Women and Voting in the Arab World: Explaining the Gender Gap

Transcription

1 Women and Voting in the Arab World: Explaining the Gender Gap Carolina de Miguel, University of Toronto Draft: April 2013 Special thanks to the panel members and audience at MPSA, April 2013 and to Mark Tessler for extremely useful comments and suggestions. The data for this paper comes from Arab-Barometer Survey (ICPSR2581- v1. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], doi: /icpsr26581.v1) 1

2 Abstract This paper examines the causes of the gender gap in electoral participation in competitive authoritarian regimes in the Arab world. I use public opinion surveys conducted in seven Arab countries (Algeria, Bahrain, Jordan, Lebanon, Morocco, Palestine, and Yemen) between 2006 and 2009 to show that women are significantly less likely to turn out to vote than men. In turn the paper develops and tests an explanation of this gender gap. I argue that the presence of individuals in the public sphere is particularly important in explaining their likelihood of turnout in the Arab World. Given that women are much less likely than men to participate in the public sphere in the Arab World (i.e. they are less likely to be employed and less likely to be members of organizations or groups), they will be less likely to turn out. In sum, the gender gap in electoral participation is explained by the differential rate of participation of women in the public sphere. To test this hypothesized causal relationship and its mechanisms I use mediation analysis. Keywords: electoral behavior; participation of women; electoral authoritarianism; Arab world; Middle East; mediation analysis.

3 Introduction Political participation is an essential part of the democratic process. Scholars of participation have long been interested in understanding who participates in a society and why in order to determine who has a voice in political affairs and who gets represented. Among such groups, women have received considerable attention (Desposato and Norrander 2009). The earliest studies of political behavior in Western Europe and North America established gender as one of the standard variables routinely used to explain levels of electoral turnout, party membership, and protest activism, alongside the most powerful predictors of age and education (Inglehart and Norris 2003, 101). Since then numerous scholars of political participation have written at length about the magnitude and the extent of this gender gap. Fewer studies however have moved beyond description to delve into the causes of this gender gap, and even fewer have tested these causal arguments empirically. Furthermore, most of the literature on the gender gap in political participation has been focused exclusively on the developed democratic world. This paper aims at addressing some of the theoretical and empirical lacuna in the literature through an exploration of the gender gap in electoral participation in the Arab World and its potential causes. The central argument of this paper is that the exclusion of women from the public sphere in the Arab World is the explanation for the gender gap in electoral participation. In other words, I claim that the mechanism explaining why women are less likely to vote in elections in the Arab World is that women are less likely to be present in the workforce and less likely to be members of formal organizations or groups. Being present in the public sphere (through participation in the workforce, trade unions, or other formal organizations or groups) is key to explain participation in the Arab World because it is through these channels that the regime mobilizes people to turn out to vote. Thus the lower presence of women in the these public places, makes it less likely that they will be mobilized, and thus makes it less likely for them to turn out to vote.i test my argument against alternative explanations using an original dataset of public opinion surveys conducted in seven Arab countries (Algeria, Bahrain, Jordan, Lebanon, Morocco, Palestine, and Yemen) between 2006 and Furthermore, since the goal of this paper is to be able to test the mechanism linking gender to vote to explain the cause of the gender gap, I develop and test a mediation model of turnout in which access to the public sphere is the mechanism linking gender to lower levels of voting. The paper is organized as follows. The first section describes the nature of political participation in the Arab World and establishes that there is a gender gap in both conventional and unconventional forms of political participation in the Arab Word. I show that women are significantly less likely than men to vote, sign petitions and protest. The second section focuses on the causes of the gender gap in voting, develops the argument of the paper and specifies some testable hypotheses. The third section describes the data and methods that I use to test the hypotheses developed in the previous section, and the fourth section presents and discusses the results. The article concludes with a review of the findings and their implications for understanding the role of women in political participation in the Arab world. The Gender Gap in Political Participation in the Arab World Electoral participation in competitive authoritarian regimes is a largely unexplored topic. The assumption is that elections are not meaningful in non-democratic contexts. Yet competitive authoritarian regimes in the Arab world regularly hold elections, and although these elections are not free and fair, both citizens and elites take these electoral contests quite seriously. Candidates hold 3

4 rallies, deliver speeches, distribute goods, post flyers, and visit local constituencies. Citizens, too, take great interest in campaigns and mobilize friends and family to turn out to vote. On the hand, elites are interested in elections as a mechanism to solidify their power and their bases of support, as well as a way to lend legitimacy to their regime in the eyes of the international community. For citizens and political parties, elections are a way of getting material benefits from the state (clientelism) (Blaydes 2007; Lust-Okar 2006) or a way to signal approval or disapproval with the regime for policy outcomes (de Miguel, Jamal and Tessler n.d.). Who turns out to vote and why is an important question in the Arab World (as well as in many other competitive authoritarian regimes). Table 1 presents aggregate figures on different forms of political participation in the seven Arab countries in the dataset that I am using. These figures suggest that participation is fairly prevalent in the Arab World and that the rates of participation are actually very similar to those of democratic countries. Turnout figures range between a low of 51 percent in Algeria to a high of 72 parent in Palestine, and as expected, conventional forms of participation (i.e. turnout in elections) are much more popular that non-conventional forms of participation (i.e. signing a petition or protesting). Furthermore, these figures suggest that there is significant within-country variation in participation, which raises the question of why some individuals decide to participate in the political process in these countries whereas others decide to abstain. Table 1: Political Participation in the Arab World Country Survey Year N Vote Petition Protest Algeria % 31.75% 22.18% Bahrain % 38.33% 36.87% Jordan % 13.23% 10.78% Morocco % 22.43% 19.78% Lebanon % 31.37% 45% Palestine % 29.49% 32.96% Yemen % 27.56% 29.28% Average 61.4% 27.56% 27.08% Source: Arab Barometer. Vote as percentage of all respondents (excluding missing cases) To contribute to the line of inquiry on the determinants of participation in the Arab World, this paper explores the particular question of whether women are less likely to participate in elections in the Arab World and if so why. There is a great deal of literature on gender and electoral participation in the developed world suggesting that women are less likely to participate than men, although this gap has been narrowing in many industrialized advanced economies (Norris and Inglehart 2003; Schlozman, Burns and Verba 1995). There is less research on gender and participation in non-developed countries, with some notable exceptions. Desposato and Norrander s (2009) article The Gender Gap in Latin America: Contextual and Individual Influences on Gender and Participation examines the relationship between gender and participation in seven Latin American countries; Chhibber (n.d.) explores the gender gap in India in Why some Women are Political Active: The Household, Public Space, and Political Participation, and Coffe and Bolzendahl (2011) have an article on the gender gap in Sub-Sharan African nations. Despite the important role of women in the Arab Spring events, there have been few studies on gender s impact on political participation in the Arab World. There is some work done on the political underrepresentation of women in public office (XXX) and on gender and political preferences (Benstead et al. 2012) but not much is known about individual level behaviour of women in elections. 4

5 Table 2 compares the average proportions of men and women s participation (across the seven Arab countries in Table 1) in three types of political activity: voting, signing petitions and protesting. The table presents the z-scores and their corresponding p-values. The table shows that the proportion of women that turn out to vote is smaller than the proportion of men. This differences is statistically significant meaning that based on the z-score and p-value we can reject the null hypothesis that the proportions are the same. Similarly, the proportion of women that sign petitions and that protest is also significantly smaller than the proportion of men engaging in these activities. The z-scores and significance levels suggest that these differences between men and women are statistically significant. It is important to note that the magnitude and significance of gender gap is larger in protest and petitioning than in voting. In other words, the gender gap is smaller in more conventional forms of participation (such as voting) than in less conventional forms of participation such as protesting or signing a petition. This seems in line with findings in developed Western countries as well as in other less-developed regions in the world (Coffe and Bolzendahl 2011). Table 3 presents the logit coefficient of gender as it is regressed on three separate dependent variables: Vote, Petition and Protest. The results corroborate the findings in Table 2 that being a women significantly reduces your likelihood of voting, signing a petition or protesting. The magnitude of the female effect is especially large and significant among non-conventional forms of participation. The last column of Table 3 present the change in probability of voting, petitioning or protesting if an individual is a women rather than a man. Being a women reduces your chances of voting by 3 percent and reduced the chances of petitioning and protesting around 15 percent. Table 2: Difference in Proportions Test Variable Male Female z-score p-value Vote Petition Protest Table 3: Bivariate Logit Regressions: Effect of Gender Dependent Variable Logit Coeff Standard Error P-Value Prob Change+ Vote Petition Protest *p<0.10,**p<0.05,***p<0.01 Sample Weights: Included Country Dummies: Included Reference Category: Algeria +From Min to Max Value Having established the existence of a gender gap in electoral participation in the Arab World the papers turns to understanding the causes of this gender gap. Why is it that women are less likely to participate in elections in the Arab World? In what follows I develop my argument and derive some testable hypotheses. In a nutshell my argument is that the exclusion of women from the public sphere in the Arab World explains their lower likelihood to turn out to vote. 5

6 Explaining the Gender Gap For a very long time, the role of women has been mostly associated with the private domain of the home; women have traditionally had a smaller presence in the public sphere than men. Although this gender divide has diminished in recent decades mostly in the developed world, women in many countries are still less likely to be in the work force, and are less likely to be members of trade unions and other political or social organizations. This inequality in the participation of public life is particularly striking in developing and democratizing regions of the world. Given that politics is an inherently public activity, scholars have suggested that the diminished presence of women in the public domain has negative consequences for their participation in political life and for their political representation. Studies in the developed world have found that individuals that are employed and that are members of organizations or groups are more likely to turn out to vote. For example, in a cross-national study, Ross shows that high rates of female labor force participation are positively (and significantly) related to percentage of seats held by women. These findings seem to be corroborated in the other regions of the world. According to Chhibber, [i]ndian women are more likely to participate in politics and elect female representatives when they have established an identity outside the household, often through work (Chhibber 2003). In many countries where women work in low-wage manufacturing including Guatemala, Taiwan, Hong Kong, India, Indonesia, Tunisia and Morocco they have formed organizations to protect their interests; often these organizations lobby for broader reforms in womens rights (Moghadam 1999) (Ross 2008, 108). In sum, this literature suggests that the presence of individuals in the public domain is key because individuals that are part of the work-force and/or that are members of organizations and groups such as political parties, trade unions and other social organization are more likely to develop an interest in politics, are better able to establish social networks that are key for collective action and tend to develop civic skills (Sapiro 1983; Schlozman, Burns, and Verba 1999 in Ross 208, 108). In sum, individuals in the work force and in formal organization tend to gain the necessary resources for political participation. This logic holds for the Arab World as well. Yet in competitive authoritarian regimes there is an additional (and perhaps even more important) reason for why for why participation in the public sphere (via participation in the labor force and/or membership in formal organizations) is key for electoral participation, which is the ability of regimes to target people for political mobilization. Much of the literature on the Arab World views electoral contests as a mechanism for elite manipulation. Elections in authoritarian regimes contribute to entrenched authoritarian rule in two different ways. On the one hand, the introduction of elections by authoritarian rulers legitimizes their rule to both domestic and international audiences. On the other hand, electoral contests allow authoritarian rulers to successfully manage the opposition as well as other governing elites by including them in the political process (Blaydes 2011). This instrumental view of elections suggests that rulers in competitive authoritarian regimes care about mobilizing people to participate in electoral contests. This is accomplished in a variety of ways from mobilizing people through tradeunions and political parties, to using clientelist networks to buy people s vote and support by giving away material benefits. This view of elections suggests that individuals in the Arab World that are in the work force and that are part of political parties, trade unions, and other organizations are more likely to be targeted and mobilized by the regime to turn out to vote. In sum, being present in the public sphere is key to being mobilized politically. We should thus expect that individuals that are more present in the public sphere through membership in organizations or participation in the labor force to be more likely to participate in the electoral process. This need not be the case for other forms of participation (especially for the most unconventional ones), since the regime is likely not waiting to increase this type of mobilization. 6

7 Given that women are significantly less likely to have access to the public domain than men in the Arab World, and that access to the public domain is key to being mobilized to turn out to vote, I argue that the key to women s gender gap in electoral participation in the Arab World is mostly due to their unequal levels of participation in the public domain, and more specifically to their lower likelihood to be part of the workforce and to participate in other organizations or groups of social and political character. In other words I posit that women in the Arab World are less likely than men to be present in the public sphere, which in turn leads to women having a smaller likelihood of turning out to vote. In other words, the mechanism explaining the negative relationship between gender and voting in the Arab World is about participation in the public sphere. This leads to the following two hypothesis: H1. Women are less likely to be employed, which in turn leads to a smaller likelihood of turning out to vote. H2. Women are less likely to be members of organizations, which in turn leads to a smaller likelihood of turning out to vote. In sum, the question motivating this paper is what explains the gender gap in electoral participation in the Arab World. I develop a set of hypotheses that get at the mechanisms of the relationship between gender and vote. I posit that the relationship between gender and voting in the Arab World is mediated by an individual s access to the public sphere, and thus that differential access to the public sphere (between men and women) is the key to explain the differences in electoral participation, that is the gender gap. Furthermore, the mediating effect of employment and membership should be significant in voting, but not in other forms of political participation. In other words, these two hypotheses hold for electoral participation, but not for non-conventional forms of political participation. H3. Participation the public sphere in the form of membership in organizations or participation in the work force should not mediate the relationship between gender and non-conventional forms of participation (i.e. protest or petition) Alternative Explanations The argument that I present in this paper builds on a broader category of arguments centred around the idea of differential resources (Desposato and Norrander 2009) between men and women. The argument that scholars tend to make is that men and women have unequal access to resources and opportunities that are key for participation and this unequal access explains why women tend to participate less than men. One of the resources that is key to participation is education, and thus scholars have argued that the fact that women are less educated means that they will also be less likely to participate politically (Inglehart and Norris 2003). I test this mediation hypothesis as an alternative hypothesis. Finally, religion has been cited as a potential explanation for the decreased participation of women in politics in the Arab World. Some scholars argue however that Islam has not deterred women in non-arab Islamic countries such as Indonesia, Bangladesh and Pakistan from reaching top elected positions. It can also be added that some Islamist political parties in the Arab world have promoted their women members as parliamentary candidates. (Sabbagh 55). 7

8 H4. Women are less likely to be educated, which in turn leads to a smaller likelihood of turning out to vote. H5. Women are more likely to be religious, which in turn leads to a smaller likelihood of turning out to vote. Data and Methods Data for this paper are drawn from an original dataset that pools public opinion surveys conducted in seven countries in the Arab World (Algeria, Bahrain, Jordan, Lebanon, Morocco, Palestine, and Yemen) between 2006 and The surveys constitute the first wave of the Arab Barometer and consist of large representative national samples. 1 While it would be interesting to examine the specificities of the gender gap in each of the seven countries, the goal of the present article is to explore general, region-wide patterns on gender and voting. Accordingly, we analyze data from the seven surveys in a pooled format, rather than country by country. The total number of observations in this pooled dataset is I weight the data to correct for unequal sample sizes across country-surveys. 3 Furthermore, although I am interested in individual-level relationships, it is important to account for country-specific factors that influence individual-level decisions about turning out to vote. 4 In order to control for these factors I run the analyses with country dummies (with Algeria being the reference category, and thus the omitted country variable). Therefore all regression analyses presented in this paper include weights (for unequal sample sizes) and fixed effects using country dummies. Method: Non-Linear Mediation The hypotheses developed in this paper present the mechanisms explaining the negative relationship between gender and vote. My main proposition is that the relationship between gender and voting is mediated by an individual s access to the public sphere. This hypothesized relationship is shown graphically in Figure 1. Following Baron and Kenny (1986) and Judd and Kenny (1981) I test this hypothesized mediation by running the following two regression models: Public Sphere Presence = β 0 + β a Gender + β d Covariates + ε (1) 1 These seven countries encompass most of the political, economic and demographic diversity of the Arab world. All of the Arab worlds geographic regions are represented, and among the seven countries are monarchies and republics, countries that are not free and countries that are partly free, more affluent and less affluent countries, some of the largest and some of the smallest countries in the Arab world, and countries with varying degrees of sectarian diversity. An Arab Barometer survey was also conducted in Kuwait in 2006 but it is not included in the present analysis because only Kuwaiti men were permitted to vote at the time of the survey. 2 The analyses that I present in the following pages have a smaller number of observations due to missing values for several of the variables included in the regression models. 3 The weight applied to each survey is 1/N*1000, where N=sample size of the survey. See Eifert et al. (2010), Karp and Banducci (2008) and Jamal and Nooruddin (2007) for a similar use of weights in a pooled analysis with several country-level surveys. 4 An important body of literature (Blais 2006; Jackman and Miller 1995; Karp and Banducci 2008) suggests that country-level institutional arrangements and conditions, such as electoral system, registration requirements, ethnic fractionalization, and level of economic development, help to explain variance in individual-level electoral behavior, including voter turnout. 8

9 Figure 1: Modeling Mediation Public sphere presence a b Gender c Political Participation Indirect Effect (a * b) Direct Effect (c ) Total Effect [(a * b) + c ] % Total Effect Mediated = Indirect Effect / Total Effect Vote = β 0 + β b Public Sphere Presence + β c Gender + β d Covariates + ε (2) Model 1 establishes that the key independent variable gender is correlated with the mediating factor public sphere presence. The coefficient on the independent variable in this model corresponds to path a in Figure 1. If there is a mediating effect I should expect this coefficient to be negative and significant, indicating the women are less likely to have a presence win the public sphere than men. Model 2 shows how the mediator variable public sphere presence affects the outcome variable vote controlling for the independent variable of interest gender. The coefficient on the mediator variable corresponds to path b in Figure 1. I expect this coefficient to be positive and significant indicating that an individual s presence win the public sphere increases the individual s likelihood of turning out to vote. Finally, Model 2 also shows how the independent variable gender affects the outcome vote when controlling for the mediator variable public sphere presence. The coefficient on the independent variable gender in Model 2 corresponds to path c in Figure 1. I expect this coefficient to be close to zero (and not significant). According to Baron and Kenny (1986) if these three conditions are met (negative and significant a coefficient, positive and significant b coefficient and not significant c coefficient) then I can reasonably conclude that the mediation is significant because it means that the indirect paths are significant and not the direct path. In the empirical section I run these two models (Model 1 and Model 2), and report the coefficients of these three paths to determine whether there is significant mediation for each of my models/hypotheses. 5 In addition, one of the most common ways to capture the size of the mediated effect is to look at the ratio of the indirect effect to the total effect, which is interpreted as the proportion of the total effect that is mediated (Preacher and Kelley 2011, 97). I use Hicks and Tingleys (2011) stata mediation package to calculate this quantity of interest. This package is adequate for non-linear mediation models and allows for the inclusion of weights, which I need to 5 I use the same covariates in both regression models since current Stata packages that test for mediation in non-linear models require the same covariates in both regressions. 9

10 correct for different survey sample sizes. 6 I will thus report on both the individual path coefficients for the indirect effect and direct effect and their significance levels as well as on the percent of the total effect that in our model that is due to mediation. Dependent Variable The main dependent variable in this study consists of the responses to the question: Did you participate in the last election? This is a dichotomous variable which I call vote. A value of 1 indicates that the respondent turned out to vote in the most recent election, and a value of 0 indicates that the respondent did not turn out to vote. Although there is a legitimate concern that respondents might over-report voting, particularly in authoritarian regimes (Karp and Banducci 2008), any such bias in our case appears to be small given the similarity between the official figures reported by IDEA and the figures calculated from the surveys. See Table 1. 7 In testing hypothesis 3 I will use protest as the dependent variable. This is also dichotomous variable based on responses to the question: Did you attend a demonstration or protest march in the past three years? A value of 1 indicates that the respondent participate in a protest or march one or more than once and 0 indicates that the respondent never participated in a protest or march. Table 4: Turnout Figures Country Survey Year N Survey Turnout Figures+ IDEA Turnout Figures++ Algeria % 46.2% Bahrain % 67.0% Jordan % 57.8% Morocco % 51.6% Lebanon % 46.5% Palestine % 77.7% Yemen % 65.2% Av. Turnout 61.4% 50.1% + Source: Arab Barometer. Vote as percentage of all respondents (excluding missing cases) ++ Source: IDEA: Vote as percentage of registered voters. 6 Testing mediation models become significantly more complicated if one or both of the dependent variables is dichotomous since the scale of the coefficients in the two models (Model 1 and Model 2) will be different. See Nataniel R. Herrs website for a detailed discussion of mediating effects with dichotomous outcome and dichotomous mediator variables. See also David D. Kennys website as well as Buis (2010). According to Kenny (2008), in a non-linear model the coefficients from the regressions should be rescaled (standardized) before computing the indirect effect. Until recently, this could only be done with software specifically designed for structural equation models (i.e. Lisrel, Mplus) but the models could not include weights. In the past two years, Stata has published two user-written packages that tackle the issue of mediation in non-linear models. These two packages are mediation (Hicks and Tingley 2011) and binary mediation (Ender n.d.). Both packages are appropriate for non-linear models and thus calculate the correct (standardized) indirect and direct effects with their corresponding standard errors (and confidence intervals). I use the former in this paper because it allows for the inclusion of weights, but for robustness, I also run our models with the Stata package, binary mediation, (Ender n.d.), which allows the use of bootstrapped standard errors but does not allow for weights. The results are very similar. 7 For a discussion on biased responses in democratic and authoritarian countries see Karp and Banducci (2008) and Wright and Stein (2010). 10

11 Independent and Mediating Variables My argument represented in Figure 1 involves testing whether presence in the public sphere is a mediating variable between gender and the likelihood of turnout. Gender is measured by a variable called female that takes the value of 1 if the respondent resorts to be female and 0 if the respondent resorts to be male. To capture the mediation factor public sphere presence I focus on two variables: member and employed. The variable member is a measure of associational activity that takes the value of 1 if the respondent is a member of any formal organization or group (i.e. trade unions, political parties, civic organizations, churches...) and 0 if the respondent is not a member of any organization or formal groups. The variable employed captures the presence in the workforce and takes the value of 1 if the respondent is employed and 0 if the responded is not employed. In sum, the ultimate dependent variable vote and the two mediators variables member and employed are all binary responses. In testing the alternative hypotheses 4 and 5 I use two different mediating variables: education and religiosity. Education has seven categories and higher values indicate more educated. The variable religiosity is based on the following survey question: In general, would you describe yourself as religious or not religious? The variable is dichotomous and takes the value of 1 if the respondent answers to be religious, and 0 if the respondent answers to be not religious. Note that both education and religiosity will also be used as control variables when testing for my main hypotheses. Controls All mediation models that I test include the same set of covariates, which aim at controlling for alternative factors that affect the likelihood of voting. Scholars of political participation in developed democracies have consistently shown that older individuals, those that are more educated and those that have more income are more likely to participate. I thus include all these variables as controls. My measure of an individuals economic vulnerability is called personal economic condition and is based on the following survey question: How would you rate the economic situation of your family today? There are four possible response options: 1=very bad; 2=bad; 3=good; 4=very good. I include age, which is a seven category variable which I treat as continuous. Higher values indicate older. In addition to these socio-demographic factors, I include an attitudinal variable that is often associated with turnout, which is trust in institutions. This variable is a factor score based on responses to the following three questions: Im going to name a number of institutions. For each one, please tell me how much trust you have in them: Prime Minister; The Courts; Parliament. For each item there were four possible responses: a great deal of trust, quite a lot of trust, not very much trust and no trust at all. The factor scores for these three variables load highly on one dimension so using these scores I create a combined measure called trust in institutions. See Table 5. Higher values indicate more trust in institutions, whereas lower values indicate less trust in the institutions of government. Finally, as discussed earlier, I include country dummies to control for contextual/country-level factors affecting individual-level decisions to turn out to vote. The reference category in all models is Algeria (which is thus excluded from the models). In the appendix I report descriptive statistics for all the variables in the models. See Table

12 Results Table 5: Factor Loadings and Cronbach s Alpha for Items of trust in institutions Factor Cronbach s Items Loadings Alfa Trust in Primer Minister Trust in the Courts Trust in Parliament Scale reliability coefficient 0.78 Table 6 presents three different regression models. The first regression model is the base model. The dependent variable is vote so the coefficients reported are logit coefficients. Focusing on the variable female the model shows that it has a negative and significant effect on the likelihood of voting. For ease of interpretation I have transformed these logit coefficients into predicted probabilities, which I present in Table 7 8. This table tells us that being female as opposed to male decreases the probability of turning out to vote about 3 precent when controlling for a set of covariates. This model confirms that there is indeed a gender gap in electoral participation in the Arab World in that female is negatively related to vote. The following two regression models in Table 6 labeled mediation model A aim at testing the mediated relationship in hypothesis 1. The first of those two regression models is the same model as the base model with the addition of the mediator variable employed. The dependent variable in this second model is also vote. This model shows how the mediator variable employed affects the outcome variable vote controlling for the independent variable of interest female. If there is indeed mediation then we expect the coefficient on the mediator variable employed to be positive and statistically significant, but we expect the coefficient on the independent variable of interest female not to be significant. In other words, the effect of female on vote should disappear when introducing the mediator variable employed. And this is exactly what we find. The third regression model in this table takes the mediating variable employed as the dependent variable and regresses it against our main independent variable of interest female and the rest of covariates. The goal of this third regression model is to establish that the key independent variable female is positively correlated with the mediating variable employed. This is the case since the coefficient on female is statistically significant in the expected direction: women are much less likely to be employed than men. Figure 2 presents the relevant coefficients of this mediation graphically, and allows us the better walk through the results of this mediation analysis. 9 The indirect effect is represented by path a and path b, and both pathways have significant coefficients in the expected direction. The coefficient in path a is negative and statistically significant indicating that women are less likely than men to be employed. Since this coefficient represents the unstandardized log odds of voting, it is difficult to interpret. I have therefore transformed these log odds into predicted probabilities for ease of interpretation. The predicted probability of being employed if female is 0.30 (holding all 8 This table shows the change in the probability of voting for each independent variable in the base model. The first column shows the change in the probability of voting when the independent variable varies from its minimum to its maximum (holding all other variables at their means). The second column shows the change in the probability of voting when the independent variable varies one standard deviation. 9 The coefficients in this figure are taken directly from the two regressions in Table 2 labeled Mediation Model A. The coefficient in path a is an unstandardized OLS coefficient and the other two coefficients (paths b and c ) in this figure are unstandardized logit coefficients. 12

13 Table 6: Mediation Model A: Employment Base Model Mediation Model A Dependent Variable Vote Vote Employed Female.122* *** (.07) (.07) (.07) Employed.359*** (.07) Trust Institutions.320***.327***.067* (.04) (.04) (.04) Good Personal Econ Condition.159***.161***.074 (.05) (.05) (.05) Religiosity *** (.08) (.08) (.08) Education.129***.110***.276*** (.02) (.02) (.02) Age.310***.299***.088*** (.03) (.03) (.03) Jordan *** (.11) (.11) (.12) Palestine.706***.762***.607*** (.12) (.12) (.12) Morocco.303**.291**.324** (.12) (.12) (.13) Lebanon.341***.264** 1.139*** (.11) (.11) (.12) Bahrain.300*.298*.021 (.15) (.15) (.16) Yemen.500***.499***.144 (.13) (.13) (.13) Constant.642***.770***.913*** (.21) (.21) (.22) No. of cases *p<0.10,**p<0.05,***p<0.01 Sample Weights: Included Reference Category: Algeria 13

14 Table 7: Predicted Probabilities of Voting: Base Model from Table 6 min to max +/- stdev Female Trust in Institutions Good Personal Economic Condition Religiosity Education Age Note: Holding other variables at their means other variables in the model at their means), whereas the predicted probability of being employed if male is 0.62 (holding all other variables in the model at their means). The change in probability of voting from being a male to a female is -0.31, which is a substantial change. The coefficient in path b is positive as expected and also significant. Again since this coefficient represents the unstandardized log odds of voting, we turn to predicted probabilities for ease of interpretation. The predicted probability of voting if employed is 0.69 (holding all other variables in the model at their means), whereas the predicted probability of voting if unemployed is 0.61 (holding all other variables in the model at their means). The change in probability of voting from being unemployed to being employed is 0.08, indicating that those employed have an 8 percent higher change of turning out to vote than those unemployed. Finally, the direct effect of being female on vote (while controlling for employed) (path c) is not statistically significant, which is what we would expect if there is to be mediation. The fact that the paths of the indirect effect (path a and path b ) are statistical significant and in the expected direction, and that the path of the direct effect (path c) is not statistically significant, provides strong support for the theory that the relationship between gender and likelihood of turnout is indeed mediated by employment or participation in the labor force. What these results tell us is that women are less likely to be employed and that this leads to a smaller likelihood to turn out to vote. In other words, that the gender gap in voting is explained by differential levels of employment between men and women. Furthermore, we can conclude that this mediation seems to be full mediation, instead of partial mediation because the direct effect (c ) is not statistically significant and the size of the coefficient is quite small.finally, in order to get a sense of the magnitude of the indirect effect I calculate the percentage of the total effect that is mediated. In this model 82 precent of the total effect of female on turnout is mediated through employment, which corroborates that this is practically full mediation, and that the effect is substantively large. The key mechanism linking gender to voting in the Arab World is an individual s presence in the public sphere, which I operationalize with two different variables: employed and member. I now proceed to test the mediated relationship with member as the mediating factor (hypothesis 2). The results for this second mediation model (Model B) are presented in Table 8 and in Figure 3 below. The results of Model B corroborate the findings in Model A that women are less likely to be present in the public sphere and that this absence explains their smaller likelihood to turn out to vote. In other words, the indirect paths a and b are both statistically significant in the expected direction, and the direct path between gender and vote is not significant. Furthermore, the percent of to the total effect that is mediated remains quite large. Taken together these two mediation models (A and B) provide strong support for the argument presented in this paper that presence in the public sphere is the mechanism explaining the negative relationship between gender and 14

15 Figure 2: Mediation Model A Employed a = -1.30*** b = 0.36*** Female c = Vote Percent of Total Effect Mediated = 82% voting in the Arab World. What I show is that the gender gap (i.e. women s smaller likelihood of turning out to vote) is due to women s smaller presence in the public sphere both in term of participation in the labor force (variable employed ) and participation in formal organizations or groups (variable member ). Figure 3: Mediation Model B Member a = -0.71*** b = 0.70*** Female c = Vote Percent of Total Effect Mediated = 48% Finally, the last implication of my argument is that it should only work to explain the gender gap in electoral participation and not in other forms of political participation such as protest (hypothesis 15

16 Table 8: Mediation Model B Dependent Variable Vote Member Female *** (.07) (.08) Member.697*** (.09) Trust Institutions.310***.100** (.04) (.04) Good Personal Econ Condition.164***.049 (.05) (.06) Religiosity * (.08) (.09) Education.115***.199*** (.02) (.03) Age.311***.021 (.03) (.03) Jordan *** (.11) (.17) Palestine.691***.293** (.12) (.13) Morocco.328***.117 (.12) (.15) Lebanon.360***.100 (.11) (.13) Bahrain.273*.137 (.16) (.18) Yemen.464***.551*** (.13) (.14) Constant.739*** 2.209*** (.21) (.25) No. of cases *p<0.10,**p<0.05,***p<0.01 Sample Weights: Included Reference Category: Algeria 16

17 3). As I argued, the presence of individuals in the public sphere is particularly relevant to elections because in competitive authoritarian regimes elites are interested in mobilizing the population to turn out to vote, and those who are in the public sphere are more easily targeted and thus mobilized by the regime. However, the regime is not necessarily interested in mobilizing people into protest, so the presence of individuals in the public sphere should not be relevant to explain the link between gender and protest. To test this proposition (hypothesis 3) I show that member and employed do not explain the relationship between gender and protest in the Arab World. I run two mediation models (similar to models A and B) where the ultimate dependent variable is protest and not vote. Figure 4 presents the results of these two mediation models (Model C and D). The full tables of results are presented in the appendix. The results show that the percent of the total effect of gender on protest that is mediated by membership in an organization is 11 percent (which is not particularly high). Similarly, the percent of the total effect of gender on protest that is mediated by employed is even smaller (4.3 percent). What these mediation models teach us is that women are less likely to engage in non-conventional forms of participation, but this gender gap is not explained by women being less likely be in the public sphere. This is consistent with my argument that participation in the public sphere is uniquely important for electoral participation (and not necessarily for other forms of participation). Testing Alternative Explanations I test for two alternative explanations of the gender gap. Hypothesis 4 states that the mechanism linking gender and vote is education, whereas hypothesis 5 states that the mechanism linking gender and vote is religion. To test these two hypotheses I run similar mediation models as in the previous section but with education and religion as mediating variables. The results of these models are presented in Figure 5. The top figure suggests that education seems to mediate some of the relationship between gender and vote, but religion does not. It is important to note that the mediating effect of education is much smaller than the mediating effect of member or employed suggesting that presence in the public sphere is a more important explanation for the gender gap than is education or religiosity. In sum, taken together all these analyses suggest strong support for the claim that the gender gap in the Arab World is in great part explained by the different presence of women in the public sphere in these countries. Finally, some might argue that the actual relationship between gender, presence in the public sphere and vote is not a meditational one but an interactive one. In other words, it might be possible that the gender gap is explained by the different effect of employed and member on the likelihood of voting for men and for women. This type of relationship is an interactive one, and would suggest that being present in the public sphere is a factor that is more important for men than for women when it comes to predicting turnout. This type of argument is common in the literature in the developed world as well as in the developing world (Desposato and Norrander 2009). To rule out this alternative specification, I run this interaction model by splitting the sample between men and women. Results are presented in Table 9. The results in this table allow me to gauge whether the effect of employed and member on vote is the same for both male and females. Table 9 shows that being employed and being a member of a formal organization or group matters positively and significantly in predicting vote for both men and women. The coefficients on these two variables are positive and statistically significant for both men and women. A more graphical representation of these results is presented in Figure 6 for the interaction between gender and employed and Figure for the interaction between gender and member. 17

18 Figure 4: Mediation Model C and D Member a = -0.72*** b = 0.90*** Female c = -0.72*** Protest Percent of Total Effect Mediated = 11% Employed a = -1.31*** b = 0.12 Female c = -0.77*** Protest Percent of Total Effect Mediated = 4.3% 18

19 Figure 5: Mediation Model E and F Education a = -0.3*** b = 0.13*** Female c = -0.12* Vote Percent of Total Effect Mediated = 24% Religious a = 0.52*** b = 0.07 Female c = -1.22* Vote Percent of Total Effect Mediated = -5.4% 19

20 The way to read these figures is the following. Figure 6 shows the probability of voting by gender and employment status. Both graphs in the figure are the same, except the bottom graph shows the 95% confidence intervals to determine whether differences are significant. The top graph shows that becoming employed increases the likelihood of voting for both males and females, but the increase is slightly higher among men. In other words, becoming employed has a larger positive effect on vote for men than for women. However, as we see in the bottom graph, the confidence intervals for the probability of voting for men and women are overlapping, which means that they are not statistically different from each other. The same is true for the effect of membership on the likelihood of voting. Being a member of a formal organization or group increases the probability of voting for both men and women. The increase is slightly higher among men than women (bigger slope for men), but once again the difference is not statistically significant. Both of these results suggests that being present in the public sphere increases the likelihood of voting for men and for women alike. Therefore the differences in voting between men and women (i.e. the gender gap) cannot be explained by a different effect of this variable. This lends further support to my argument that the gender gap can be explained by the differential presence of men and women in the public sphere. Discussion and Implications This paper is concerned with exploring the causes of the gender gap in electoral participation in the Arab World. After establishing that there is a gender gap in participation, the paper develops and tests an explanation for this gender gap. I argue that the presence of individuals in the public sphere is particularly important in explaining their likelihood of turnout in the Arab World. Given that women are much less likely than men to participate in the public sphere in the Arab World (i.e. they are less likely to be employed and less likely to be members of organizations or groups), they will be less likely to turn out. In sum, the gender gap is explained by the differential rate of participation of women in the public sphere. Although some research has proposed that differential resources of women (in terms of education, income, participation in organizations... ) might explain why they are less likely to participate in politics, few studies have actually tested this explanation empirically. This paper contributes to the body of literature on political participation by empirically testing these propositions. I argue that the differential resources argument is an argument about mechanisms and that in order to test this argument we should use mediation analysis, and not interactions. Using mediation analysis I show that differential access to the public sphere between men and women is the mechanism that explains women s smaller likelihood to turn out to vote in the Arab World. I also show that alternative explanation are not as important as the one I posit. Although education is quite significant in explaining the gender gap, it is not as important as being employed or being a member of organizations. Furthermore, I show that religiosity has not explanatory effect on the gender gap. Finally, I show that the interactive story (as opposed to the meditational story) cannot explain the gender gap. In other words, both membership in organizations and employment are equally important for men and for women. Finally, this paper wishes to contribute to a new research agenda on the gender gap in participation in less developed and less democratic countries. 20