The gender gap in African political participation:

The gender gap in African political participation: Testing theories of individual and contextual determinants This version: 2013-06-03 Ann-Sofie Isaksson *, Andreas Kotsadam **, Måns Nerman *** * University of Gothenburg ** University of Oslo *** University of Gothenburg Abstract: This paper aims to test whether existing theories of what factors underlie the gender gap in political participation apply in an African context. Empirical estimations drawing on recent data covering over 27,000 respondents across 20 African emerging democracies suggest that whereas several of the investigated factors structural differences in individual resource endowments and employment, and cultural differences based in religious affiliations are found to be important determinants of participation, they explain only a very modest share of the observed gender gaps. Suggestive evidence instead point to the role of clientelism, restricted civil liberties, economic development and gender norms. Keywords: Political participation, Gender gap, Africa, Afrobarometer. JEL classification: D01, D72, J16, O12, O55. Corresponding author: Måns Nerman, Department of Economics, University of Gothenburg, Box 640, 405 30 Gothenburg, Sweden. E-mail: mans.nerman@economics.gu.se, Tel. +46-(0)31-7864720. We wish to thank the editor and two anonymous referees for useful comments. We are also grateful to Arne Bigsten, Niklas Bengtsson, Andy McKay, Måns Söderbom, and seminar participants at the 2011 CSAE conference in Oxford for valuable suggestions.

1 Introduction Political participation tends to be unequally distributed across citizens (Bartels, 2005; Brady et al., 1995; Griffin and Newman, 2005; Isaksson, 2010; Lijphart, 1997; Verba et al., 1995). Thinking of political participation as citizen acts to influence the selection of and/or the actions taken by political representatives, participatory inequalities may affect what policy issues are brought to the agenda (see for example Bartels, 2005; Gilens, 2005; and Griffin and Newman, 2005), potentially reinforcing existing economic and social inequalities. Hence, broad-based political participation is important due to its intrinsic democratic value as well as from an inequality perspective. The present paper investigates the gender gap in African political participation. Can gender inequality in political participation, traditionally implying lower participation among women than among men, be explained by individual observable characteristics, such as women being less educated, or is it attributable to, say, gender variation in participatory norms? Given that gender differences in participation could reproduce gender inequalities in other domains, understanding this participatory inequality is central. Furthermore, considering the millennium development goal to promote gender equality and empower women, the issue is arguably particularly pertinent in the emerging African democracies, where resources are scarce and women often suffer from severe inequalities in important dimensions such as health and education (World Bank, 2011). Against this background, there is surprisingly little research on the determinants of gender differences in African mass political participation. Turning instead to evidence from Western countries, leading explanations of the gender gap in participation focus on structural differences in individual resource endowments, often viewing female employment as the crucial factor (see for example Iversen and Rosenbluth 2008; Ross 2008), and on cultural differences, often with religion as main focus (Norris and Inglehart 2001; Norris 2009). However, while in Western countries the traditional gender gap in political participation is in the process of closing (Inglehart and Norris, 2000; Norris, 2002), the sparse evidence available for developing countries indicates that there are still important gender differences in mass political participation. A number of recent studies exploring the patterns of political participation in Africa note that women tend to vote and participate politically in between elections to a lesser extent than men (Bratton, 1999; Bratton and Logan, 2006; Bratton et al., 2010; Kuenzi and Lambright, 2010; Isaksson, 2010), yet we have little knowledge about to what extent the commonly suggested explanations mentioned above are applicable to Africa. In light of this, our aim is to test whether existing theories of what factors underlie the gender gap in political participation apply in an African context. In particular, we evaluate the 2

explanatory power of three commonly suggested determinants of the gender gap, namely resources, employment and religion, taking account of their contextual as well as their individual variations. Empirical findings drawing on data covering more than 27,000 respondents from 246 regions in 20 African countries suggest that while several of the individual and contextual characteristics considered are important determinants of general political participation, existing theories explain only a modest share of the gender gap in African participation. Interesting in the sense that it conflicts with common suggestions in previous literature, it turns out that religious affiliations neither at the individual nor the contextual level, seem to increase the gender gap in participation. Finding that the leading explanations of the participatory gender gap in Western countries explain only a limited share of the gender gap in Africa, we briefly explore a number of alternative explanations that may be particularly relevant in an African context. The results suggest that clientelism, restricted civil liberties, economic development and gender norms are potentially important determinants of the participatory gender gap in Africa. To our knowledge, this is the first study focusing exclusively on exploring the factors that underlie the gender gap in African mass political participation, assessing the explanatory power of both individual and contextual determinants of participation. As such, it should contribute to our understanding of a central form of inequality. 2 Understanding the gender gap in political participation In this section we discuss possible determinants of the gender gap in participation implied by the literature on the general determinants of participation and by previous studies specifically addressing gender variation in the same. In particular, we focus on the role of three factors highlighted in the literature: individual resources, labour market participation and religion, taking account both of their possible individual and contextual influences. 2.1 Individual level At the individual level, previous studies of gender variation in political participation have stressed the role of structural inequalities in individual resource endowments and employment, and of cultural differences originating in religious affiliations. The former perspectives focus on the traditional role of women in the family and the labour market, the idea being that gender gaps in other areas of society hinder women s participation in politics. If political participation is costly, and the resources relevant for meeting these costs are differentially available between the genders, this could give rise to gender differences in political participation. However, the conventional finding that citizens with low incomes and little education participate less than their richer and more educated counterparts (Verba and 3

Nie, 1972; Wolfinger and Rosenstone, 1980; Brady et al., 1995, and Verba et al., 1995) does not necessarily apply when studying political participation in developing countries. Studies of political participation in Africa, Asia, and Latin America suggest that whereas education is often positively associated with participation, poor people participate politically no less (if anything, they seem to participate more) than more well-off citizens (Bratton, 1999, 2008; Yadav, 2000; Krishna, 2002; Bratton and Logan, 2006; Booth and Seligson, 2008; Bratton et al., 2010; Kuenzi and Lambright, 2010; Isaksson, 2010). Hence, it is interesting to investigate if individual resource differentials are important determinants of the gender gap in political participation in the African context. Furthermore, the role of education is twofold; while it helps the individual develop the human capital needed to meet the costs of participation, it also affects what people he/she comes in contact with and thus what participatory norms and networks he/she will face (La Due Lake and Huckfeldt, 1998). Hence, in terms of explaining a gender gap in participation, the influence of a gender gap in education is likely to go beyond that of gender variation in human capital. A similar story applies to employment a factor often pointed out as central for female participation. Employment is thought to positively impact the individual resource base relevant for political participation (such as economic standing and human capital acquisition), access to recruitment networks, and motivational factors stimulating engagement (Schlozman et al., 1999; Norris, 2009). Studying political participation in the US, Schlozman et al. (1999) find that women lack these participatory factors relative to men since women are less likely to be employed, work full time, and hold high-level jobs. Women who are full-time homemakers have their traditional gender roles reinforced, the argument goes, and domestic isolation hinders activism since women are cut off from political discussion and networks (Schlozman et al., 1999). Female labour force participation, on the other hand, is argued to make women informed about their interests and more capable of acting on them (Iversen and Rosenbluth, 2008). Through processes of socialization in the work place, leaving home and joining the paid labour force is suggested to affect women s views and identities (Ross, 2008). i The focus on structural inequalities in individual resource endowments and employment has been challenged by a cultural perspective focusing on religious traditions and their impact on attitudes toward gender equality in attempts to explain the relatively low number of women engaged in politics (Norris, 2009). The argument is that religious traditions affect social values, which in turn are crucial for the role of women in politics (Inglehart and Norris, 2003a). Put differently, religion is thought to affect gender-specific participatory norms and thus the motivational factors stimulating engagement. Although this argument has been criticized (see for instance Charrad 2009; Rizzo et al. 2007), it is not uncommon to single out 4

Islam as a religion reinforcing traditional gender norms and thus negatively affecting female participation (Inglehart and Norris, 2003a,b; Blaydes and Linzer 2008). 2.2 Contextual level Thus far we have considered the suggested individual level influences of resources, employment and religion on gender differences in political participation. Taking account of the literature on the effects of social capital and participatory norms, however, it is also reasonable to assume that these factors could have aggregate or contextual effects. Several empirical studies suggest a positive influence of social capital and participatory norms on political participation (La Due Lake and Huckfeldt, 1998; Knack and Kropf, 1998; Krishna, 2002; Norris, 2002; and Gerber et al., 2008). Social capital, often understood as the social networks and norms of reciprocity and trustworthiness that arise from connections among individuals (Putnam, 2000), is described as the glue binding citizens together so as to enable collective action as well as the gear that directs citizens toward political activity (Krishna, 2002). It is suggested that individuals through repeated interactions with the surrounding social network family, friends, colleagues, community members, and so forth learn civic norms that stimulate participation, and that this can constitute a powerful motivation for participation (Knack and Kropf, 1998; La Due Lake and Huckfeldt, 1998). With respect to gender differences in participation, gender-specific participatory norms might vary across regions depending on systematic regional variation in the individual level determinants of participation discussed above. For instance, it has been argued that once a sufficient number of women have entered into the paid labour force, this will stimulate female political participation (Iversen and Rosenbluth, 2008; Ross, 2008; and Schlozman et al., 1999). Chhibber (2002) argues that since both paid employment and political life take place in the public sphere, more women working will also imply a more woman-friendly political sphere. According to Iversen and Rosenbluth (2008), as women enter the labour market, they become part of networks and organizations (such as unions) where they are more likely to be exposed to political discussion and advocacy, which in turn encourages interest and involvement in politics (p. 486). More women entering the labour market is also argued to have political consequences since the increased density of working women increases the likelihood for women s organizations (Ross, 2008). Against this background, it seems reasonable to test whether individual level factors also have aggregate effects; if a sufficient number of women get an education and become involved in paid employment, it should affect the participatory norms applying to women. Similarly, it has been suggested that religious traditions shape attitudes both at the individual and societal levels (Norris and Inglehart, 2001; Norris, 2009). Norris (2009) 5

specifically proposes that both the individual Muslim identity and living in an Islamic society even as, say, a Christian or a non-believer strengthen traditional gender norms. According to this line of reasoning, not only individual religious affiliation but also the composition of religions in society could potentially affect political participation. To sum up, we intend to evaluate the explanatory power of three commonly suggested determinants of gender differences in participation resources, employment and religion taking account both of their individual and contextual variation. 3 Data and empirical setup To investigate the importance of factors possibly underlying gender differences in African political participation, we use recent data from the Afrobarometer survey. The Afrobarometer is a multi-country survey project collecting data on political and economic attitudes and behaviour of African citizens. As such, it provides a unique opportunity to study the gender gap in African political activity in a large multi-country sample. Round 4 of the Afrobarometer, conducted in 2008-2009, covers over 27,000 respondents from 20 African countries Benin, Botswana, Burkina Faso, Cape Verde, Ghana, Kenya, Lesotho, Liberia, Madagascar, Malawi, Mali, Mozambique, Namibia, Nigeria, Senegal, South Africa, Tanzania, Uganda, Zambia, and Zimbabwe. The country samples range from 1,200 to 2,400 respondents and are representative of each country s voting age population. ii 3.1 Dependent variables As dependent variable we consider electoral as well as inter-electoral political participation, that is, voting and political activity taking place between elections. Thinking of political participation as citizen acts to influence the selection of and/or the actions taken by political representatives, it is a multidimensional concept that encompasses a wide and heterogeneous set of activities; on top of voting, citizens can work in election campaigns, engage in the local community, contact political leaders, attend demonstrations, etc. (for further discussion see for instance Verba et al., 1995; and Lijphart, 1997). Since we are interested in political participation in Africa, where political activity often takes place through informal channels (Hirschmann, 1991; Bratton et al., 2005), taking into account both electoral and inter-electoral political participation should be especially important. To capture electoral participation, we create a dummy variable taking the value one if the respondent reports to have voted in the last national election, and zero otherwise. Those who report to have been too young to register to vote are excluded from the analysis. iii To measure inter-electoral participation, we use a dummy for whether the respondent got together with others to raise an issue in the past year. This participation measure has several attractive 6

properties: first, it is rather universal in the sense that it does not require any particular institutional context (as opposed to, for example, having attended a village meeting), making it suitable for country comparisons; second, it is arguably a more active form of participation than voting, thereby broadening the types of political participation that we capture; and third, it is a relatively common activity (compared to, for instance, the alternative measure of having attended a demonstration). 3.2 Explanatory variables On top of the gender dummy, which is our main explanatory variable, and some basic individual controls (age in years, age squared and a dummy for living in a rural area), our selection of independent variables is based on the discussion in Section 2, and thus includes individual level resources, employment and religious affiliation, and contextual (region) level averages of these. Since our sample is limited to 20 countries, there are limited possibilities for analysis of country-level variables such as political systems or income levels. Instead, we control for country level variation using country fixed effects, capturing all variation across countries and effectively leaving that analysis outside the scope of this paper. iv With respect to the individual resource base, we measure the individual s educational attainment using dummies indicating whether the respondent has completed primary school, secondary school, and whether (s)he has attended post-secondary education. To capture economic standing, we construct a poverty index as the first principal component of four questions asking how often, if ever, the respondent s family has gone without enough food, clean water, medicines/medical treatment and fuel. The index is constructed for each country separately, meaning that it has a mean of zero and a standard deviation of one within each country, with higher values indicating that the respondent is poorer (for variable descriptions and summary statistics, see Tables A.1-2 in the online appendix). To proxy for information access, we include a dummy for owning a radio, and for employment we use an indicator variable equal to one if the respondent has paid part- or full-time employment. To capture religious affiliations, we use a set of dummies indicating whether the respondent is an active member of a religious group, with separate dummies for being of Christian, Muslim, or some other faith. Non-members and non-active members serve as the base category. While membership in religious groups may enhance the social capital and network of the respondent, so can membership in other community groups. To control for this, we proxy for the availability of recruitment networks using the share of the other respondents in the respondent s region who are members of a non-religious community group. Next, we want to account for contextual variation in our key variables resources, employment and religion. To do so we aggregate the individual level variables by taking 7

averages at the region level for each respondent excluding his or her own observation. Moreover, considering the suggested wide-reaching effects of women taking part in education and employment (see the discussion in Section 2), implying that gender gaps in these variables may reproduce gender gaps in political participation, we also split the region level education and employment averages by gender. By region we mean the first-order administrative division in a country, in the survey manual denoted region/province (Afrobarometer Network, 2007). In the total 20-country sample there are 246 regions, with an average of 112 observations in each. While the number and size of regional units vary across countries meaning that they are not strictly comparable, they provide the most reasonable unit for sub-national aggregation. v 3.3 Estimation strategy Given our binary dependent variables, we initially run probit regressions on a pooled sample consisting of both men and women, of the form: (1) Pr(y i = 1) = Φ(α c + δd i + H i β + R i γ + X i θ) where y i is our dependent variable, α c are country fixed effects, D i is a dummy for being female, H i is a vector of the individual variables that we test (resources, employment and religion), R i is a vector of the region level resource, employment and religion variables derived as aggregates from the other individuals within the region, and X i is a vector of controls. By inspecting the marginal effect of being female rather than male (at the mean of all other explanatory variables), we assess to what extent the possible gender gap in participation can be explained by differences in the included variables. While the regional context does not vary across men and women, and hence should not affect the size of the gender gap, regional averages are included as a benchmark test of the hypothesized importance of contextual variables for participation. Expecting that there may not only be gender differences in the concerned variables, but also in their effects, in a next step we relax the pooling assumptions of equality of parameters for men and women. For reasons laid out in Ai and Norton (2003), using interaction terms in probit regressions results in marginal effects that are very difficult to interpret, vi and for this reason we estimate the above equation for men and women separately using a linear probability model (that is, an OLS on a binary dependent variable) vii of the form: (2) y i = α c + H i β + R i γ + X i θ + D i (α c F + H i β F + R i γ F + X i β F ) where notations are the same as in equation (1), and an F superscript denotes parameters of interactions with the female dummy. viii 8

4 Results As is evident from Figure 1, the gender gap in political participation varies across countries as well as between the different forms of participation. The gap in electoral participation is smaller than that in inter-electoral participation, and is not present in all surveyed countries. In fact, in six of the countries, the share of women who vote exceeds that of men, although only in Botswana is this reverse gender gap statistically significant. Looking at inter-electoral participation in terms of joining others to raise an issue, on the other hand, with the exception of Namibia participation rates are consistently significantly lower among women, the difference being more than five percentage points in all countries and as high as 24 percentage points in Ghana. Moreover, the two measures of a gender gap in political participation seem to be correlated across countries. For instance, the countries where women report to vote more than men are also among the countries with the smallest gaps in inter-electoral participation. Hence, it seems that the two measures indeed pick up a more general concept of political participation. <<< Figure 1 about here >>> 4.1 Main results Tables 1 and 2 present the results of probit regressions, focusing on electoral and interelectoral participation, respectively. In a naïve estimation, controlling only for country fixed effects, age, and rural settlement (Table 1, Column 1), women are 3.4 percentage points less likely than men to vote. ix For inter-electoral political participation, as measured by whether the respondent got together with others to raise an issue the equivalent estimation (Table 2, Column 1) suggests a larger gender gap; here women have an approximately 12 percentage points lower participation rate than men. << Table 1 about here >>> << Table 2 about here >>> Introducing the individual resource variables in Column 2, we see that in line with previous findings for Africa (see Isaksson, 2010), whereas information access is positively related to voting, education and economic standing are not, seemingly suggesting that a lack of resources in terms of education or money does not constrain participation to any larger extent. For inter-electoral participation, on the other hand, the individual resource variables 9

stand out as important determinants. Education and information access display sizeable positive effects, possibly reflecting that compared to voting this is a more active form of participation thus requiring more in terms of resource inputs. The fact that poverty is positively related to inter-electoral participation (also in line with previous findings, see Isaksson, 2010), can presumably be explained by motivational forces distinct to poorer groups. Most relevant for our purposes though, including the resource variables seemingly helps reduce the size of the observed gender gaps, albeit modestly; the gender gap in interelectoral participation is lowered by approximately two percentage points to about 10 percentage points. Hence, it appears that gender variation in the individual resource base can explain some of the difference between men s and women s participation, but that the lion s share of the gaps remain. Both employment and religious membership may help build social capital and break domestic isolation, exposing individuals to new sets of norms and recruitment networks, and introducing the individual level employment variable (Column 3) and the dummies for religious affiliations (Column 4), both come out positively related to our measures of participation. For both voting and inter-electoral participation, employment implies a relatively modest increase in participation, and religious affiliations a more marked. Being an active religious believer increases the propensity to vote by around 4-8 percentage points. For inter-electoral participation the equivalent figures are as high as 13-18 percentage points, presumably pointing to the importance of co-operation and thus socialization and networks for this form of participation. As it turns out, though, the individual employment and religious affiliation variables do very little to explain the gender gaps in participation. If anything, taking account of individual religious affiliations makes the unexplained gender gap even more pronounced. As of yet, however, we have not explored the effects of these variables at the aggregate level or for men and women separately. Introducing the regional averages of the individual level variables in Column 5, we see that, contrary to the individual estimates, living in a region with a high share of active Christians or Muslims is negatively related to both forms of participation (the differences between Christians and Muslims are not statistically significant in either estimation). Hence, it is interesting to note that religious affiliations seem to have a two-fold effect; while being religiously active Christian, Muslim or of some other faith increases the likelihood that one will participate politically, living in a society where many are active Christians or Muslims seems to have the opposite effect. At the individual level, religious affiliations could as noted bring a social network in turn enabling political participation. The regional share with religious affiliations could (conditional on the respondent s own connection to religious groups) presumably pick up contextual variation in participatory norms. Still, the pooling of 10

men and women in these regressions may obscure important gender differences in the effects of religion, an issue which we will deal with in the next section. Most of the regional resource and employment measures are not significantly related to individual participation. Living in a region with a greater share of people with primary education is, however, associated with a lower probability to vote. This negative association is offset by a positive association between the share of people with secondary education, which has a positive marginal effect of the same magnitude. This non-linearity may indicate less mobilized voting as people get a basic education but that reaching a certain level of education changes the context where the political participation takes place. For instance, at higher education levels, education may spur more political competition, higher motivation to participate and/or better opportunities for policy debates. For inter-electoral participation the pattern is similar but less significant. It is also worth mentioning that the positive correlation between individual political participation and the share of other people in the region engaged in some non-religious community group seems to support the importance of access to recruitment networks. Furthermore, this control variable is as expected more important for the more active form of participation taking place in-between elections than for voting. To sum up the results so far, we can note that the gender gap in political participation is considerably larger for inter-electoral participation than for voting. Arguably, the former here measured in terms of how often the respondent gets together with others to raise an issue constitutes a more active form of political participation. Moreover, it takes place in groups rather than individually, why it is not surprising that having access to a political network seems more important than for voting. Most importantly, whereas several of the included individual and regional explanatory factors stand out as important determinants of participation, as it turns out, they do relatively little to explain the observed gender gaps in electoral and inter-electoral political activity. Gender inequality in participation remains even when accounting for religious affiliations and for women being less educated and less often employed than men. 4.2 Digging deeper Given that accounting for differences in resource endowments, employment levels and religious affiliations that is, leading explanations of the participatory gender gap in Western countries explains only a modest share of the gender-gap in African political participation, we need to dig deeper. We do this in two steps. First, we relax the pooling assumption of equal parameters across men and women and explore whether the key to understanding the gender gap in African political participation lies in that the investigated determinants of 11

participation affect men and women differently. Second, finding that the leading explanations of the participatory gender gap in Western countries largely fail to explain the gender gap in Africa, we briefly explore a number of alternative determinants that may be particularly relevant in an African context. 4.2.1 Gender-specific regressions Thus far we have restricted our analysis to models where the parameters of the commonly suggested explanatory factors have been equal for men and women. However, it might well be that it is not, say, differing resource endowments across men and women that are most important for understanding the gender gap in participation, but rather differences across the two groups in the effects of having these resources. Running separate regressions for men and women allows all parameters to vary across the two sub-samples. Furthermore, there is reason to believe that women s political participation depends on the capabilities of, and interactions with, other women in society, and that the participation of women in other areas of society may help advance women s participation in politics. Hence, we also introduce gender-specific regional averages for education and employment x in our estimations. Table 3 presents the results of gender specific regressions. Columns 1-2 and 4-5 present the parameter estimates for the male and female subsamples for electoral and inter-electoral participation respectively, and Columns 3 and 6 present the parameter differences between the two groups. <<< Table 3 about here >>> As it turns out, we see almost no statistically significant gender differences in the individual resource, employment, and religion parameters for either measure of participation. The only exception is our information access proxy in the inter-electoral participation estimations; while owning a radio has a positive and statistically significant parameter in both sub-samples, it is about twice as large for men as it is for women, perhaps signifying that men to a larger extent utilize this information source to become politically informed. Furthermore, although the parameter difference across the two sub-samples is not statistically significant, it is worth noting that the relation between labour market participation and voting is larger and only statistically significant for women, possibly pointing to the importance of breaking their domestic isolation. Also, considering that the positive associations observed between individual religious affiliations and electoral and interelectoral participation apply to both women and men (if anything, they tend to be stronger for women), these results do not support the idea that religious norms reinforce gender inequality 12

and thus work against female participation. Rather, they point to possible positive effects of religious activity, such as an increased social network. Turning to the parameters of the contextual variables, these too are similar for men and women, suggesting that they are of limited importance for explaining the gender gap. Still, some interesting findings stand out. Since it has been suggested that female political participation is negatively affected by traditional norms in more religious societies, the parameters of the regional shares of respondents active in religious groups are of special interest. Considering that we observe no (individually or jointly) statistically significant difference between the female and male parameters on these variables, our results do not lend any support to this claim. Rather, for both men and women we find the two-fold effect of religion discussed earlier, that is a positive relationship between individual religious affiliation and participation, and a negative relationship between participation and the regional share with religious affiliations. As for the gender-specific regional education averages, it is interesting to note that men s education affects women s participation. In particular, a higher share of men with primary education is, while not significantly related to male participation, associated with a smaller likelihood that women participate politically. For inter-electoral participation, this parameter difference is statistically significant, while at the same time the positive parameter on the share of women with primary education is significantly larger in the female than in the male sample. As men tend to have more education than women in the sample, this implies that a gender gap in primary education is also correlated with one in political participation, both presumably reflecting the same underlying gender norms in society. For secondary education, on the other hand, a higher share of men with secondary education is, while again not significantly related to male participation, associated with a greater likelihood that women vote. Having more people with secondary education is arguably correlated with less traditional gender norms, even if men are the ones being educated. It is also worth noting that we find no support for the claim that the level of education of other women is an important determinant of women s participation. When it comes to gender-specific regional employment, we similarly want to test the suggested importance of women s labour market participation in (re-)shaping gender roles. For voting, the results indicate significant parameter variation across the male and female subsamples; while female voting decreases with male employment and increases with female employment, there are no corresponding employment effects on male voting. Hence, in line with the pattern observed for primary education, a gender gap in employment is seemingly correlated with a gender gap in voting through women s voting behaviour. 13

To sum up, comparing the effects of individual and region level variables on electoral and inter-electoral political participation, the parameters differ surprisingly little between men and women. Some interesting findings do however stand out. In particular, while we find no support for the claim that traditional gender norms in more religious societies can help explain the gender gap in participation, the results seem to indicate that gender gaps in education and employment are negatively related with female participation. 4.2.2 Alternative determinants Finding that the leading explanations of the participatory gender gap in Western countries notably structural differences in individual resource endowments and employment, and cultural differences based in religious affiliations explain only a limited share of the gender gap in African political participation, it is relevant to explore possible alternative explanations. The fact that our sample countries are young and evolving democracies that still struggle with important problems (for a snapshot, see for instance the Freedom House and Polity IV rankings) brings possible alternative determinants of the gender gap to the forefront. We focus specifically on measures intended to capture the perceived prevalence of clientelism and political intimidation in the regions (both from the Afrobarometer), and country economic development (the log of country Gross National Income, GNI, per capita) and gender norms (the OECD s SIGI gender index). Results from pooled sample regressions where we introduce the new variables one at a time along with their interactions with the female dummy are presented in the online appendix. xi The parameters of the interaction terms tell us to what extent the alternative determinants can help explain the gender gaps in participation. African politics is often described as clientelist in the sense that rulers rely on the distribution of material incentives and personal favours in exchange for political support (Wantchekon, 2003; Christensen and Utas, 2008; Lindberg and Morrison, 2008; and Vicente, 2008). Relevant for our purposes, it has been suggested that clientelist promises stimulate political participation (Christensen and Utas, 2008; Vicente, 2008), but also that they tend to have an important gender dimension in that they are often directed to men and are not equally available to women (Wantchekon, 2003; Vicente and Wantchekon 2009). In line with this, our results indicate that a higher perceived prevalence of clientelism xii in the region involves larger gender gaps in electoral and inter-electoral participation. Comparing the countries with the highest and lowest averages in terms of perceived clientelism, the latter is predicted to have an approximately 5.8 percentage points smaller gender gap in voting. Moreover, several of our sample countries have been described as having restricted civil liberties (see for instance Freedom House, 2013). This is likely to affect people s political 14

activity. Reasonably, an individual could abstain from voting due to voter intimidation or as a result of perceiving the election as unfair (Lindberg, 2004b; and Collier and Vicente, 2009). And potentially, political intimidation could have differential effects on male and female participation. Using fear of violence in connection with elections and beliefs on whether you need to be careful when talking about politics as measures of political intimidation, we find that higher levels of perceived intimidation imply larger gender gaps in voting. Comparing the countries with the highest and lowest averages in terms of fearing violence, the latter is predicted to have an approximately 9.6 percentage point smaller gender gap in voting. Interestingly, the political intimidation variables do not to the same extent seem to impact the gender gap in inter-electoral participation, arguably suggesting that the illegitimate practices captured focus primarily on the electoral process. Hence, fear of violence in connection with elections seems a potentially important determinant of gender bias in voting. As discussed in Section 3.2 the limited number of sample countries restricts the possibilities for analysis of country level variables. Nevertheless, due to the failure of commonly suggested individual and regional determinants to explain the gender-gap in African political participation it is interesting to explore the explanatory power of key country level indicators. The role of economic development is interesting in this context; as a country develops, the structural transformation process arguably erodes traditional gender roles and thereby reduces gender disparities (see for example Inglehart and Norris, 2000). In line with this, the interaction effect between the GNI measure and the female dummy is positive and statistically significant, seemingly implying that as countries grow richer, the gender gaps in participation tend to grow smaller. The predicted differences are sizeable; comparing the richest and poorest countries in the sample their predicted gender gaps in voting differ by as much as 8.5 percentage points. Likewise, introducing the OECD s social institutions and gender index (SIGI), measuring formal and informal gender equality norms, predicts a 5.5 percentage points smaller gender gap in voting in the most gender equal country compared to the least equal, indicating that this gender imbalance is part of a wider system of gender inequality in society (for inter-electoral participation the interaction effect is not statistically insignificant). While we cannot draw causal conclusions based on the data at hand, the above estimations provide suggestive evidence that clientelism, restricted civil liberties, economic development and gender norms are potentially important determinants of the participatory gender gap in an African context. Further research is however needed to establish the role of these factors for explaining the gender gap in African political participation. 15

5 Conclusions Several studies document the existence of a gender-gap in African political participation. Yet there is a lack of studies exploring what factors explain this important source of inequality in the African context. Against that background, this paper explored the factors underlying the gender gap in African electoral and inter-electoral political participation, testing the explanatory power of determinants of participatory inequalities suggested in studies from other regions of the world. Specifically, we considered the role of individual resources (arguably relevant for meeting the costs of participating), employment (proposed to affect women s resource endowments, participatory norms, and access to recruitment networks), and religion (suggested to act as the carrier of traditional gender roles). Considering the commonly suggested influence of social capital and participatory norms, we argued that there is a need to go beyond individual determinants of participation and also consider their contextual influences. Hence, we considered the contribution of our key indicators measured both at the individual and the contextual level. Empirical analysis of a recent and comprehensive data material, covering political and economic attitudes and behaviour of over 27,000 respondents across 20 African countries, suggests that while there is a gender gap in both electoral and inter-electoral participation, the gender gap in the latter, that is in political participation taking place in between elections, is considerably larger. Compared to voting, getting together with others to raise an issue our measure of inter-electoral participation takes place in groups rather than individually and arguably constitutes a more active form of political participation. As such, it presumably requires more in terms of resource inputs, motivations, and access to political networks, presumably working to the disadvantage of women. While several of the tested individual and contextual variables stand out as important determinants of general political participation, they do little to explain the gender gap in participation. Some interesting findings stand out, however. In particular, while we find no support for the claim that traditional gender norms in more religious societies can help explain the gender gap in participation, we observe a two-fold effect of religion, with individual religious affiliation being positively related to participation presumably reflecting better access to political networks and living in a society where many are active Christians or Muslims seemingly having a negative effect, possibly capturing contextual variation in participatory norms. Moreover, the results seem to indicate that gender gaps in education and employment are negatively related to female participation, presumably pointing to the impact of community gender norms. 16

Finding that the leading explanations of the participatory gender gap in Western countries resource endowments, employment, and cultural differences based in religious affiliations explain only a very small share of the gender gap in political participation, we moved on to explore possible alternative determinants that are arguably particularly relevant in an African context. Our findings suggest that clientelism, restricted civil liberties, economic development and gender norms are potentially important determinants of the participatory gender gap in Africa. For instance, it seems that the gender gap in voting is heavily influenced by the fear of political violence in connection with elections. Further research is needed to explore the role of these factors in the emerging African democracies. In particular, we could learn from case studies providing in-depth information on the causal mechanisms linking the above determinants to female participation. To address the millennium development goal of promoting gender equality and empowering women, we need to better understand why the political participation of women lags behind that of men in the emerging African democracies. 17

Tables and figures Figure 1. Gender gaps in participation Gaps (%) -10 0 10 20 30 Botswana Cape Verde South Africa Senegal Lesotho Malawi Namibia Benin Mozambique Voting Tanzania Liberia Ghana Zambia Madagascar Uganda Raised issue Mali Kenya Zimbabwe Burkina Faso Nigeria Notes: Gaps calculated as the participation rate of men less the participation rate of women. Countries are ordered by voting gap. 18

Table 1. Pooled sample regressions (probit marginal effects). Dependent variable: voted. (1) (2) (3) (4) (5) Female -0.034*** -0.029*** -0.032*** -0.036*** -0.030*** (0.007) (0.007) (0.007) (0.007) (0.007) Resources Poverty index 0.002 0.003 (0.003) (0.003) Primary 0.006 0.008 (0.008) (0.008) Secondary 0.001-0.003 (0.010) (0.009) Tertiary -0.027-0.021 (0.018) (0.018) Own radio 0.035*** 0.032*** (0.007) (0.007) Employment Employed 0.015** 0.013* (0.007) (0.007) Religion Christian 0.044*** 0.046*** (0.007) (0.007) Muslim 0.055*** 0.054*** (0.011) (0.011) Other religion 0.077*** 0.071*** (0.017) (0.018) Regional averages Poverty index -0.026* (0.014) Own radio -0.015 (0.044) Christian -0.117** (0.047) Muslim -0.154** (0.062) Other religion 0.019 (0.105) Primary -0.189*** (0.050) Secondary 0.150** (0.063) Tertiary -0.157 (0.152) Employed -0.018 (0.043) Community group 0.106* (0.061) Observations 23,646 23,646 23,646 23,646 23,646 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Additional controls: age, age squared, rural dummy, country fixed effects. 19

Table 2. Pooled sample regressions (probit marginal effects). Dependent variable: raised issue. (1) (2) (3) (4) (5) Female -0.123*** -0.102*** -0.117*** -0.133*** -0.110*** (0.008) (0.008) (0.008) (0.008) (0.008) Resources Poverty index 0.028*** 0.025*** (0.006) (0.005) Primary 0.070*** 0.062*** (0.010) (0.010) Secondary 0.036*** 0.029*** (0.011) (0.011) Tertiary 0.097*** 0.091*** (0.019) (0.019) Own radio 0.074*** 0.070*** (0.009) (0.009) Employment Employed 0.049*** 0.031*** (0.010) (0.009) Religion Christian 0.179*** 0.165*** (0.012) (0.012) Muslim 0.152*** 0.169*** (0.020) (0.018) Other religion 0.129*** 0.123*** (0.036) (0.038) Regional averages Poverty index 0.015 (0.023) Own radio -0.122 (0.083) Christian -0.127* (0.065) Muslim -0.241*** (0.090) Other religion -0.300 (0.224) Primary -0.104 (0.077) Secondary -0.038 (0.106) Tertiary 0.397** (0.190) Employed 0.003 (0.068) Community group 0.519*** (0.087) Observations 26,371 26,371 26,371 26,371 26,371 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Additional controls: age, age squared, rural dummy, country fixed effects. 20

Table 3. Gender specific OLS regressions and the differences in parameters. (1) (2) (3) (4) (5) (6) Dependent: voted raised issue Sample: Men Women Difference Men Women Difference Resources Poverty index 0.005-0.001-0.006 0.022*** 0.022*** -0.000 (0.004) (0.005) (0.006) (0.005) (0.006) (0.007) Primary -0.002 0.011 0.013 0.049*** 0.058*** 0.009 (0.010) (0.011) (0.014) (0.012) (0.012) (0.016) Secondary -0.006-0.008-0.002 0.034*** 0.015-0.019 (0.012) (0.015) (0.018) (0.011) (0.016) (0.019) Tertiary -0.011-0.031-0.020 0.079*** 0.084*** 0.004 (0.023) (0.029) (0.034) (0.019) (0.026) (0.027) Own radio 0.035*** 0.026*** -0.009 0.082*** 0.045*** -0.037*** (0.011) (0.008) (0.014) (0.011) (0.010) (0.014) Employment Employed 0.011 0.024** 0.013 0.028*** 0.031*** 0.003 (0.009) (0.010) (0.012) (0.010) (0.012) (0.014) Religion Christian 0.037*** 0.057*** 0.020 0.143*** 0.161*** 0.018 (0.009) (0.011) (0.013) (0.013) (0.014) (0.016) Muslim 0.050*** 0.045** -0.005 0.152*** 0.158*** 0.006 (0.013) (0.018) (0.021) (0.022) (0.022) (0.026) Other religion 0.053** 0.095*** 0.042 0.095** 0.136*** 0.041 (0.023) (0.033) (0.042) (0.041) (0.047) (0.044) Regional averages Poverty -0.019-0.036* -0.016 0.002 0.028 0.026 (0.015) (0.018) (0.018) (0.022) (0.024) (0.021) Male primary -0.066-0.155*** -0.089 0.055-0.187** -0.242*** (0.056) (0.058) (0.063) (0.079) (0.089) (0.072) Female primary -0.049-0.076-0.027-0.098 0.043 0.141** (0.059) (0.059) (0.064) (0.075) (0.084) (0.071) Male secondary 0.032 0.282*** 0.250*** -0.108 0.056 0.164* (0.076) (0.075) (0.080) (0.118) (0.130) (0.084) Female secondary 0.066-0.076-0.142* 0.019-0.040-0.058 (0.081) (0.081) (0.082) (0.092) (0.099) (0.088) Male tertiary -0.162-0.122 0.040 0.154 0.125-0.028 (0.159) (0.155) (0.152) (0.214) (0.256) (0.172) Female tertiary -0.032-0.058-0.025 0.231 0.187-0.044 (0.243) (0.194) (0.190) (0.226) (0.263) (0.189) Own radio 0.015-0.067-0.082-0.064-0.112-0.048 (0.053) (0.062) (0.073) (0.081) (0.087) (0.081) Male employed 0.011-0.140*** -0.151*** -0.126* -0.066 0.060 (0.051) (0.053) (0.053) (0.075) (0.084) (0.074) Female employed -0.038 0.098* 0.136** 0.104 0.118 0.014 (0.055) (0.056) (0.057) (0.084) (0.103) (0.094) Christian -0.080* -0.132** -0.053-0.102* -0.165** -0.063 (0.047) (0.053) (0.049) (0.060) (0.073) (0.060) Muslim -0.131* -0.110* 0.021-0.189** -0.215** -0.026 (0.067) (0.059) (0.067) (0.089) (0.091) (0.087) Other religion -0.060 0.052 0.113-0.338* -0.191 0.147 (0.101) (0.123) (0.137) (0.186) (0.275) (0.187) Community group 0.067 0.108 0.041 0.446*** 0.441*** -0.006 (0.059) (0.069) (0.066) (0.076) (0.104) (0.100) Observations 12,026 11,620 23,646 13,254 13,117 26,371 R-squared 0.090 0.107 0.102 0.130 0.105 0.131 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Additional controls: age, age squared, rural dummy, country fixed effects, constant. 21