country; for instance, while the United Kingdom has 466 ‘Local Authorities’, Latvia has 110 different ‘Dome’. We are well aware that the mandates and responsibilities of these councilors differ across countries. Therefore, we control for the four major historical traditions of these councils, which should also account for differences in their power (Sellers and Lidström, 2007; Smith et al., 2012).
To capture these various council types, we follow the typology of Bennett (1993). In more detail, this typology classifies these councils according to their organization by outlining the scope of their responsibility, their power in relation to municipal administration, and the influence of mayors. These three components of the typology are factors that may determine the ‘desirability’ of a seat in such a council, which in turn may have gendered effects. This council typology distinguishes between (1) the Franco-Napoleonic type, (2) the North-Middle European type, (3) the Anglo type, and (4) the Central-East European type (see Egner and Heinelt, 2008; Hesse, 1991). 5
We operationalize our five regional-level covariates, our six national-level institutional factors, and the different council types as follows: To measure women’s labor force participation, we use the ratio of women to men active in the formal labor force for each region (see United Nations Development Programme, 2013, for an equivalent operationalization and Matland, 1998). The data for this indicator are from the Eurostat Regional Statistics as well as national statistical authorities. 6 Second, we gauge a region’s level of urbanization by the population density (i.e. the average number of inhabitants per square kilometer). We collect these data from the Eurostat Regional Statistics. 7 Whenever this information is not available, we use data from national statistical offices. Third, we create an indicator for economic development, which we operationalize by the log GDP per capita per region that we obtained from the Eurostat Regional Statistics. 8
Finally, we employ two proxy variables to determine the ideological composition of a region’s electorate. In more detail, we calculate the regional average vote share in legislative elections between 2007 and 2012 of (1) left-leaning parties (i.e. communist, socialist, social democratic, and green parties) and (2) radical right-wing parties. While ideally we would include the vote share of these parties in elections at the local level, these data are not currently available for researchers in any ordered form. Nevertheless, we maintain that the parties’ regional vote share in national elections adequately captures the political ideology of these regions.
To define these party families, we use the coding from the Party Manifesto (Volkens et al., 2013). While the attribution of parties into the left-wing category is relatively straightforward, it is sometimes more difficult to differentiate a (very) conservative party from a radical right-wing party and to correctly attribute regional parties. While radical right-wing platforms include a mixture of populism, anti-Europeanism, anti-Islamism, nationalism, authoritarianism, and cultural monism, we code a party as belonging to the radical right if it comprises three features: authoritarianism, populism, and the issue ownership of national identity against foreign influences (Ignazi, 2003; Klandermans and Mayer, 2006). Data for the vote share of these parties are from the European Election Database (2013). 9
Our five national-level factors – PR electoral system, gender quotas at the subnational level, federalism, the share of women in national parliaments, and the time since women gained suffrage – are operationalized as follows: To measure a country’s electoral system type, we create a dummy variable that codes PR systems, including mixed-member proportional (MMP) systems, as one and non-PR systems as zero. 10 We base our coding on the established categorization by the International Institute for Democracy and Electoral Assistance (2014). To gauge gender quotas, we use two indicators. First, we code any country that has legal provisions requiring a certain share of women to be elected at the subnational level as one and all other countries zero. Second, we calculate the percentage of parties per country that have implemented quota rules in their statutes using data from the Quota Project (2014). To ensure that these quota provisions were not new (i.e. established after the local election data used here), we cross-referenced the information for each country to make sure that these institutions were in place for several years (a majority of parties and countries installed quotas in the 1990s).
Retrieved from the Forum of Federations, the indicator federalism is a dummy variable that gauges if a country is a federal (coded one) or a centralized state (coded zero). We collect data from the Inter-Parliamentary Union (2013) to measure the share of women in the national parliament (i.e. the lower house, in case they are two chambers). To temporally account for any possible trickle-down effect from the national to the subnational level, we collect the data on the share of women in national parliament from the electoral cycle prior to figures on the respective local elections. The women’s suffrage variable measures the years a country had universal suffrage at the time of the respective election. These data are based on official records. Finally, to capture the four council types, we create three dummy variables, with the Franco-Napoleonic type serving as the reference category (see Bennett, 1993). For summary statistics of the variables included in this study, see Table 2.
Summary statistics of the dependent and independent variables.
GDP: gross domestic product.
Methods
Our statistical analysis proceeds in several steps. First, we present some univariate statistics of our dependent variable to investigate and compare women’s representation at the local and national levels. Second, we present a so-called null model that allows us to determine the amount of variance explained by regional- and national-level covariates. To do so, we use a well-known formula and divide the variance component corresponding to the random intercept by the total variance (see Rabe-Hesketh and Skrondal, 2012). Third, using our independent variables, we construct a multivariate multilevel regression model to explore the determinants of women’s local representation. Fourth, we run the same model excluding women’s representation at the national level. We do this because there is a strong correlation between women’s representation at the local and national levels (the Pearson correlation coefficient is 0.68). We also cannot be certain whether there is an endogeneity problem between the share of women at the local and national levels, despite lagging women’s representation at the national level by one election cycle. Running the model without women’s national representation gives us an idea of whether the findings we get from Model 2 are robust.
Because we are interested in how regional-level and national-level covariates influence our dependent variable, we use hierarchical linear modeling for our three models. This technique explicitly accounts for the two-level structure of the data. In other words, it allows us to model the subnational indicators and the institutional national context in which the regions in our data set are embedded (Wells and Krieckhaus, 2006). As level 1 factors, we include our five regional covariates. As level 2 factors, we include the six national-level variables, as well as the dummy variables for the various council types. Directly measuring these institutional factors’ influence on the regional average share of female local councilors avoids an overestimation of the regional-level factors’ influence, which in turn guarantees a purer representation of the effect of these level 1 factors (O’Connell and McCoach, 2008).
Results
The first finding that sticks out is that the proportion of women’s representation at the local level is relatively similar to the share of women’s representation at the national level. The respective means from our data are 23.43 percent for national-level representation and 24.96 percent for local-level representation. Second, our univariate statistics indicate that there is wide variation in the dependent variable: The average share of female local councilors in a region ranges from 1.86 to 46.48 percent for the whole data set. In the bottom of this dispersion (the areas with the lowest share of female councilors), we find the regions belonging to Turkey and Croatia. Correspondingly, the subnational units from countries such as Sweden, Spain, and Norway comprise the largest share of women (Appendix 2 lists the average share of elected women for each region).
Turning to the inferential statistics, Model 1 (Table 3) indicates that national-level factors explain around 85 percent of the variance in women’s representation at the local level, and regional-level covariates explain around 15 percent. (As explained in the method’s section, the national variation in women’s representation is calculated as follows: 80.67/(80.67 + 14.26).) Despite the fact that the national context accounts for the majority of the variance, there is still within-country variation that can only be captured by the level 1 covariates. As indicated by Figure 1, the variation in women’s local councilors aggregated at the regional level fluctuates by more than 5 percentage points from the median, on average, and for some countries, including Germany and Austria, by more than 15 percentage points.
Multilevel models measuring women’s representation of local councils aggregated at the regional level.
GDP: gross domestic product.
Standard errors in parentheses.
p < 0.1; **p < 0.05; ***p < 0.01.

The average share of female local councilors in the regions of 29 European countries (percentage).
Models 2 and 3 (Table 3) illustrate that level 1 factors capture some of this regional variation in the percentage of female councilors. We find that four of our five regional covariates have the hypothesized impact. These four covariates, high levels of female labor force participation, a high support base for leftist parties, weak support for radical right-wing parties, and a high degree of urbanization positively impact the representation of women at the local level. The influence of all these covariates is moderate. For example, holding everything else constant, the model predicts that for every 1000 inhabitants more per square kilometer, women’s representation increases by 1.8 points. In other words, a very urban area with a population density of 3000 is expected to have more than 5 percentage points more female deputies than a rural area with a population density of 100 or 200.
Similarly, the variable women’s labor force participation indicates that for every 0.12 points that women close the gap in their contribution to GDP compared to men, their representation increases by nearly one point. Given that the women’s participation rate ranges from 0.24 to 1.06 in the regions of the data set, Models 2 and 3 predict that regions where men and women contribute roughly equally to the economy have more than 6 percentage points more female councilors than regions where women, as compared to men, only contribute one-fourth to the region’s economic performance. The two party variables – support for the left and the radical right – also have a moderate to weak influence. For example, the models predict that for every 13 points of votes for the left, the percentage of female councilors increases by one percentage point. Finally, the models suggest that for every 10 percentage points the radical right wins more votes, women’s representation declines by one percentage point. Numerous empirical examples support these patterns. For example, in Austria, the Vienna region is the most populated, with higher-than-average support for the left and the second highest ratio of women to men active in the labor force – the region with the highest percentage of local female councilors in the country. However, we do not find support for the proposition that women’s local representation increases with the economic development of a region. This relationship is absent when examining cross-regional variation in the share of female councilors.
Pertaining to the national-level factors, there is surprisingly only one indicator that has an, albeit very strong, effect on women’s local representation. As expected, the national context influences women’s representation at the local level: that is, high percentages of women in the national legislature trickle down to the local level. In fact, the model predicts a very strong relationship. According to Model 2, a one-point increase in the share of women at the national level coincides with a 0.9 point increase at the local level. Contrary to our expectations, there seem to be no effects on the local presence of elected women from having a PR election system, gender quotas, from being a federal state, or the time passed since women gained suffrage. Finally, our results indicate that the historical tradition of the council has no influence on women’s local representation.
Importantly, the coefficients of the regional-level indicators are robust whether or not we remove the variable women’s representation at the national level (see Model 3). The same applies to a range of alternative specifications. For example, we get the same findings if we exclude Romania and Slovakia or if we include a dummy variable for Western Europe instead of the dummies for the different council types. Also, when we alter some of the level 2 covariates, our results remain robust. For instance, if we code the dummy variable of electoral systems differently – that is, setting MMP systems as a single category – no substantial change takes place in the model. This also applies for our variable party quotas. For example, when we instead construct a dichotomous measure that gauges whether or not the biggest party in the country has a gender quota for the subnational level, the results remain identical.
The theoretical implications from these results are twofold. First, we confirm that women’s representation at one level (the national level) influences women’s representation at another level (the local level). For advocacy, this implies that if women succeed in increasing their presence at one level, they will most likely see an increase at another. Second, our study indicates that between regions within the same country, there is some variation in women’s representation. Regional factors such as population density and support for left-wing parties can partially account for this variation. We further assume their effect would increase if we were to engage in an even more fine-grained analysis on the city or municipal level.
Conclusion
This article makes two contributions to the literature on women’s representation. For one, we find that women’s representation levels at the local level are not independent of their national context (i.e. if women’s representation is high nationally, it also tends to be comparatively high at the local level). However, we also discover that despite this trickle-down effect, there is subnational variation. At least partially, this subnational variation can be explained by variation in levels of urbanization, women’s workforce participation, and the dominant political ideology. The second and related contribution is more methodological and pertains to the conceptualization of women’s representation. Perhaps we should stop thinking of countries as homogenous entities. Within countries, there are large differences in the degree of urbanization, party support patterns, wealth, and how much women contribute to the economy. These differences are meaningful and explain variation in the percentage of women representatives at the local level.
More broadly, this study is a plea for more research at the subnational level. Our research shows that structural and cultural processes, such as modernization, are complex phenomena that occur at different paces within one country. Only regional- and local-level studies can account for these complexities. This tendency should apply even more for countries outside of Europe. For example, in Asian or Latin American countries, such as China or Brazil, the differences in values and beliefs between the rural and urban centers might be larger than differences within countries in Europe. For instance, in China’s hubs like Beijing or Shanghai, the dominant values of the population are cosmopolitan, free-market oriented, and consumerist. In contrast, inhabitants in rural areas might lack most, if not all, features of a modern society such as TVs and the Internet. These differences in living standards and development might impact individuals’ goals and values, which in turn might influence women’s roles in politics and society.
Footnotes
Appendix
The share of female local councilors in 272 European regions (NUTS codes and average share of elected women for each region).
| SE32 | 46.49 | BE2 | 36.14 | DEF | 29.5 | NL13 | 25.43 | NL42 | 20.82 | ITE4 | 12.98 |
| SE12 | 44.66 | FR24 | 36 | IE022 | 29.43 | DE2 | 25.2 | AT34 | 20.7 | RO21 | 12.62 |
| SE33 | 44.63 | FR53 | 36 | HU21 | 29.42 | PL22 | 24.95 | SI01 | 20.65 | IE025 | 12.33 |
| SE11 | 44.01 | FR81 | 36 | HU31 | 29.41 | DE8 | 24.8 | SK04 | 20.53 | RO42 | 12.21 |
| NO01 | 43.83 | FI1A | 35.8 | CZ02 | 29.4 | PT11 | 24.73 | NL34 | 20.51 | GR23 | 12.1 |
| SE31 | 43.82 | EE007 | 35.61 | PL42 | 29.17 | LT006 | 24.36 | PL31 | 20.44 | CH07 | 11.89 |
| ES30 | 42.5 | AT13 | 35 | FR21 | 29 | GR30 | 24.34 | ITC1 | 20.33 | TR10 | 11.83 |
| ES21 | 42.4 | FR61 | 35 | FR43 | 29 | BG32 | 24.33 | ITC4 | 20.33 | ITG1 | 11.57 |
| SE23 | 42.15 | BE3 | 34.94 | FR83 | 29 | UKM | 24.04 | PL34 | 20.32 | GR14 | 11.41 |
| FR10 | 42 | PT17 | 34.36 | NL32 | 28.82 | NL31 | 24.04 | AT33 | 20.2 | TR31 | 11.38 |
| DE3 | 41.9 | FI19 | 34 | DEA | 28.8 | NL33 | 24.00 | DEC | 20 | GR21 | 11.24 |
| ES43 | 41.9 | FR23 | 34 | HU33 | 28.79 | CZ04 | 23.98 | ITG2 | 19.8 | ITF4 | 10.9 |
| ES23 | 41.6 | FR30 | 34 | PL62 | 28.66 | CH01 | 23.98 | ITD4 | 19.77 | RO32 | 10.71 |
| ES42 | 41.5 | FR63 | 34 | BG34 | 28.57 | LT005 | 23.91 | LT003 | 19.72 | GR25 | 10.59 |
| BE1 | 41.46 | NO04 | 33.85 | NL12 | 28.51 | CZ03 | 23.65 | DEE | 19.3 | RO12 | 10.41 |
| ES61 | 41.2 | UKE | 33.8 | DK03 | 28.5 | SK03 | 23.58 | AT11 | 19.26 | GR11 | 10 |
| ES62 | 40.9 | LV009 | 33.75 | UKG | 28.1 | UKN | 23.54 | DED | 19.2 | TRC2 | 9.76 |
| NO02 | 40.82 | FR26 | 33 | EE006 | 27.94 | CH06 | 23.54 | SK01 | 19.1 | ITF3 | 9.06 |
| ES53 | 40.7 | FR72 | 33 | HU32 | 27.93 | UKL | 23.50 | AT31 | 18.92 | TR51 | 8.32 |
| SE21 | 40.67 | DE7 | 33 | PT16 | 27.70 | NL22 | 23.48 | DE1 | 18.7 | TRC3 | 8.07 |
| ES52 | 40.5 | PT15 | 32.89 | HU22 | 27.68 | ITD5 | 23.46 | ITC3 | 18.67 | TR22 | 7.07 |
| ES22 | 40.4 | PL61 | 32.83 | LT002 | 27.61 | SI02 | 23.40 | RO31 | 18.46 | TR42 | 7.06 |
| ES70 | 39.9 | LV007 | 32.81 | LT00A | 27.57 | DEG | 23.3 | GR42 | 17.54 | TR21 | 6.57 |
| ES41 | 39.8 | DE6 | 32.7 | PL43 | 27.48 | CH03 | 23.30 | PL32 | 17.54 | TR41 | 6.52 |
| SE22 | 39.38 | LV003 | 32.22 | CZ05 | 27.46 | NL23 | 23.27 | ITD3 | 17 | TR81 | 6.05 |
| IS002 | 39.36 | LV008 | 32.21 | CZ08 | 27.22 | CZ06 | 23.24 | GR12 | 16.87 | TR62 | 5.90 |
| NO06 | 39.15 | FR25 | 32 | CZ07 | 26.64 | CH04 | 23.24 | RO22 | 16.83 | TRB2 | 5.58 |
| FR82 | 39 | FR62 | 32 | DEB | 26.5 | LT009 | 22.9 | ITE2 | 16.68 | TR90 | 4.98 |
| DE5 | 38.8 | PT20 | 31.66 | NL11 | 26.38 | SK02 | 22.84 | IE021 | 15.73 | TRB1 | 4.90 |
| NO07 | 38.73 | UKC | 31.4 | EE001 | 26.35 | LT004 | 22.68 | GR41 | 15.72 | TRA2 | 4.89 |
| DK01 | 38.6 | DK04 | 31.2 | NL21 | 26.34 | DE9 | 22.60 | GR13 | 15.7 | TR32 | 4.79 |
| ES13 | 38.6 | FR22 | 31 | AT12 | 26.22 | NL41 | 22.51 | IE023 | 15.67 | TR82 | 4.61 |
| NO03 | 38.32 | FR42 | 31 | AT22 | 26.2 | CH02 | 22.43 | RO41 | 15.18 | TR83 | 3.47 |
| ES12 | 38.2 | UKD | 30.8 | LT001 | 26.19 | CZ01 | 22.22 | RO11 | 14.64 | TR52 | 3.19 |
| FI20 | 38 | PT30 | 30.79 | PL11 | 26.18 | PL41 | 22.12 | IE012 | 14.58 | TRA1 | 2.78 |
| FI13 | 37.9 | PT18 | 30.77 | LT008 | 26.08 | ITC2 | 22.09 | GR24 | 14.4 | TR61 | 2.66 |
| ES11 | 37.9 | UKK | 30.7 | DK05 | 26 | BG42 | 22.07 | IE011 | 14.27 | TR33 | 2.32 |
| ES24 | 37.9 | UKH | 30.6 | DE4 | 26 | PL21 | 21.83 | GR43 | 14.02 | TRC1 | 2.11 |
| FI18 | 37.8 | DK02 | 30.5 | PL12 | 25.87 | CH05 | 21.71 | GR22 | 13.81 | TR72 | 1.95 |
| NO05 | 37.57 | HU23 | 30.47 | PL63 | 25.83 | ITE1 | 21.71 | HR03 | 13.74 | TR71 | 1.90 |
| IS001 | 37.14 | LV005 | 30.47 | HU1 | 25.79 | LV006 | 21.67 | ITF1 | 13.69 | TR63 | 1.87 |
| ES51 | 37.1 | EE004 | 30.44 | BG41 | 25.74 | AT21 | 21.59 | IE013 | 13.5 | ||
| FR51 | 37 | FR41 | 30 | BG31 | 25.62 | LT007 | 21.57 | IE024 | 13.4 | ||
| FR52 | 37 | UKJ | 30 | UKF | 25.6 | AT32 | 21.5 | ITF2 | 13.35 | ||
| FR71 | 37 | EE008 | 29.77 | PL52 | 25.57 | PL33 | 21.21 | ITF5 | 13.11 | ||
| UKI | 36.4 | PL51 | 29.73 | BG33 | 25.47 | ITD2 | 21.17 | HR04 | 13.04 |
Each NUTS region has a code: The first two letters refer to the country, followed by a regional denominator. ITF3, for instance, refers to the south Italian region Campania.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
