The Turnout Gender Gap

The literature on female enfranchisement can essentially be divided into those works trying to understand when and why female suffrage was implemented (Hicks Reference Hicks2013; Przeworski Reference Przeworski2009; Teele Reference Teele2018) and those examining its consequences on a variety of outcomes: expansion of the welfare state, educational attainment, persistence or erosion of gender norms, or their impact on political behaviour (Abrams and Settle Reference Abrams and Settle1999; Corder and Wolbrecht Reference Corder and Wolbrecht2006, Reference Corder and Wolbrecht2016; Kose, Kuka and Shenhav Reference Kose, Kuka and Shenhav2021). In studying the effect of female enfranchisement, albeit with varying degrees, works focusing on political behaviour have generally documented the presence of a turnout gender gap. For instance, in Finland or Austria the gender turnout gap was below 4 percentage points (pp) during the interwar period, while in Sweden women turned out to vote around 12–15 pp. less than men in 1921 (Morgan-Collins and Natusch Reference Morgan-Collins and Natusch2022), around 25 pp in Iceland (Tingsten Reference Tingsten1937), and around 30 pp in the US (Corder and Wolbrecht Reference Corder and Wolbrecht2016).

Why was the gender turnout gap wider in some countries than in others? Different potential explanations to this question arise. General explanations—not specific to post-female enfranchisement literature—trying to understand differences in political participation between men and women have highlighted several factors. The most broadly acknowledged is the role of individual characteristics and resources. The idea is that women have traditionally had lower levels of education, income, or other resources to deal with politics (for example, Burns, Schlozman and Verba Reference Burns, Schlozman and Verba2001; Iversen and Rosenbluth Reference Iversen and Rosenbluth2006) and this led to lower levels of turnout. Yet, given similar characteristics in terms of education or income, women appear to participate at different rates across different contexts. Besides resources, institutional reforms that affect the costs or incentives to vote can also shape the extent to which there are gendered differences in turnout (Kittilson and Schwindt-Bayer Reference Kittilson and Schwindt-Bayer2012; Beauregard Reference Beauregard2014). Compulsory voting provisions that increase the costs of abstaining reduce turnout gender gaps (Córdova and Rangel Reference Córdova and Rangel2017), and in referendum votes, where the individual sense of political efficacy is larger, gender gaps tend to be smaller Kim (Reference Kim2019).

Some of these individual and institutional factors have also been highlighted by the literature focusing on gendered turnout gaps upon enfranchisement. For instance, individual characteristics mattered and were often aggravated, upon enfranchisement, by some barriers to voting, such as literacy tests, or poll taxes, which made participation even more costly for women (Corder and Wolbrecht Reference Corder and Wolbrecht2006). However, the most recent explanations have focused on the importance of institutions to understand turnout gender gaps once women were granted the right to vote. For instance, a recent contribution by Skorge (Reference Skorge2021) has documented how the transition from plurality to proportional electoral rules in Norway increased turnout among women and Teele (Reference Teele2023) shows that the effect was larger in those districts that shifted from being noncompetitive under plurality to competitive under PR.

Multiple contributions have emphasised the importance of institutions in part because these can shape parties’ incentives and mobilisation strategies. Women, as “peripheral voters’’ (Teele Reference Teele2023), were more sensitive to political mobilisation by political organisations than men under specific circumstances. Several contributions pinpoint that women were more likely to turn out in those places where elections were more competitive (Corder and Wolbrecht Reference Corder and Wolbrecht2016; Morgan-Collins Reference Morgan-Collins2023) Yet, the argument, as noted by Teele (Reference Teele2023), is conditional on the type of institution in place. While PR is “automatically” associated with increases in political mobilisation and thus reductions in gender turnout gaps, the impact of female enfranchisement under majoritarian rules depends on the levels of competitiveness and party mobilisation in the district.

Other works have pointed out additional sources of female political mobilisation, mostly related to the existence of social networks. On the one hand, political pre-enfranchisement female activism through voluntary networks had long-term consequences; women anti-slavery canvassers had a higher likelihood to become suffragettes (Carpenter and Moore Reference Carpenter and Moore2014), and female pre-suffrage petitioning also boosted female turnout after enfranchisement Carpenter et al. (Reference Carreras2018). The pre-enfranchisement strength posited a credible threat and ensured the representation of women’s interests in parliament (Morgan-Collins Reference Morgan-Collins2021). On the other hand, horizontal and “non-political’’ social networks can help overcome voting barriers, especially for working-class women, and increase female turnout (Morgan-Collins and Natusch Reference Morgan-Collins and Natusch2022).

Recognising the importance of institutional designs, political competitiveness and party mobilisation, we complement previous works by looking at the effect on turnout of being a polling officer. By studying a case that keeps constant almost all institutional features, we complement existing explanations by focusing on an additive factor that could help in mobilising women. In addition, we build on previous works on social networks and examine whether an important political socialising event spills over to the immediate networks, thereby helping to close the gender turnout gap among women who did not formally participate in the management of the elections.

All in all, we focus on an institutional design that induces a direct individual-level political/electoral experience. In turn, this event can be crucial to activate women politically and spur their turnout in subsequent elections. Moreover, as previous literature has emphasised, we acknowledge that these individual experiences cannot be detached of certain contextual conditions. We additionally examine how the jump-start in political awareness among women polling officers interacted with the importance of political mobilisation efforts and the existence of social networks. We contend that both elements—individual engagement and social mobilisation—should further help in closing the turnout gender gaps by enhancing women’s individual levels of political participation.

Learning Democracy, Voting Habit, and Civic Duty

To understand why women vote less upon female enfranchisement, it is important to focus on the factors that are related to the decision to turn out. Previous evidence shows that previous voting behaviour is a good predictor of turnout in subsequent elections (Gerber, Green, and Shachar Reference Gerber, Green and Shachar2003). Recently enfranchised individuals inherently lack a habit of voting and therefore their turnout rates trail that of the other groups in the first elections after enfranchisement. For some of these new voters, the habit of voting can arise as a natural consequence of their resources, political interest, or socialisation (Verba and Nie Reference Verba and Nie1972), but in other cases, it is only developed through the experience of voting (Coppock and Green Reference Coppock and Green2016; Green and Shachar Reference Green and Shachar2000). Previous evidence has shown that the development of pro-mobilisation attitudes or voting habits among new voters can be influenced by the nature of the first election in which young people can vote (Franklin and Hobolt Reference Franklin and Hobolt2011), and how close in time elections take place when becoming eligible to vote (Meredith Reference Meredith2009).

Similarly, citizens may be more likely to develop a sense of civic duty after they participate in the electoral process (Dinas Reference Dinas2012; Gerber, Green, and Shachar Reference Gerber, Green and Shachar2003)—yet, the direction of causality is being debated (Feitosa, Stiers, and Dassonneville Reference Feitosa, Stiers and Dassonneville2022). Some contributions relate the closing of the gender turnout gap to increasing levels of civic duty among women (Carreras Reference Carreras2018; Galais and Blais Reference Galais and Blais2019). Therefore, serving as a poll officer has the potential to arise as a powerful politicisation tool. When new voters face formal and informal barriers to their electoral participation —political, economic, or social —exposing them to the functioning of democracy through election management can represent an efficacious tool to learn about democracy, and to induce a habit of voting and a sense of civic duty on these new voters.

Electoral Engineering and Polling Officers

Electoral engineering (Norris Reference Norris2004) or electoral ergonomics (Bruter and Harrison Reference Bruter and Harrison2020) examines how different institutional set-ups affect individual electoral behaviour, including turnout inequalities. The main idea is that the election administration process shapes voters’ participation costs —e.g., Orford et al. Reference Orford, Railings, Thrasher and Borisyuk2011; Potrafke and Roesel Reference Potrafke and Roesel2020—but also how individuals interact with elections and political events more broadly.

Poll officers play an important role in elections as they are essential to ensure the transparency and integrity of electoral procedures (Clark and James Reference Clark, James, Norris and Nai2017) and can influence voters’ perception on the integrity of elections (Hall, Monson and Patterson Reference Hall, Monson and Patterson2009). The selection procedure of polling officers is crucial to guarantee their independence and neutrality, and good practice is to elect polling officers by lot and to make the appointment binding (ACE Project 2012)—which is not always straightforward (Cantú and Ley Reference Cantú and Ley2017).

The selection of poll officers by lot implies that citizens can be randomly exposed to the functioning of elections. Recent works analysing these lotteries in contemporary settings have found effects on turnout (Artés and Jurado Reference Artés and Jurado2023) and on vote choices made by voters (Neggers Reference Neggers2018). By focusing on the effect of polling officers’ selection on the eve of female enfranchisement, this article examines whether this role contributed to developing a participatory culture among individuals who had never been directly involved in the functioning of democratic elections before.

Expectations

We contend that a direct political experience in elections as a poll officer represents a political socialisation event among recently enfranchised groups that should minimise turnout gaps. If individual resources matter to understand turnout (Burns, Schlozman and Verba Reference Burns, Schlozman and Verba2001), experiencing direct exposure to the functioning of an election for the first time acts as a political socialisation experience, as a political activation device, and endows an individual with resources. Direct involvement in the management of elections should make women learn about politics and elections, and develop faster a sense of civic commitment.

We expect that being exposed to election management lowers the cost of participating and has a positive effect on political activation as a result of three different processes. First, being involved in the management of elections implies that an individual has to learn and implement several concepts related to core democratic rules. Second, women serving as officers might update their prior thoughts that politics is a ‘male business’. Women become more aware that their political rights and duties are formally no different than the ones in the hands of men. Third, a larger political knowledge among officers might motivate them to consume more political information, which enhances turnout (Verba, Schlozman and Brady Reference Verba, Schlozman and Brady1995). Overall, serving as a poll officer and being exposed to the intricacies of a democratic electoral engineering process should not only give women a direct political experience on how the elections work, but it also lays the grounds to jump-start their political mobilisation. Women poll officers have the chance of increasing their political information and political knowledge, and hence gather the necessary components positively related to political mobilisation. Therefore, the effect is arguably not circumscribed to the election when they serve, but it rather sticks with them and travels to subsequent elections.

We also dig into whether the extent of the reduction of the gender turnout gap is conditional on different contextual factors. First, we expect the effects of being a polling officer to be larger for women in contexts of greater political mobilisation (Corder and Wolbrecht Reference Corder and Wolbrecht2016; Morgan-Collins Reference Morgan-Collins2023; Skorge Reference Skorge2021; Teele Reference Teele2023). Even when there are no institutional differences across geographical units in terms of the electoral system in place, the sociodemographic composition of different localities can lead some parties to be more active in certain places. Hence, the individual experience of being an officer can be reinforced when there are active organisations interested in bringing in women to the ballots to secure their support. In other words, once women polling officers become politically activated, their mobilisation capacity is likely to be higher.

Second, we also expect that the presence of women as poll officers can have positive spillovers, not only for their turnout but also for their peers. Women poll officers might have—unintentionally—become role models for other women (Wolbrecht and Campbell Reference Wolbrecht and Campbell2007). The presence of women involved in politics can encourage other women to participate in politics in close social networks (Goyal Reference Goyal2023). Along the lines of Morgan-Collins and Natusch (Reference Morgan-Collins and Natusch2022), we would expect that networks of women should contribute to enhancing the political participation of women. Networks should be closer in smaller and tighter communities, and we should expect that women poll officers to explain their experience to other women, thereby contributing to their political socialisation. In localities where no woman assumed poll officer duties, the spread of the information of how an election is conducted was less likely to have been a topic for discussion among women. Thus, the larger the number of women poll officers, the more likely that other women were exposed to political information through other women. Through an informational channel, female turnout should be higher among all women in localities where women served as poll officers.

Female Enfranchisement in Spain

In a situation that resonates elsewhere, female suffrage in Spain was only approved after very heated debates, especially within the left. Some left-wing MPs feared women would be manipulated by priests or their husbands to vote conservative (Martín i Berbois Reference Martín i Berbois2013). Two leftist women led the anti- vs. pro-female suffrage debate in Congress: Clara Campoamor, in favour of female suffrage on the basis of equality of rights, and Victoria Kent, who stood against the suffrage because, she argued, confessor priests were still the main intellectual influence for most women (Pujol-Soliano Reference Pujol-Soliano2018). The female suffrage bill eventually passed in October 1931 with 161 votes in favour and 121 against and women would vote for the first time in legislative elections in November 1933Footnote ² and they did it again in February 1936, before the Spanish Civil War (Fig. 1).

Figure 1. Timeline of the main historical events.

Female enfranchisement was granted despite the absence of a largely organized suffragette movement (Teele Reference Teele2018). The widespread fears among left-wing voters of larger proportions of women being more conservative were reinforced after right-wing parties won the first elections in which women could vote in 1933. Yet, recent works have shown that this was largely a belief rather than an empirical pattern (Rodon Reference Rodonn.d.), also present in other contexts (Teele Reference Teele2024). Later, under larger turnout rates and also under full universal suffrage a left-wing coalition won the 1936 election.

Elections and Polling Officers

Voting rules in Spain at that time were the following. Each locality was composed of one or more census tracts, according to the number of registered voters. All voters in a census tract cast their vote in the same polling station —usually located in a public building, for example, a school —, where citizens could find a ballot box and rooms or cabins to maintain the secrecy of the ballot (Vilanova Reference Vilanova1986). Each ballot box had to be supervised by three individuals randomly selected among all literate registered voters in the census tract. One individual was appointed as the presiding officer and the other two as supervising officers, all of them performing very similar tasks on election day. Three extra individuals —one for each position —were also drafted as backups to replace the main members who did not show up in election day, for example, due to illness.

The selection of these members was conducted through a lottery,Footnote ³ and female enfranchisement did not only grant the right to vote to women but also their right to be selected as poll officers. Despite the rules and requirements to become a poll officer were not de facto gender neutral,Footnote ⁴ for the first time in Spanish history, in the 1933 legislative election both (literate) women and men were included in the list of potential poll officers. The exact procedure to select the polling officers was as follows.

First, local civil servants compiled a list of all literate individuals in each census tract. Second, literate individuals were classified into three different groups: (a) individuals with ‘academic or professional titles’,Footnote ⁵ (b) ‘major taxpayers’,Footnote ⁶ and (c) the rest of literate individuals —the majority. Then, all individuals were alphabetically ordered (by surname) within each group. Third, two letters were chosen to create conglomerates of individuals. The three first individuals whose surname started with these letters (or a subsequent if there were not enough people) in each of the three groups were eligible. To ensure rotation over time, the letters chosen changed every two years; if in one year the lists for the draw used letters A and U, two years later the draw would employ letters C and S. Fourth, the oldest person among the individuals in the first conglomerate —e.g., surname starting with A —in all three lists was appointed as president, and the oldest in the second conglomerate —e.g., surname starting in U —in the three lists was appointed alternate president.Footnote ⁷ Finally, each time an election was called, the remaining two officers —and their alternates —were selected among those individuals who belonged to different group lists. For instance, if the president was a major taxpayer, one officer would be selected from the ‘academic’ group and the other from the ‘only literate’ group. Table B.2 in the Supplementary Information (SI) shows a naive example of a randomised selection of polling officers. Once the lottery had been conducted, the electoral commission made the decision public by posting the list of names and their responsibilities on the City Hall noticeboard or on other public buildings, by personally notifying polling officers —sending them a letter —and by publishing their names in the official gazettes.

Data

We draw our data focusing on poll officers from the province of Girona. We chose this province for three reasons. First, archives in Girona contain—almost completely—all individual-level electoral records for the 1930s elections; this means that it is possible to identify whether an individual living in this province did turn out to vote in any elections in the period we are interested. Second, we can compare our data to individual-level real voting behaviour data for a randomly selected sample of individuals in this province from Muñoz et al. (in preparation). Third, fine-grained municipality-level data are available from multiple sources and, due to the diversity of economic activities and geographic characteristics, the analyses in the Girona province are likely comparable to other regions.Footnote ⁸

Sampling

Our sampling procedure was as follows. We first collected individual information on all drafted polling officers in the Girona province that were reported in the official gazettes the weeks before the 1933 elections (an example in Figure A.1, SI). We distinguish from these lists the individuals who were drafted into acting or alternate roles. Acting members were those appointed to serve on election day, while alternates were only backups in case an acting member was unable to carry out his/her responsibilities. Unfortunately, archival records do not allow us to disentangle to which group —‘taxpayers’, ‘academic’, or ‘literate’ —each individual belonged to. We then selected all census tracts in which at least one woman was drafted as a poll officer, either acting or alternate, to analyse the whole population of women poll officers in 1933 in the Girona province. Therefore, we capture the entire universe of women poll officers in the province. Finally, we randomly selected an equivalent number of census tracts in which no woman was drafted to avoid any possible source of bias, as census tracts with women as polling officers might differ from other census tracts. Unfortunately, a map of the census tracts is not available, but Figure 2 maps all those municipalities from which we collected poll officers’ individual-level turnout data in our sample. Polygons in dark grey reflect municipalities where at least one woman was selected as an acting or alternate polling officer in one of its precincts, localities in middle gray, are those municipalities where we collected data for a random sample of the precincts, and the light grey municipalities are not in our study. The precincts we analyse belong to a diverse typology of localities, with variations in their geographical and socioeconomic characteristics.

Figure 2. Map of Girona’s municipalities and the sample.

Note: the map shows all municipalities in the province of Girona in 1933 as precinct maps do not exist. The poll officers’ data we analyze comes from 84 precincts that were located in the municipalities shaded in mid and darker grey on the map; poll officers’ data from white municipalities was not analyzed.

As displayed in Table 1, we exploit individual-level data of drafted poll station officers in 84 census tracts (out of 487 in the province).

Table 1. Census tracts data (1933)

In fifty-five census tracts, at least one woman was drafted and in twenty-nine no women were drafted. There is balance in multiple contextual covariates across all census tracts —regardless of the presence of women as poll officers (Table C.1, SI). This is important because a potential threat to validity is that women were only selected, for example, in richer municipalities. This is not the case, and hence we will be comparing poll officers from similar contextual environments.

As for the gender composition, around 18.5 per cent of the individuals in our sample are women. This low figure is likely to be a consequence of the lower literacy, education, and income levels of women, which made them less likely to be drafted as poll officers. We are aware that women were less likely to be elected in groups 1 (professionals) and 2 (major taxpayers), yet when looking at the occupational categories of drafted women we observe that around 20% were classified as “owners” or “professionals”. Moreover, the existence of 5 census tracts with more than one woman reflects that we are also capturing some women in those two groups. Although our results might not be extensible to all enfranchised women —most notably to illiterate women —all literate women in our sample should be comparable in their observed and unobserved characteristics, except for the fact that some of them acted as polling officers. In fact, our scenario represents a hard case to test our expectations because enfranchised literate women who were drafted should have on average more individual resources and hence be more likely to vote than illiterate women; thus, we should expect a lower turnout gender gap among these women.

Aggregate-level data

We complement individual-level data with contextual-level information. We first include a group of indicators capturing the sociodemographic characteristics of the municipality from official censuses: logged population, literacy rates, and female marital rates. A set of variables considers the role of political contextual covariates: aggregate turnout and female share of turnout in 1933 at the census tract level. Unfortunately, separate turnout is not available in official statistics and the only way to capture turnout by sex has been to manually count the number of women in the voter lists in all census tracts in our sample.Footnote ¹¹ Finally, we measured the feminization of poll boards by counting the number of women effectively serving as poll officers on election day.

Other indicators capture potential political mobilisation dynamics at the municipality level. First, two indicators identify whether a municipality had a local political branch (1) or not (0). Among the left, we consider anarchist organisations (Cucó Giner Reference Cucó Giner1970) and ateneus —cultural and political socialisation organisations (Amat et al. Reference Amat, Boix, Muñoz and Rodon2020). Second, we also calculated whether the Lliga —the main right-wing conservative party in Catalonia —organized a political rally in the municipality in 1933 and the percentage of Catholic associations in each municipality using original archival data on associations.Footnote ¹² We also consider the levels of political competition by including in the analysis either the levels of support to the left and/or right parties, or the political competitiveness.Footnote ¹³

We finally take into account the role of economic development at the locality level using fiscal data in the Girona province in 1923 (Vall-Prat Reference Vall-Prat2023) and an indicator of population concentration that considers the number of inhabited population nuclei and the number of inhabitants in each of them (Esteve Palós Reference Palós2003). Table 2 displays descriptive statistics of the main variables of interest, showing balance in all covariates when comparing individuals who were drafted as alternates and those drafted as acting.

Table 2. Summary statistics

Empirical Strategy

As described in section ‘Elections and Polling Officers’, we rely on a historical natural experiment. The basic principle of our case is that treated individuals —acting poll officers —and untreated ones —alternate polling officers —should have in expectation similar characteristics except for their treatment status (Gerber and Green Reference Gerber and Green2012). The comparison between the two groups allows us to be close to the Average Treatment Effect on the Treated. Yet, because of non-compliance, the average effect on the outcome depends on being assigned to treatment —acting as a polling officer —rather than experiencing the treatment. Thus with this design, we are closer to the Intent-To-Treat effect (ITT) (Gerber and Green Reference Gerber and Green2012).

We classify individuals in our database in three different categories as a function of their exposure to treatment: ‘alternates’, ‘poll officers’, and ‘defiers’ (see Table 3). The ‘alternates’ category corresponds to those individuals who were drafted as backups and they did not have to stay in the polling station board during the election day. ‘Poll officers’ are those individuals who effectively performed the tasks as members of the polling board and spent the day supervising the elections; they were the individuals effectively treated. Finally, ‘defiers’ are those individuals who were drafted as acting but did not assume their duties as poll officers. Table 3 shows compliance status by sex.Footnote ¹⁴

Table 3. Treatment compliance (sample)

Since we are interested in whether random selection contributed to making women more prone to vote —reducing the gender gap —in the subsequent election after serving as poll officer, we initially focus on the interaction between the sex of poll officers and their exposure to treatment. Hence, we rely on the following estimation:

(1) $$Vote{36_i} = {\beta _1}Pol{l_{it-1}} + {\beta _2}Woma{n_i} + {\beta _3}Pol{l_{it-1}} \times Woman + X_i^\prime + \zeta _c^\prime + {\epsilon _{ic}}$$

The dependent variable $(Vote{36_i})$ reflects whether an individual (i) voted (1) or not (0) in 1936, the second legislative election after female enfranchisement. Our main independent variables are two binary indicators capturing a) whether individuals effectively serve as poll officers in the 1933 election $(Pol{l_{it-1}})$ and b) if an individual is a man (0) or a woman (1) $(Woma{n_i})$ . ${\beta _1}$ shows whether men serving as poll officers in 1933 were more likely to vote in 1936, ${\beta _2}$ depicts the turnout differential between non-serving women and non-serving men, and the ${\beta _3}$ reflects differences between women poll officers and women who did not serve as poll officers. We also include in the model a set of individual-level $(X_i^\prime )$ and contextual $(\zeta _c^\prime )$ covariates. An important covariate is whether an individual voted in the 1933 election, which captures existing levels of voting habits (Dinas Reference Dinas2012; Green and Shachar Reference Green and Shachar2000) and estimates coefficients net of individual baseline levels to participate in elections. We additionally present models in which we replace the $Pol{l_{it-1}}$ covariate with the categorical variable distinguishing ‘alternate’, ‘poll officers’, and ‘defiers’.Footnote ¹⁵

To deal with non-compliance, we implement an instrumental variable (IV) approach using the original draft status in official Gazettes (acting vs. alternate) as the instrument in a 2SLS model. The instrument enables us to estimate the likelihood of an individual being a poll officer in 1933 (1st stage dependent variable) as a function of whether s/he was drafted as acting or alternate and other individual-level covariates —including sex to capture that there might be different gendered probabilities of self-selection into compliance. The first-stage equation is the following:

(2) $$Pol{l_{it-1}} = {\gamma _1}Draf{t_{it-1}} + {\gamma _2}Woma{n_i} + X_i^\prime + \zeta _c^\prime + {\epsilon_{ic}}$$

In the second stage, we use the predicted values of becoming an acting poll officer $(\widehat {Pol{l_{it-1}}})$ as our main independent variable in the second stage.

(3) $$Vote{36_i} = {\beta _1}Woma{n_i} + {\beta _2}\widehat {Pol{l_{it-1}}} + {\beta _3}Woman \times \widehat {Pol{l_{it-1}}} + X_i^\prime + \zeta _c^\prime + {\epsilon_{ic}}$$

We employ the randomised allocation between acting and alternate positions to estimate the ITT effect, net of other possible confounders leading to self-selection into treatment.

We also use data from Muñoz et al. (in preparation) to create a control group as similar as possible to all the drafted individuals in our original sample using nearest matching methods and taking into account individual and contextual characteristics. Thus, all individuals drafted as poll officers were paired with a similar control unit.Footnote ¹⁶

The final part of the empirical analysis delves into the reasons for the closing of the gender turnout gap by exploring whether contextual characteristics trigger heterogeneous effects of the treatment. These models are based on triple interactions between gender, the predicted likelihood to become a poll officers, and contextual characteristics.

Instrumental Variable

As described before, a potential threat to validity is that the effects shown in Figure 3 might be biased because of non-compliance with treatment assignment. We isolate variation in the treatment status related to self-selection into treatment through an IV approach in Table 4.

Table 4. Instrumental Variable Analysis

*p < 0.1, **p < 0.05, ***p < 0.01

Comarca and Occupation FE.

Standard errors clustered at the municipality level.

The first-stage results show that sortition in an acting role is the main variable driving exposure to treatment, as it increases by 66 percentage points the likelihood of effectively becoming a poll officer. Crucially, there are no gendered differences on the likelihood of becoming poll officers, giving us further confidence that we are estimating the effect of becoming a poll officer—and not a gendered self-selection effect. In the second stage, the main effect is in line with our expectations (see Figure F.1, SI). Women not predicted to become poll officers in 1933 were 30 percentage points less likely to vote in 1936 than men, resulting in a statistically significant gender gap.Footnote ¹⁸ In turn, women whose likelihood of becoming polling officers in 1933 was high were more likely to participate in 1936. Moreover, given that the ‘Pred. Poll Officer’ coefficient is not statistically significant and the magnitude of the interactive coefficient is as high as the coefficient for ‘Woman’, we can assert that being a poll officer contributed to closing the gender turnout gap.Footnote ¹⁹ The effect holds even when controlling for turnout in 1933.

Robustness: Matching and Placebo

To show the robustness of our findings we also implemented matching techniques to create an equivalent control group of non-drafted individuals. This test allows us to assess whether being drafted is driving our results. Individuals in the control sample should behave as individuals in the ‘Draft’ category because they share individual and contextual characteristics and they never became poll officers. Figure 4 shows that this intuition is partially confirmed (see SI Table G.3). While men in the control group were less likely to vote than drafted men, women in the control group had similar predicted levels of turnout than women in the drafted group. Moreover, the gender turnout gap exists in both categories, albeit smaller in the control group.Footnote ²⁰

Figure 4. Predicted turnout in 1936, including control group (matching).

Note: Regular OLS estimates (not an IV strategy), see Model 5 in Table G.3 (SI).

A placebo test (SI section G.3) estimating whether being elected as a polling officer affected turnout in pre-treatment elections (1931) confirms that being a poll officer in 1933 was not associated with larger turnout rates before the treatment.Footnote ²¹ This strengthens the validity of the previous findings.

Draft vs. Serving

To better understand how being drafted or serving as a poll officer changed the levels of turnout, we think it is useful to observe the evolution over time of the turnout across all the groups in our data. In Figure G.2 (SI) we observe that the gender turnout gap was large in the control group in 1933 but narrowed in 1936 as a consequence of the high polarisation in the 1936 election (Vilanova Reference Vilanova1986). Among alternates, the gender turnout gap was present both in 1933 and 1936, but being drafted as an alternate had short-term effects and increased turnout of women vis-à-vis the control group in 1933. While for men being alternate had a positive effect on turnout in both 1933 and 1936, alternate women were more likely to participate in 1933 but not in 1936. Being drafted, thus, had enduring effects on turnout for men but not for women Finally, almost all poll officers voted in 1933 and in 1936 turnout among them slightly decreased. However, while men poll officers voted in similar numbers as alternate men, women poll officers voted in larger proportions than women in the control and alternate groups. All in all, we find some short-term effects among drafted women but the effects were larger and more durable among women effectively acting as poll officers. This points out that being drafted can be sufficient to boost turnout in the long-term for individuals with previous electoral experience —men —, while for recently enfranchised voters a more intense treatment —serving as poll officer —was necessary for their electoral activation.

Mobilisation

Once female enfranchisement was enshrined, political parties in Catalonia —both left-wing (ERC) and right-wing (Lliga) —adapted their strategies, created women’s sections, directly appealed for the women’s vote, and included female speakers in their meetings to mobilise a new pool of voters (Martín i Berbois Reference Martín i Berbois2013). Beyond parties, two other organisations could also influence women’s turnout. On the one hand, anarchist organisations played a crucial role in demobilising workers in the 1933 election to punish the left-wing government and in mobilising them in 1936 in order to prevent the rule of reactionary forces (Amat et al. Reference Amat, Boix, Muñoz and Rodon2020). On the other hand, the Catholic Church —at odds with the Republican governments’ secular policies —disagreed with the new role given to women in the public sphere and used priests’ recommendations to influence political mobilisation, especially among women. Thus, we argue that in contexts where these organisations were strong, they could have influenced the participation of women poll officers. Here, it is vital to note that we control for the specific levels of political competition, as previous literature (Morgan-Collins Reference Feitosa, Stiers and Dassonneville2023; Teele Reference Teele2023) has noted that, beyond political mobilisation, competitiveness can also spur turnout among recently enfranchised women.

Figure 5 shows to what extent mobilisation patterns influenced the 1933 drafted poll officers turnout in 1936.Footnote ²² The estimates point out that women poll officers were not more likely to vote in contexts where political mobilisation was low, but when there were mobilisation efforts differences arose. For us the most indicative differences are those that can be detected among men and women who were predicted to be poll officers under different contextual scenarios. For instance, the top-left panel shows that women predicted to become poll officers were more susceptible to vote in 1936 if there was a local anarchist organisation (p-value 0.0265). In other words, women’s turnout was substantially higher when they acted as polling officers and when there was a political organization that tried to coordinate their votes compared to women predicted to be poll officers who lived in localities where anarchist mobilisation efforts were absent.

Figure 5. Heterogeneous effects. Mobilisation resources.

Note: We use the IVs second stage (models in Table G.5, SI) and we categorize individuals as predicted poll officers if their predicted value to become poll officers is above 0.5 –according to the IVs first stage.

We observe similar predictions when focusing on the mobilisation capacity of the right-wing Lliga —top-right panel —and on the presence of left-wing Ateneus —bottom-left panel. In all these cases, women’s turnout among predicted poll officers remains low(er) when mobilisation capacities are low, and rises when these organisations were present in the municipalities. Yet, when we explore the religious dimension we observe a slightly different story. More concretely, we see that where religiosity was more dominant, women likely to be poll officers in 1933 were less prone to participate in 1936. In municipalities with more religious social capital, there were still hurdles to female political participation. One potential explanation is that women voted the first time they were allowed to, but then they were relegated to the previous political role. Another one is that left-wing women in religious contexts might have been more disappointed than the average because of the victory of the right, thus being more demobilised.Footnote ²³ Overall, these variations in women’s turnout given specific mobilisation contexts is related to the closing of the gender gap in turnout, since men’s turnout seems to be unrelated to contextual mobilisation characteristics.

Peer Effects

In line with our expectations, we should also observe peer effects and observe higher female turnout among the general population in places where there had been women poll officers. In places where women became poll workers, these women learned about politics, became a sort of “role models” for their neighbouring women, and should have been more influential regarding politics among their peers. This should have prompted female turnout. In turn, in places where no women assumed poll officer duties, women should have felt less compelled to vote in 1936.

Although we cannot directly test the diffusion of information, we can assume that for those women who became poll officers in 1933, this represented a clear jump-start in their political knowledge that they could have shared through their networks —as shown by Morgan-Collins and Natusch (Reference Morgan-Collins and Natusch2022) networks were important for women’s mobilisation. We test this using two different strategies. On the one hand, we simply distinguish between those census tracts where there was at least one woman poll officer and all-male poll boards. Figure 6, shows how in places where there were no women in the polling stations in 1933, women were still less likely to turn out to vote in 1936. Nonetheless, when there was a woman polling officer, the gender turnout gap in the control group closed.The presence of a woman poll worker motivated political involvement of their peers. Figure G.4 (SI) displays the robustness test of this argument by showing that the larger the number of women in the polling station, the lower the gender turnout gap, especially in the control group.

Figure 6. Peer effects through the presence of women poll officers in the past.

Note: Estimates derived from Table G.6 (SI).

On the other hand, we should expect that political information travels faster in places where people are similar or live close to one another (Morgan-Collins and Natusch Reference Morgan-Collins and Natusch2022; Sinclair Reference Sinclair2012). The information on who was acting as a poll officer and their experience should have been shared and diffused more widely in places where people lived close to each other. In these settings it was more likely that other people could observe changes in gender roles and news about women’s presence in polling stations could politically mobilise other women.

To capture the idea of social proximity, we employ an index of population concentration from Esteve Palós (Reference Palós2003), which is a proxy for population density, i.e., whether all inhabitants live close one to another. The index takes a value of 1 if all inhabitants in a locality live in the same nucleus and it would take a value of 0 if each inhabitant lived isolated in different nuclei. Given that rural towns can be concentrated or dispersed and largely inhabited localities could also have their population dispersed, this indicator captures the proximity of inhabitants to one another and is different from the classic rural-urban distinction.Footnote ²⁴ Figure 7 shows that population concentration —controlling for the total population —was an important factor contributing to reducing the gender turnout gap, but only in those localities where there was a woman poll officer. Differences between men’s and women’s turnout decreased when the population concentration was larger and there was a woman poll officer in 1933. In turn, in all-male polling stations, gender gaps in turnout persisted regardless of the level of population concentration.Footnote ²⁵

Figure 7. Heterogeneous effects. Social proximity (90 per cent CI).

Note: Estimates from Model 1 in Table G.7 (SI).

Given the limited sample size, the results presented in this section regarding mobilization and peer effects need to be taken with caution. Nonetheless, the results pinpoint that there are relevant contextual factors that can shape the effect of serving as a poll officer.