Gender quotas and the quality of politicians by 六 Audinga Baltrunaite ** Piera Bello *** Alessandra Casarico *** Paola Profeta

Documento de Trabajo 2013-11

October 2013

* IIES, Stockholm University.

** Università della Svizzera Italiana.

*** Università Bocconi and CESifo.

Los Documentos de Trabajo se distribuyen gratuitamente a las Universidades e Instituciones de Investigación que lo solicitan. No obstante están disponibles en texto completo a través de Internet: http://www.fedea.es. These Working Paper are distributed free of charge to University Department and other Research Centres. They are also available through Internet: http://www.fedea.es. ISSN:1696-750

Gender quotas and the quality of politicians^∗

Audinga Baltrunaite¹, Piera Bello², Alessandra Casarico³, and Paola Profeta³

¹IIES, Stockholm University ²Università della Svizzera Italiana ³Università Bocconi and CESifo

October 26, 2013

Abstract

We analyze the efects of the introduction of gender quotas in candidate lists on the quality of elected politicians, as measured by the average number of years of education. We consider an Italian law which introduced gender quotas in local elections in 1993, and was abolished in 1995. As not all municipalities went through elections during this period, we identify two groups of municipalities and use a Diference in diferences estimation. We find that gender quotas are associated with an increase in the quality of elected politicians, with the efect ranging from 0.12 to 0.24 years of education. This efect is due not only to the higher number of elected women, who are on average more educated than men, but also to the lower number of low-educated elected men. The positive efect on quality is confirmed when we measure the latter with alternative indicators, it persists in the long run and it is robust to controlling for political ideology and political competition.

Keywords: gender quotas, municipal elections, Diference in diferences, average years of education

JEL classification codes: D72, J45

^∗We wish to thank for their comments and suggestions Toke Aidt, Manuel Bagues, Thomas Braendle, Pamela Campa, Maria De Paola, Olle Folke, Bruno Frey, Vincenzo Galasso, Simone Ghislandi, Osea Giuntella, Wojciech Kopczuk, Fabrizio Mazzonna, Tommaso Nannicini, Johanna Rickne and seminar participants at the Dondena Research Meeting, at the Workshop on gender at Università Bocconi, at the CESifo workshop on Political Economy, at the Alp-Pop Conference, at the University of St.Gallen, at the 26th Annual Conference of the European Society for Population Economics, at the 21st Silvaplana Workshop in Political Economy and at Columbia University. All remaining errors are ours. Paola Profeta thanks CEPRA and IDEP at University of Lugano for hospitality during her visiting academic year 2012-2013 and acknowledges Project No. 87/12 funded by the Spanish Institute for Women (Ministry of Health, Social Services and Equality).

1 Introduction

This paper studies the efects of the introduction of gender quotas in candidate lists on the quality of elected politicians. Some countries have recently introduced gender quotas in order to achieve gender balance in political institutions, in which women are largely under-represented. In fact, in 2010 women represented only 19.8% of members of Parliaments in the world. In the European Parliament women are 35% of the members. When we consider the diferent European countries, in the Lower (or Single) House around 40% of members are female in Belgium, Denmark, Finland, Norway, Sweden and only 22% in Italy. The figures for the Upper House are not very diferent. Even in local governments, which are typically considered a first step for politicians’ career, and possibly easier to access for women, in many countries the female presence is much lower than their share in the population would predict. In Italy, for instance, women represent 11% of mayors, 20% of members of the municipal councils and 21% of the members of executive committees in municipalities.

There are many arguments in favor of increasing female presence in politics. First, given that women represent half of the population, equal participation in political decision-making contributes to legitimizing the democratic body (Stevens, 2007). Second, as women’s needs may be diferent from the men’s ones, a larger female presence may be justified as a way to redirect policy implementation and public spending towards specific areas (Funk and Gathmann, 2010; Rehavi, 2007). Additionally, female political leadership may also be beneficial in itself, if women have diferent behavior and practices which have a positive impact on the quality of institutions and organizations (Epstein et al., 2005).

There are, however, also arguments against the introduction of afirmative action measures, such as gender quotas. Some of them are not specific to politics but also apply to business. One of the main claims is that female under-representation is just the result of individual choices, especially those related to fertility and motherhood. Thus, by equalizing outcomes rather than opportunities, afirmative action policies risk to promote less qualified individuals who will very likely perform poorly: gender quotas may increase equity at the expense of eficiency (Holzer and Neumark, 2000). More specifically, as gender quotas do not necessarily obey to meritocracy, the average quality of those responsible for decision-making may decrease. Since the quality of politicians is crucial for good governance and consequently for performance, this consideration may be used against the introduction of gender quotas in politics.

In this paper we reverse the argument that gender quotas may have an adverse efect on the quality of selected representatives. On the contrary, focusing on politics, and measuring the quality of politicians primarily by their years of education, we show that gender quotas may increase the quality of elected politicians.

We perform an empirical analysis focused on the temporary adoption of gender quotas in candidate lists in Italy. Gender quotas were introduced in Italy in 1993 by the Law No. 81, and were abolished in 1995 by the Constitutional Court. The law imposed that in candidates’ lists neither sex could represent more than 2/3 of the total candidates. Given that not all municipalities voted during the years 1993-1995 when the law was in force, we can identify two groups of municipalities, one afected by the quota and another one never afected. This allows us to use a Diference in diferences approach to investigate the efect of gender quotas on the quality of politicians. Given the reasonable assumption that there is a positive relationship between educational attainment and private sector’s success, and between market and political skills (Galasso and Nannicini, 2011), we measure the quality of politicians in terms of human capita (see also Kotakorpi and Poutvaara, 2011). We thus compare the change in the average education of politicians across the two groups of municipalities before and after the policy is enforced. In this way we disentangle the efect of the quotas on politicians’ quality from the temporal trend, which we assume to be common to the two groups. We find that, due to the introduction of the reform, the average education of elected politicians increased significantly more in municipalities afected by the policy. Namely, municipal councilors invested in their education around 2-3 months more in the treated group than in the control one. The efect is driven not only by the increase in the number of elected women, who are on average more educated than men, but also by the reduction in the number of low-educated elected men. Quality should therefore be an argument in favor rather than against the introduction of gender quotas. We propose alternative measures of the quality of politicians and several robustness checks to support our results.

The rest of the paper is organized as follows. Section 2 discusses the related literature. Section 3 describes the Italian institutional framework and the data. Sections 4 and 5 present the estimation strategy and the empirical results, respectively. Section 6 shows evidence in support of the validity of our identifying assumptions. Section 7 analyzes other outcome variables and performs some robustness checks. Finally, Section 8 concludes.

2 Related literature

Our paper crosses and combines two strands of the literature: the one on gender quotas and the one on the selection of politicians.

Starting from the first one, although there is a wide literature on diferent ways of implementing quotas and on the electoral success of women in legislative bodies (see Krook, 2009), analytical studies on the impact of gender quotas in politics are still few. Some papers focus on the reduction of gender stereotypes (e.g. Beaman et al., 2009). Others (e.g. De Paola et al. 2010; Campa, 2011 and Casas-Arce and Saiz, 2011) analyze the impact of gender quotas on female representation. De Paola et al. (2010) show that the introduction of gender quotas in local elections in Italy in the period 1993-1995 increased female presence and, by exposing voters to female leadership, broke down negative stereotypes, generating a higher percentage of elected women, even after termination of the policy. In this paper we study the same policy experiment, but focus on the quality of politicians rather than on the number of elected women. Folke and Rickne (2012) suggest that in contexts where political competition is weak, gender quotas may reduce barriers for women to get access to higher political ofices. Gender quotas may also have an impact on local policies. Women and men have diferent preferences both as voters (Edlund and Pande, 2002; Lott and Kenny, 1999; Aidt et al., 2006; Bertocchi, 2011) and as policy-makers. Women seem to prefer a diferent allocation of public funds, favoring projects that support female needs (Chattopadhyay and Duflo, 2004; Rehavi, 2007; Funk and Gathmann, 2010; Clots-Figueras, 2011) and that provide more public goods (Duflo and Topalova, 2004).

The literature also looks at why gender quotas may be introduced in politics. Maniquet et al. (2008) show that in single-member districts incumbent politicians may want to introduce gender quotas to increase the probability of running against a woman and of being reelected, given voters’ bias in favor of men candidates.¹

Our paper is also related to the growing literature on the selection and quality of politicians. Several contributions emphasize that the association between political competition and politicians’ quality is a crucial determinant of the quality of government. Besley and Preston (2007) show how the electoral contestability of a district, in terms of electoral bias in favor of one party, afects policy choices. As politicians’ quality is not an easily quantifiable concept, diferent measures have been proposed in the literature. Galasso and Nannicini (2011) measure quality by years of schooling, previous market income, and local government experience and study the efect of political competition on the quality of government. They find that members of the Italian Parliament with higher ex-ante quality are more likely to run in contestable districts and their subsequent performance is better. Quality is measured by education and occupational qualifications in Kotakorpi and Poutvaara (2011).² Using data on Finnish politicians, they find that higher salary attracts better quality female candidates, while no efect is detected for males. Focusing on US mayoral elections in the period 1950-2005, Ferreira and Gyourko (2011) find that female mayors have higher political skills than male, and thus have an advantage as incumbents over comparable male candidates.

Other works show that education is positively correlated with the quality of government and discuss the impact of the latter on economic development (Djankov et al. 2003, Glaeser et al. 2004, Fortunato and Panizza, 2011). Recent studies also show that the identity of leaders, and especially their education, matter for growth (Jones and Olken, 2005; Besley et al., 2011). Gagliarducci and Nannicini (2012) find that better paid politicians are more skilled individuals and that they size government expenditure down. Merlo et al. (2009) show that there is a negative association over time between the salary of Italian politicians and their quality.

¹Bagues and Esteve-Volart (2012) however provide evidence that challenges this view of the voters bias in favor of men candidates.

²These measures may in general raise an issue of representation, i.e. to what extent more qualified politicians are representative of electorate. However, if quality matters for voters, high quality politicians do not necessarily have incentives in contrast with the goal of appropriately representing their electorate.

To the best of our knowledge, the relationship between the introduction of a law on gender quotas and the quality of politicians has not yet been tested empirically. The only exception is Besley et al. (2012), who construct a new dataset on Swedish municipalities and test the impact of the imposition of gender quotas by the central social-democratic party on its municipality groups on politicians’ skills. There are however influential theoretical studies and experimental evidence. Julio and Tavares (2010) challenge the idea - as one could gather from Caselli and Morelli (2004) - that gender quotas in politics, by inducing women with lower opportunity cost on the private labor market to become candidates, may reduce the average quality of politicians. They argue that this reduction of quality is only a short-term efect, that dominates when the rewards from public ofice are low, or when they are high but women are significantly more discriminated in the political market than in the labor market. Otherwise, quotas may even increase the average quality. In the experimental literature, afirmative actions, especially gender quotas, have been associated with the participation of high performance women with no eficiency losses (Niederle et al., 2008). Having an enlarged pool of candidates is positive for the quality of selected individuals (see the review by Croson and Gneezy, 2009).

3 The institutional framework and the data

3.1 Italian municipalities and the Law 81/1993

There are approximately 8,100 municipalities in Italy. They vary significantly in terms of geographic, demographic and economic indicators. The municipal administration manages the registry of births and deaths, the registry of deeds, contracting for local roads and public works and, most importantly, social services. It is headed by a mayor, who is assisted by a legislative body, the municipal council (Consiglio Comunale), and an executive body, the executive committee (Giunta Comunale).

In 1993 a law concerning the electoral system for municipalities and provinces was approved. According to the Law 81/1993, neither sex could represent more than 2/3 of the total candidates in electoral lists for municipal councils.³ The quota system was introduced to balance the gender composition in representative institutions at local level. Namely, since usually the majority of the candidates are male, the law established that at least 1/3 of the positions in candidates’ lists had to be reserved to women. In case the list did not comply with the quota requirement, it was refused.

³For municipalities with less than 15,000 residents neither sex could represent more than 3/4 of the total candidates. As to the presentation of candidates’ lists, note that a party has to present a list that consists of at most as many candidates as the number of seats in the council and at least as large as 3/4 of the number of seats for municipalities up to 15,000 inhabitants, and 2/3 for larger ones. The number of seats depends on the size of the resident population.

The Law 81/1993 included other provisions besides gender quotas. Namely, it established that mayors were directly elected by their own constituents, whereas previously they had been appointed by the politicians elected to municipal councils. It also prescribed that in municipalities with less than 15,000 inhabitants mayors were elected according to a single ballot rule, whereas a dual ballot was mandatory in municipalities with more than 15,000 inhabitants.⁴ The Law also substantially increased mayors powers, as it allowed them to nominate their own executives from outside the elected council members, while previously their choice had been constrained to the pool of the elected politicians.⁵ In 1995 the provision regarding gender quotas was abolished by the Constitutional Court. The Judgment 422/1995 claimed that this provision was unconstitutional because in violation of the principle of equality between sexes. All the other reforms included in the Law were not modified.

As a result of the ruling, the provision on gender quotas was in place for a short period of time between March 25, 1993 and September 12, 1995. Local elections take place every five years and municipal governments cannot afect their schedule.⁶

Given this fixed term feature, not all the Italian municipalities were afected by the reform: some municipalities voted with gender quotas, and others never did so, as if the law had never been enacted. Thereby, we identify the former as the treatment group (where treatment is defined as being exposed to gender quotas) and the latter as the control group.⁷ The first group is composed of 7,643 municipalities, which voted at least once during the period when the law was in force; the second group consists of the rest of the municipalities, in line with the strategy used by De Paola et al. (2010).

⁴In single-ballot municipalities, the candidate who gets the relative majority in the single election is appointed to be the mayor. Under this scheme, each candidate for the mayor position can be backed by one list only, with a substantial victory bonus: the list supporting the winner gets 2/3 of the seats in the council, while the rest of the seats are assigned to the remaining lists according to a proportionality criterion. In dual-ballot localities instead each candidate can be backed by a number of lists and not just one, i.e. there is no direct link between lists and mayoral candidates. If a candidate obtains an absolute majority (i.e. more than 50% of the votes cast) he or she becomes the mayor; if no candidate wins an absolute majority, then those ranked first and second go to a second round, in which they can seek the support of lists whose candidates have been eliminated. After having appointed the mayor, the council is formed. If the lists supporting the winning candidate receive over 50% but less than 60% of the votes, then they obtain 60% of the seats in the Council; otherwise, seats are assigned according to a proportionality rule.

⁵Other changes involve the reduction in the maximum number of seats in municipal councils and the change in the format of the electoral ballot.

⁶In the period 1993-1999 the mandate was shortened to four years. Note also that elections take place only in a specified time window of a year. If the term of the municipal council expires after this period, elections are delayed. This explains why in some municipalities the term may be longer than 4 or 5 years depending on the election we are focusing on.

⁷We may have a mixing between the two groups if the electoral campaign is run right before the adoption or the abolition of gender quotas. Taking into account that electoral campaigns oficially last for 30 days and candidates’ lists must be presented on the first day of the campaign, no such mixing occurs in our sample since no municipalities voted during the 30 days after March 25, 1993 and in the

As only gender quotas were removed, while the other provisions introduced by Law 81/1993 remained in place, we can safely claim that gender quotas are the only diferent institutional feature between our treatment and control group municipalities. There are no reasons to suspect that the 1993 reform had a diferential impact on the two groups of municipalities other than the gender quotas’ provision.

3.2 The data and descriptive analysis

In our analysis we use administrative data provided by the Italian Ministry of the Interior on gender, education level and previous jobs of all politicians elected in Italian municipalities from 1985 to 2009.⁸ We use these data also to identify the political ideology of the majority in each municipal council and to build a measure of political competition. In addition, data regarding the size of the resident population over age 15, the employment rate (overall and female) and the education level in the municipality are taken from the 1991 and 2001 Italian Census of Population.⁹

To calculate the overall employment rate (Employment rate), the female employment rate (Female employment rate) and the education level defined as the share of university graduates (Local education level ) we use the resident population over age 15. We measure political ideology using dummies for the political leaning of the majority in the council.¹⁰ The variable Left-wing majority is a dummy for a left-wing parties’ majority; the variable Centre-right majority is a dummy for a center-right parties’ majority; the variable Civic list majority is a dummy for councils where the majority of members are politicians elected through civic lists and the variable Coalition majority is a dummy for councils where seats are shared equally by two or more parties with diferent ideology. The degree of political competition is measured by the diference between the share of seats of the winning majority and that of the second largest group.

Table 1 reports averages of these variables in treated and control municipalities in 1991. The statistics show that, while the two groups do not difer in terms of resident population size, control municipalities are more likely to be in the South and they have a lower overall and female employment rate. Also, they are less likely to be governed by a left-wing majority and show less competition in local politics. This is further discussed in Section 6.

30 days after September 12, 1995.

⁸There are a few observations with electoral date falling prior to 1985. Given that data collection started in 1985 and covered all Italian municipalities, some localities at this point of time were governed by municipal councils elected previously.

⁹In line with the methodology adopted by the Ministry of Interior in compiling data on local politicians in Italy, we use 1991 Census data for elections up to 1997, and 2001 Census data for election from 1998 onwards.

¹⁰Since Italian local elections are characterized by a large number of parties running for and winning the seats, in order to identify the majority we group parties according to their political leaning and then determine the majority in a given municipal council and also the remaining minority groups. Notice therefore that in a municipality with, for example, a left-wing majority we may often have councilors from several diferent left-leaning parties.

Table 2 shows descriptive statistics on the average years of education of the elected councilors in the two groups of municipalities, which we use as a proxy for the quality of politicians. The averages are calculated using data on the politicians elected in the last election before the adoption of the law (Before) and in the first election immediately after it (After ).¹¹ In Panels B and C of Table 2 we distinguish between female and male politicians.

[Table 2 here]

The statistics show that the education level of the elected councilors is on average higher in the control group, both in the Before and After periods. The temporal change is positive for both groups of municipalities, and it seems to be larger for the treatment group. All diferences are statistically significant to conventional levels. A similar pattern occurs when looking at the years of education of the elected male politicians. Interestingly, the elected female councilors have on average completed roughly two years of schooling more that the male ones for all the groups considered.¹² There is little evidence that the temporal change in the years of education of the elected females was positive.

We finally look at how female presence evolved in the two groups of municipalities before and after the adoption of gender quotas. The data reveal that in the post-reform period it increased in both groups of municipalities. The rise in municipalities that voted with gender quotas is more pronounced than in municipalities that did not. Since women have on average more years of education than men, this finding indicates one potential channel through which gender quotas may have afected the quality of the elected politicians.¹³

¹¹However, 86 municipalities voted twice during the period when the reform was in place. For them, we keep both observations after the introduction of quotas in order to take into consideration all the elections where gender quotas were applicable.

¹²Note that women have a higher level of education also when we look at the entire population and not just at elected politicians: for instance, in the Nineties the share of women holding a secondary school degree was between 6 and 12 percentage points higher than the share of men.

¹³The efects of gender quotas on female presence are documented in detail by De Paola et al. (2010). We have checked the impact of gender quotas on female presence using our diferent empirical strategy, based on the last election before the policy and the first election right after it, as described in Section 4. We also find the positive efects reported in De Paola et al. (2010). The results are shown in the Appendix, Table A.1.

4 Empirical Strategy

We analyze the impact of the reform on the average quality of politicians, as measured by years of education of the elected municipal councilors. We identify municipalities which were exposed to gender quotas as the treatment group and municipalities which never voted with gender quotas as the control group and use a Diference in diferences estimation. Namely, we run a number of municipality-level regressions and compare the change in the average education level of municipal councilors across the two groups of municipalities in elections right before and right after the introduction of the reform. We focus on the short term efects of the policy, as this makes us more confident in claiming that within a short time period there were no sharp changes in the institutional environment other than the reform which could have diferentially afected the quality of elected politicians.¹⁴ Finally, we also analyze the efects on the education level of the elected female and male politicians separately.

Since being exposed or not to gender quotas was induced by an exogenous change in the institutional setting, we consider the treatment and the control group status unrelated to other unobserved municipality characteristics afecting the dependent variable. Moreover, in our setting selection into treatment or control groups is especially unlikely, because the timing of voting cannot be manipulated by the municipality itself due to the fixed term of local governments.

The key identifying assumption in our analysis is the existence of parallel trends prior to the reform between the two groups of municipalities, which we state as follows:

Assumption 1. In the absence of the reform the diference in the outcome between the treatment and the control group would have remained the same.

\[E [ \varepsilon_ {i t} | T r e a t m e n t _ {i}, A f t e r _ {t}, X _ {i t} ] = 0\tag{1}\]

where is a dummy variable for municipalities afected by gender quotas; is a dummy variable for elections taking place after the introduction of the reform; is a vector of municipal characteristics and is an error term.

We point out that the timing of the elections in both groups of municipalities difers by construction. This implies that when comparing electoral results in treatment and control municipalities, we will consider diferent years for the two groups. Figure 1 shows that the fixed term feature induces a cyclical pattern in the election dates, with dis-proportionally many elections in some years, and very few in others. We argue that this is not a specific feature of elections around the time of the adoption of the reform, because such pattern is already visible as early as in the mid-Eighties. The diference in the timing of elections across municipalities can be explained by the fact that some local governments terminated before the end of the mandate in the past. These observations support our claim that the timing of the elections around the gender quotas’ adoption is not due to local politicians manipulating the date of elections with the specific goal to hold elections with or without gender quotas.

¹⁴This empirical strategy is diferent from De Paola et al. (2010), who look at the long run efects of the reform. We also exploit all the available data rather than two observations for each locality in the robustness analysis. Note that some municipalities voted twice during the period when the reform was in place. For them, we keep both observations after the introduction of quotas. Our results are broadly robust to the exclusion of these elections and can be made available upon request.

We also note that to verify Assumption 1 we would need at least two time observations for each locality before the introduction of gender quotas. Our data start in 1985. Since the mandate of local governments is five (four) years and the reform took place in 1993, we do not have enough data points for all municipalities to verify if the trends in the outcome variable are parallel in the two groups prior to the assignment of the treatment. However, we attempt to examine developments in the outcome variable for the two groups using the available data (1985-2009). Figure 2 plots the pattern in the average years of schooling of councilors for the two groups of municipalities over this period.¹⁵ The available data show that the pre-treatment changes in the outcome variable for the two groups of municipalities are fairly parallel. We further address the validity of this identifying assumption separately in Section 6.1.

Finally, due to staggered election dates, in our analysis we generally do not compare voting outcomes for the treatment and the control group in the same year. Roughly, “after reform elections” for the treatment group municipalities happen in the period March 1993 - September 1995, while “after reform elections” for the control group happen during the period October 1995 - July 1999. Hence, in our analysis we also implicitly assume that, in the absence of the reform, the treatment group would have had the same change in outcome as the control group during diferent time periods, as formally stated in the following.

Assumption 2.

\[\Delta_ {t} E \left(Y _ {0 i} | T _ {i} = 1, t = \tau_ {i} ^ {B e f o r e} - \tau_ {i} ^ {A f t e r}\right) = \Delta_ {t} E \left(Y _ {0 j} | T _ {j} = 0, t = \tau_ {j} ^ {B e f o r e} - \tau_ {j} ^ {A f t e r}\right)\tag{2}\]

¹⁵To calculate these averages we divide our data into five-year intervals in such a way that the 1993 reform falls strictly in between two intervals. Then, we calculate the average years of education of municipal councilors for each interval and each group. This procedure guarantees that we observe all municipalities voting at least once in each interval. Therefore, the first data point for the treatment and the control group is the average over elections taking place before 1988, the second data point - over elections taking place during the period 1988-1992, the third - 1993-1997, the fourth - 1998-2002, the fifth - 2003-2009. Given the diferent temporal structure of elections in treatment and control municipalities, this way of constructing the averages is the most appropriate, in our opinion. First, it is virtually meaningless to look at the averages in each calendar year to verify the presence of a parallel trend, because subsamples of the treatment and control municipalities observed in each calendar year are not representative of the composition of the two groups (see Figure 1). Second, comparing treatment and control municipalities according to the order with which elections take place , first election for each treatment municipality and each control municipality; second election for each treatment municipality and each control municipality, so on and so forth) would often entail comparing elections occurring at very distant points in time. Our method has the advantage of relying on the comparison between elections which are not very distant in time.

where stands for the variable T reatment as defined before; i and j are indeces for treatment and control group municipalities; and are the dates in which, respectively, treatment and control group municipalities vote.

In other words, we assume that the change in the (untreated) outcome in the control group, for instance, from year 1992 to year 1997 can be used as a proxy for the change in the untreated outcome in the treatment group, for instance, from year 1989 to year 1994. In Section 6.1 we attempt to compare elections closer in time for the two groups of municipalities in order to indirectly validate this assumption.

The baseline Diference in diferences estimator is of the form:

\[Y _ {i s t} = \alpha + \gamma T r e a t m e n t _ {i} + \varphi A f t e r _ {t} + \beta T r e a t m e n t _ {i} * A f t e r _ {t} + X _ {i t} \delta + \mu_ {s} + \pi_ {s t} + \varepsilon_ {i s t}\tag{3}\]

where is the outcome of interest and measures the average years of schooling of politicians elected in locality i, in province in year t; T reatment_i and are, respectively, dummy variables for municipalities afected by gender quotas and for elections taking place after the introduction of the reform, as already described. The variable T reatment allows us to control for the unobserved time-invariant characteristics that may difer across municipalities in the two groups, while the variable Af ter captures the temporal trend common to both groups. More precisely, the variable does also account for the impact of other provisions of the 1993 reform that afected all Italian municipalities. is the interaction term between the two dummies and measures the treatment efect of our interest. is a vector of control variables including the population size, its square, the share of university graduates and the share of employed resident population at the municipal level. We also include dummies for each province to account for the characteristics that are common to municipalities in the same province and are constant over time and indicate them with is the interaction between province dummies and the dummy Af ter accounting for time and province-varying shocks in politicians’ education. The inclusion of guarantees that time-varying unobserved characteristics common to municipalities within the same province do not drive our results. Put it diferently, the inclusion of province-specific shocks in the education of politicians allows to control that these do not contribute to the identification of our parameter of interest . Finally, is an error term.

5 Results

Table 3 presents the results of our main specification. Panel A focuses on the efect of the reform on the average years of education of all members of the municipal council; Panel B and Panel C look at female and male politicians, respectively, to investigate the existence of diferential efects according to the gender of the elected politician. In all panels, column 1 reports estimates of equation 3 without considering control variables, province and province-After dummies; column 2 includes province dummies; column 3 adds control variables and, finally, column 4 uses the full specification in equation 3. Standard errors are clustered at the province level.¹⁶

[Table 3 here]

In Table 3, Panel A, the coeficient on the T reatment variable is statistically significant and negative: this indicates that the members of the municipal councils have on average more years of education in non-gender-quota municipalities. However, the coeficient becomes virtually zero if we include province dummies, i.e., within provinces, variation in the average education of the elected politicians across treatment and control groups seems essentially random. Second, we observe a positive time trend in the level of education of the elected politicians. The Af ter coeficient is statistically significant and positive in columns 1 to 3, indicating an improvement of the quality of elected politicians in the After with respect to the Before elections. Most importantly, the coeficient on the interaction term Treatment*After is statistically significant and positive in all columns. The estimates suggest that the reform improved the average level of education of the elected municipal councilors. Namely, after the introduction of the reform the average education of municipal councilors in the treatment group municipalities increased by 0.12 to 0.24 years more than in the control group. This corresponds to approximately 2-3 additional months of education on average.¹⁷

Our estimates of the efect of the reform would be upward biased in case there were a catching up of the treatment group, given the lower level of education it starts from. On the contrary, estimates would be downward biased if the control group showed a more pronounced increasing trend in education. We note that the coeficient of interest is smaller in column 1 than in columns 2 to 4, where we gradually strengthen our identification. Given that an improved specification should allow to remove biases, this pattern is consistent with the presence of a downward bias in our initial estimate. Moreover, a catching-up efect for treatment municipalities on a hypothetically concave path for years of education is not very likely, given that the average years of education of the elected politicians are around 11 - 2 years less than those required for the completion of high school and 6 years less than those necessary for the achievement of a tertiary degree.

Focusing on female politicians, Table 3, Panel B, shows that the coeficient of interest is not significant. Thus, we cannot claim that in the treatment municipalities the education level of the elected women did evolve diferently after the reform compared to the control group.

The more interesting results come from Table 3, Panel C. The regressions on the education level of the elected male show that the quality of male politicians increased more in municipalities afected by the reform than in the control group. The results indicate that the education level of elected municipal councilors rose by 0.12-0.24 years more in the treatment group than in the control group. Therefore, our estimates in the baseline specification are mainly driven by the improvement in the years of education of the elected men rather than of the elected women.

¹⁶The results are broadly robust to clustering the standard errors at the municipal level and can be made available upon request.

¹⁷Results in Table 3 are broadly confirmed when we weight the average years of education in the municipal council by the number of municipal councilors elected in a given municipality.

To better qualify our result, we compute the variance of the years of education in municipal councils and use it as a dependent variable in our regressions. Table 4 shows that in the treated municipalities the variance increased less than in the control group municipalities. Given that the average level of education increases and the variance goes down, gender quotas guarantee a higher number of educated councilors. In other words, the increase in the average education of politicians does not hinge on the entry of a small number of individuals with an exceptionally high level of education.

[Table 4 here]

Next, we try to decompose the following changes that lead to our findings in Table 3: i) more women, who are on average more educated, are elected in municipalities with gender quotas, and ii) the pool of the elected men in municipalities which voted with quotas have on average more years of education. We argue that the latter outcome may be due to two diferent mechanisms: new female politicians replace the lowest skilled male politicians for a given distribution of male education and/or gender quotas lead to a change in the education distribution of males. To disentangle the diferent forces, we propose a simple numerical calculation. We focus on the efect of the quota estimated in the most restricted empirical specification in Table 3, Panel A, column 4, in which the point estimate of Treatment*After is 0.24. The gender quotas-driven increase in female presence estimated in Table A.1 is around 4%. Since female politicians have, on average, approximately 1.5 years more of education than an average politician in the period before the implementation of the policy (See Table 2, Panels A and B) having more women in municipal councils can contribute to the increase in the level of education by 0.06 (1.5 years* 4%). In other words, only one fourth of the improvement in the average quality of politicians may come from the fact that more women were elected, whereas the remaining part of the efect must come from the increase in elected men’s quality. This may result from the replacement of men in the lowest percentiles of the education distribution by new female entrants, for a given male education distribution, and/or from changes in the nomination procedure of male candidates. The first efect is maximum when female entrants substitute men in the four lowest percentiles in the male education distribution. In this case, we calculate its contribution to the total efect of gender quotas as the diference between the male average education level in the period before the policy was implemented and the average male education level in the distribution obtained by eliminating the four lowest education percentiles, weighted by the share of men in the municipal councils. This weighted diference is equal to 0.133.

We note that this value cannot explain the entire efect which must come from the increase in quality of the elected men. Therefore, we can attribute the remaining 0.05 (out of 0.24) increase in the average education of the elected politicians to the change in the nomination procedure of candidates, as suggested also in Besley et al. (2012). In other words, when gender quotas are in place, the pool of candidates who run for local elections is more qualified. This accounts for roughly one fourth of the efect.

To further complement the regression analysis in Table 3, we plot the kernel densities of the education levels of elected male politicians in the treatment group in the pretreatment period and in the post-treatment period.

[Figures 3a and 3b here]

Figure 3a shows that the entire distribution shifts to the right. More precisely, the largest diference between the two kernel densities lies in the right tail of the distribution.

We next replicate the same analysis for all elected politicians (males and females) in the treatment group in the pre-treatment and post-treatment periods. Note that the changes showed in Figure 3b are very similar to those in kernel densities showed for the only males sample. This is consistent with the gender quotas efect coming mainly through the election of more educated males.¹⁸

To conclude, our results show that the introduction of gender quotas increased the average education level of individuals who became politicians in the municipalities affected by the policy:¹⁹ the quality of the overall political body increased by more in the treatment group municipalities than in the control. This result is driven by the fact that fewer male politicians with a low level of education were elected.

6 Validity tests

In the following section we provide further evidence to support our identifying assumptions. To address the validity of Assumption 1 we present an alternative exercise based on municipal level covariates. Next, we show that our findings are not driven by systematic diferences between the two groups of municipalities in their geographic location and population size. Further, we perform a placebo test. Last, and more importantly, we show evidence supporting our result that it was gender quotas to increase the average level of education of municipal councilors. Concerning Assumption 2, we check the robustness of our results with respect to the time period of the elections.²⁰

¹⁸This exercise is limited to some extent, since we are not taking into consideration changes in the control group. Note that we do not show the education distribution for female councilors, since it is implicit from the comparison between Figures 3a and 3b that the change for women is virtually null. This is, in fact, consistent with the zero estimate of the efect of the reform in Table 3, Panel B.

¹⁹We also implemented the analysis using the share of politicians that have acquired at least a high school diploma as the dependent variable. The results do not change and are shown in the Appendix, Table A.2. We cannot use the share of elected politicians that acquired a tertiary degree because their presence is extremely low in our sample.

6.1 Selection into treatment and control group

To address the concerns regarding the selection into treatment and control groups, we start by providing further evidence on the existence of parallel trends between the two groups of municipalities. To do so, we focus on analyzing the behavior over time of our main covariates. Namely, we use the 1981, 1991 and 2001 Italian Census of Population data and check whether the employment rate, the local education level and the population follow a parallel development pattern. Figure 4 shows that, despite the diferences in absolute levels, treatment and control municipalities follow a parallel trend in terms of these variables.

[Figure 4 here]

To further limit concerns about selection into the treatment versus the control group we implement a test first introduced by Altonji et al (2005), adapted to the continuous case by Bellows and Miguel (2009). The results of the test allow us to conclude that the selection on unobservables should be twice as large as the selection on observables in order to completely eliminate the estimated efect of the gender quota. Since this is unlikely, we are more confident in causally interpreting our findings.

We further show that the efects of gender quotas do not depend on systematic diferences between treatment and control municipalities, for instance, on their geographical distribution or on their population size. In our sample 74% of the treatment group and 50% of the control group are formed by municipalities located in the Center-North. We thus replicate our main regression including a dummy for municipalities located in the South (and all the relevant interactions) and results are broadly consistent with the ones reported in Table 3. The interaction coeficient between our variable of interest and the dummy for southern municipalities is not significant, which suggests that our results do not depend on the systematic diference in geographic location between treatment and control group municipalities.

We also check the distribution of municipalities according to their population size.²¹ More precisely, the law prescribes seven population thresholds to determine the size of the municipal council. We check the distribution of treatment and control municipalities in each of the seven groups. We find that such distribution is similar to the overall distribution of municipalities across the treatment and the control group, with the exception of the largest municipalities (above 250,000 inhabitants) which all belong to the treatment group. When we run the regression excluding them, the results are unchanged.

²⁰All the results that are not reported in the paper are available upon request.

²¹Namely, we consider the seven population thresholds prescribed by the law which correspond to a diferent number of seats in the municipal council and check for the distribution of treatment and control municipalities in each of the seven groups.

6.2 Placebo Diference in diferences

In this section we replace the actual date of the introduction of gender quotas by a "placebo" date to show that our results are arguably not driven by a mere diference in trends between the two groups of municipalities. We keep the two groups of municipalities fixed and compare the change in the average education of the elected politicians in control and treatment group municipalities across elections during a five-year period before the "placebo reform" on January 1, 1988 and a five-year period after it.²² Note that due to the limited availability of pre-1985 data, we cannot observe the entire sample of Italian municipalities in both periods. Yet, this exercise contributes, at least in part, to the credibility of the identifying assumption of a parallel trend between the two groups. In Table 5 we replicate the same regression specifications as in Table 3, Panel A. Note that in this part of the analysis After is a dummy for elections after January 1, 1988.

[Table 5 here]

The results show that the placebo Treatment*After coeficient is, if anything, negative. Note also that in the specifications in columns 3 and 4 the coeficient is not significant.

6.3 Heterogenous efects of quotas

We provide a number of checks to indirectly support our findings that it was gender quotas to afect the quality of municipal councilors. In performing this analysis, we encounter a major dificulty: data on candidates are not collected in any municipality and therefore it is not possible to have any information on those who are running for ofice. Thus, we use data on elected politicians and rely on investigating the presence of heterogeneous efects to support our estimated impact of the quota. Namely, we expect the quotas to be efective in municipalities in which they are binding, and to have a zero efect in municipalities in which they are not binding. In order to determine whether the quotas were binding or not we propose alternative measures. We start from focusing on the share of female municipal councilors in the pre-treatment election and consider the policy to be binding in municipalities in which this share is below the quota requirement.²³

²²We choose this date to make sure that none of the municipalities vote with gender quotas during a five year period after the "placebo reform".

²³Since the data on candidates are not available, we implement this analysis on the actual number of elected politician, even though the quota was imposed on the candidate lists.

There are 355 municipalities out of 8069 where the share of women is larger or equal to the quota requirement. The majority of municipalities with high shares of women are in the treatment group - 338 out of 355. Therefore, we cannot use regression analysis to implement a falsification test showing that the quotas efect was indeed null in these municipalities. Instead, we drop from the sample the municipalities where the quota was already satisfied prior to the treatment. Since in these municipalities quotas should have a zero efect, we expect that the coeficient on Treatment*After remains virtually the same in this restricted sample. Table 6, Panel A shows the regressions that replicate Table 3, Panel A in a subsample of only the municipalities where the quota should have been efective. The coeficients are almost unchanged in columns 1-4 and, if anything, are slightly larger than in the regressions implemented in the complete sample. This is in line with the idea that the efect should be null in localities where the gender quotas are not binding.

[Table 6 here]

Next, we propose a more indirect indicator of binding quotas based on female labor market participation. Female labor market opportunities may capture the degree of gender bias in a given society. Therefore, we may expect gender quotas to be more binding in places where female employment is low and the bias is stronger.

In Table 6, Panel B, we interact our variable of interest Treatment*After with a measure of the absence of bias against women in a given municipality. We build a dummy variable which takes the value of 1 for municipalities whose female employment rate is above the 95th percentile of the distribution of female employment rates in the pre-treatment period as an indicator of the absence of such bias. The results show that in these municipalities the efect of the introduction of gender quotas on the quality of politicians is significantly smaller than in the others. In fact, a formal statistical test cannot reject the null hypothesis that the sum of the two coeficients is zero. In other words, in municipalities which do not sufer from high initial level of bias against women, gender quotas are less binding and increase the quality of elected politicians by a much smaller amount. The efect can be reasonably approximated to zero in these municipalities.

Last, we focus on the number of seats won by a single party in a given municipality as a measure of binding quotas. We expect the efect of the policy to be more pronounced in municipalities with larger majorities. We recall that candidates’ lists can at most consist of as many candidates as the number of seats in the council. If a given party wins more seats than the quota requires, it is forced to include in the council also the politicians of the other, minority, gender. We first select the municipalities where a single party’s list wins a large majority in the elections in the After period. There are 6842 municipalities out of 8069 in which the winning party gains a strong enough majority to make the quota binding. To investigate the heterogenous treatment efects, we then estimate a fully interacted model with a dummy for these municipalities. While the main coeficient of interest remains of the same order of magnitude as in Table 3, Panel A, we find no diferential efect in municipalities where the winning party gained a share of seats large enough to make the quota binding. The absence of diferential efects could be rationalized by the argument proposed by Besley et al. (2012): strong parties’ elites may cling to their power and prevent qualified candidates from gaining access to political institutions.

6.4 Staggered elections

To corroborate Assumption 2, we replicate the analysis in Table 3, Panel A, focusing on a subsample of elections ranging from 1989-1990 to 1994-1995 for the treatment group and from 1991-1992 to (the end of) 1995-1997 for the control group. This is an attempt to minimize the time distance between elections in the two groups. The results shown in Table 3, Panel A, are confirmed.

7 Other outcome variables and robustness analysis

When using education as the measure of politicians quality we follow the previous studies. To investigate whether our results are confirmed using alternative proxies for quality, we examine the efects of gender quotas on other outcomes: the skill-intensity of their previous occupation and their future political careers.²⁴ Next, we analyze the long-run efects of quotas and examine the robustness of the results to controlling for political variables.

7.1 Previous occupation

We have information on the previous occupation of the elected politicians and we use it to build a measure of politicians’ quality. More precisely, we consider all politicians who were engaged in entrepreneurial, professional or other skill-intensive activities²⁵ before obtaining a seat in a municipal council.²⁶ Such choice of the dependent variable is motivated by the potential correlation of labor market ability and skills in the political arena (Galasso and Nannicini, 2011).²⁷ A higher proportion of politicians whose previous occupation is skill-intensive would be interpreted as an indicator of a higher quality of the political body.

²⁴Gender quotas may also afect political participation. The impact of this gender quotas’ reform on turnout is examined by De Paola et al. (2012), who adopt our identification strategy.

²⁵The full list of occupations included in this category is in the Appendix, Table A.3.

²⁶Some papers analyze the presence of professional groups in politics as a potential determinant of the quality of governance: for instance, Braendle and Stutzer (2010) focus on the presence of public servants in German Laender Parliaments; Gehlbach et al. (2010) look at businessmen; Rosenson (2006) at lawyers.

²⁷The importance of the professional background for politicians’ quality and their performance is confirmed by Dreher et al. (2009), who show that the professional background of a head of government

First, we replicate the same regression specifications as in our main analysis. We report the estimated coeficients of the interaction term from the full specification of equation 3 in Table 7. Column 1 shows the results for all politicians, columns 2 and 3 - for women and for men, respectively. The efect of gender quotas is positive and significant and amounts to an additional 3 percentage point increase in the share of highskill politicians elected in the councils of treated municipalities. When we distinguish between male and female politicians, the coeficients in columns 2 and 3 suggest an increase of both male and female politicians’ quality. Note that we did not find an efect on the quality of female politicians when the latter is measured by the years of education. This diferent result is consistent with the fact that in Italy gender diferences in the labor market are much larger than the ones observed in education. Women with high education indeed are not perfectly matched into high-skilled jobs. Thus, measuring the quality of female politicians by years of schooling or by type of occupation may deliver diferent results.

[Table 7 here]

7.2 Political careers

The reelection probability is considered a good proxy of politicians’ quality, since more able individuals, at least in theory, are rewarded by votes from the electorate (see, for instance, Besley, 2005).²⁸ Thus, large shares of reelected politicians may signal their high quality. However, being repeatedly reelected can also have a diferent interpretation: strong political persistence may hamper the entrance of more competent politicians, who may threaten the survival of the existing, and often men-dominated, elites (Besley et al., 2012). This interpretation may be more salient in localities characterized by low levels of political competition, in which strong single party majorities prevail. Therefore, high reelection rates in such municipalities may be a signal of low political quality rather than the opposite.

We incorporate this idea in our analysis of the reelection rates of municipal councilors and check separately how the gender quotas’ efect varies according to the level of electoral competition in municipalities. The dependent variable is the share of municipal councilors that are re-elected in a subsequent municipal election. The measure of electoral competition is the diference between the number of seats of the winning majority and that of the second largest party group. The larger this diference is, the lower the level of political competition. We interact the competition measure with our variables of interest and implement a full triple interaction model on the reelection probability.

matters for the implementation of market-liberalizing reforms.

²⁸Some studies suggest that the time in ofice could also be a good measure of the quality of politicians (see Gagliarducci and Paserman, 2012). However, in our set-up, duration in ofice and belonging to the treatment or the control group are intertwined. For this reason, we do not consider the duration in ofice as an appropriate outcome variable in our analysis.

Results in Table 7, columns 4-6, show that gender quotas did not afect the share of reelected politicians, since the coeficient of Treatment*After is not significant. The coeficient on the triple interaction Treatment*After*Compete, instead, shows that the efect becomes negative as the level of electoral competition decreases. In other words, gender quotas reduced the reelection probability in places in which the level of political competition is low. As in these environments it is more likely that political elites are entrenched, we interpret this result as a sign of improvement of politicians’ quality. To illustrate the size of the quotas’ efect, moving from a perfectly competitive council (the majority and the minority parties’ shares of the seats are close to 50%) to a very uncompetitive council (one party occupies all the seats) reduces the share of reelected politicians by 10 percentage points.

Another possibility that we explore is that the municipal council may be a steppingstone in political careers and some councilors may, in fact, advance at province, regional or national level. To track their further careers, we complement our dataset with the data on politicians elected to these higher levels. We then measure the quality of a given municipal council by the share of politicians elected at a higher level. The regression analysis shows that the gender quotas do not significantly afect the share of politicians reelected at a higher level. This holds true also when we look at female and male politicians separately. These findings are likely due to the fact that positions at provincial, regional and national bodies are much fewer than at local level.

Finally, we look at the turnover rates, defined as the share of politicians that are new to municipal councils.²⁹ Table 7, column 7 shows that gender quotas increased turnover. We further decompose the efect by gender in columns 8 and 9 and discover that this finding is mainly driven by male politicians. Women turnover rates, instead, appear to be unafected by the gender quotas reform. This is consistent with the idea that gender quotas helped to renew the existing class of male politicians.³⁰

7.3 Long run analysis

We analyze the long run efects of the introduction of the gender quotas and study whether the positive efects on the quality of politicians are persistent over time. So far, in our analysis we have not yet fully exploited the longitudinal dimension of our data. In this section we thus run regressions using all the available data at hand. We repeat the same regression specifications of equation 3. Table 8, columns 1 to 3, reports the coeficient of interest from the full specification for all politicians, and then separately by gender. We find evidence that the gender quotas’ efect persisted over a long time horizon. In line with the results in Section 5, the coeficient of interest in regressions for all members of councils is positive and statistically significant. The results for female and male politicians are also confirmed.

²⁹We cannot create this variable for 1391 municipalities due to a limited time span of our data. In these municipalities, the first election observed in the data is the last election before the gender quotas and thus is used in Diference in diferences estimation. Due to the fact that our dataset does not cover earlier elections, we cannot identify politicians who are new to the council.

³⁰The increase in female presence is typically associated with a decrease of the average age of politicians. Thus, gender quotas may also have an interesting negative relationship with political gerontocracy. A thorough analysis of this issue is presented in Baltrunaite et al. (2013).

[Table 8 here]

Further, we adopt an alternative strategy to estimate the long-term efects of the gender quotas. Since in this part of analysis we are exploiting a longer time dimension of the data, we estimate regressions with municipality fixed efects. Table 8, columns 4 to 6, reports the coeficient of interest from fixed-efects regressions with control variables and five-year trends, for all politicians and then separately by gender. The coeficient of interest remains significant and broadly stable across diferent specifications.

7.4 Political variables

We check the robustness of our results to the inclusion of political variables which may have an impact on politicians’ quality, mainly political ideology and political competition. First, one may argue that, given the staggered timing of municipal elections in the treatment and control group, it is possible that the efects on the politicians’ quality are mainly driven by changes in the winning parties’ ideology. We therefore explicitly control for the political ideology of the majority in municipal councils. Namely, the vector of municipality-level control variables is augmented with dummies for the political leaning of the majority in the council. The distribution of councils according to this dimension can be found in Table 1.

[Table 9 here]

Table 9 shows that our results are robust to controlling for the political ideology of the majority and suggests that they are not driven by diferences in the winning parties’ composition across elections taking place on diferent dates.

Second, we control for our measure of the level of political competition. As shown in previous studies, the latter has a positive impact on the quality of politicians. We therefore want to check whether the impact of quotas on quality remains significant when controlling for the extent of political competition. Table 9, columns 4 to 6, shows that indeed a stronger political competition is associated with a higher average level of education of municipal councilors and that our results on the efect of quotas are robust.

8 Concluding remarks

We investigate the efect of gender quotas on the quality of politicians. We analyze the temporary adoption of gender quotas in candidate lists in Italian municipalities and we try to disentangle the causal efect of the afirmative action policy from other factors. We show that the introduction of gender quotas in candidate lists increased the average education level of elected politicians, primarily by increasing the number of elected women and reducing the number of low-educated elected men. The positive efect on quality is confirmed also when we measure the latter with alternative indicators, it is robust to the inclusion of political ideology and political competition and it persists in the long run. Overall, our results suggest that gender quotas are not per se detrimental to quality, rather the opposite.

As women have caught up and often overtaken men in some areas of educational participation and performance, the existence of gender gaps in politics may represent a considerable loss for society, coming from an unexploited female potential. Even when the introduction of gender quotas has a limited impact on the gender balance in political bodies, it may represent an efective mechanism to have more educated individuals elected. Thus, as long as we expect more educated individuals to perform better as politicians, gender quotas may be beneficial for the entire society.

References

1. [1] Aidt, T.S., Dutta, J. and Loukoianova, E. (2006). "Democracy Comes to Europe: Franchise Extension and Fiscal Outcomes", European Economic Review, 50, 249- 83.
2. [2] Altonji, J. G., Elder, T. E. and Taber, C. R. (2005). "Selection on Observed and Unobserved Variables: Assessing the Efectiveness of Catholic Schools", Journal of Political Economy, 113 (1).
3. [3] Baltrunaite, A., Casarico, A. and Profeta, P. (2013). "Gender quotas and the power of the elderly", mimeo, Università Bocconi.
4. [4] Bagues, M. and Esteve-Volart, B. (2012). "Are Women Pawns in the Political Game? Evidence from Elections to the Spanish Senate", Journal of Public Economics, 96, 387-399.
5. [5] Beaman, L., Chattopadhyay, R., Duflo, E., Pande, R. and Topalova, P. (2009). "Powerful Women: Does Exposure Reduce Bias?", Quarterly Journal of Economics, 124(4), 1497-1540.
6. [6] Bellows, J. and Miguel, E. (2009). "War and Local Collective Action in Siera Leone", Journal of Public Economics, 93(11-12), 1144-1157.
7. [7] Bertocchi, G. (2011). "The Enfranchisement of Women and the Welfare State", European Economic Review, 55, 535-553.
8. [8] Besley, T. (2005) "Political Selection" The Journal of Economic Perspectives, 19(3): 43-60.
9. [9] Besley, T., Folke, O., Persson, T. and Rickne, J. (2012). "Forced to Succeed: Gender Quotas, Candidate Quality and Political Competition", mimeo, Columbia University.
10. [10] Besley, T., Montalvo, J.G. and Reynal-Querol, M. (2011). "Do educated leaders matter?", The Economic Journal, 121(554), 205-227.
11. [11] Besley, T. and Preston, I. (2007). "Electoral Bias and Policy Choice: Theory and Evidence", Quarterly Journal of Economics, 122(4), 1473-1510.
12. [12] Braendle, T. and Stutzer, A. (2010). "Public Servants in Parliament: Theory and Evidence on Its Determinants in Germany", Public Choice, 145(1-2), 223-252.
13. [13] Campa, P. (2011). "Gender Quotas, Female Politicians and Public Expenditures: Quasi-Experimental Evidence", Econpubblica Working Paper No.157.
14. [14] Casas-Arce, P. and Saiz, A. (2011). "Women and Power: Unwilling, Inefective, or Held Back?", IZA DP No.5645.
15. [15] Caselli, F. and Morelli, M. (2004). "Bad Politicians", Journal of Public Economics, 88, 759-782.
16. [16] Chattopadhyay, R. and Duflo, E. (2004). "Women as policy-makers: Evidence from a randomized policy experiment in India", Econometrica, 72(5), 1409-1443.
17. [17] Clots-Figueras, I. (2011). "Women in politics: Evidence from the Indian States", Journal of Public Economics, 95(7-8), 664-690.
18. [18] Croson, R. and Gneezy, U. (2009). "Gender Diferences in Preferences", Journal of Economic Literature, 47(2), 448-474.
19. [19] De Paola, M., Scoppa, V. and Lombardo, R. (2010). "Can gender quotas break down negative stereotypes? Evidence from changes in electoral rules", Journal of Public Economics, 94(5-6), 344-353.
20. [20] De Paola, M., Scoppa, V. and De Benedetto, M. A. (2012). "The Impact of Gender Quotas on Electoral Participation: Evidence from Italian Municipalities", Working Paper No. 5, Dipartimento di Economia e Statistica, Universita della Calabria.
21. [21] Djankov, S., Glaeser, E., La Porta, R., Lopez-de-Silanes, F. and Shleifer, A. (2003). "The new comparative economics", Journal of Comparative Economics, 31(4), 595- 619.
22. [22] Dreher, A., Lamla, M., Lein, S. and Somogyi, F. (2009). "The Impact of Political Leaders Profession and Education on Reform," Journal of Comparative Economics, 37, 169-193.
23. [23] Duflo, E. and Topalova, P. (2004). "Unappreciated Service: Performance, Perceptions and Women Leaders in India", mimeo, MIT.
24. [24] Edlund, L. and Pande, R. (2002). "Gender politics: The political salience of marriage", Discussion Papers 0102-56, Columbia University, Department of Economics.
25. [25] Epstein, M. J., Niemi, R. G. and Powell, L. W. (2005). "Do Women and Men State Legislators Difer?" in S. Thomas and C. Wilcox, Women and Elective Politics: Past, Present and Future, II Ed. New York, Oxford University Press
26. [26] Ferreira, F. and Gyourko, J. (2011). "Does Gender Matter for Political Leadership? The Case of U.S. Mayors", NBER Working Paper No. 17671.
27. [27] Folke, O. and Rickne, J. (2012), "Female Representation but Male Rule? Party Competition and the Political Glass Ceiling", mimeo, Columbia University.
28. [28] Fortunato, P. and Panizza, U. (2011). "Democracy, education and the quality of government", POLIS Working Papers No. 155.
29. [29] Funk, P. and Gathmann, C. (2010). "Gender Gaps in Policy Making: Evidence from Direct Democracy in Switzerland", Working paper, Universitat Pompeu Fabra.
30. [30] Gagliarducci, S. and Nannicini, T. (2012). "Do Better Paid Politicians Perform Better? Disentangling Incentives from Selection", Journal of the European Economic Association, forthcoming.
31. [31] Galasso, V. and Nannicini, T. (2011). "Competing on Good Politicians", American Political Science Review, 105(1), 79-99.
32. [32] Gagliarducci, S. and Paserman, D. (2012). "Gender Interactions within Hierarchies: Evidence from the Political Arena", Review of Economic Studies.
33. [33] Gehlbach, S., Sonin, K. and Zhuravskaya, E. (2010). "Businessman Candidates", American Journal of Political Science, 54(3), 718 736.
34. [34] Glaeser, E.L., La Porta, R., Lopez-de-Silanes, F., and Shleifer,A. (2004). "Do Institutions Cause Growth?", Journal of Economic Growth, 9, 271-303.
35. [35] Holzer, H. J. and Neumark, D. (2000). "What Does Afirmative Action Do?", Industrial and Labor Relations Review, 53, 240-271.
36. [36] Jones, F.B. and Olken, A.B. (2005). "Do leaders matter? National Leadership and Growth Since World War II", Quarterly Journal of Economics, 835-864.
37. [37] Julio, P. and Tavares, J. (2010). "The good, the bad and the diferent: Can gender quotas raise the quality of politicians", CEPR Discussion Paper No.7917.
38. [38] Kotakorpi, K. and Poutvaara, P. (2011). "Pay for politicians and candidate selection: An empirical analysis", Journal of Public Economics, 95(7-8), 877-885.
39. [39] Krook, M.L. (2009). "Quotas for Women in Politics: Gender and Candidate Selection Reform Worldwide", New York: Oxford University Press.
40. [40] Lott, J. R., and Kenny, L. W. (1999). "Did Women’s Sufrage Change the Size and Scope of Government?", Journal of Political Economy, 107(6), 1163-1198.
41. [41] Maniquet, F., Frechette, G.R. and Morelli, M. (2008). "Incumbents’ interests and gender quotas", American Journal of Political Science, 52, 891-907.
42. [42] Merlo, A., Galasso, V., Landi, M. and Mattozzi, A. (2009). "The Labor Market of Italian Politicians", in T. Boeri, A. Merlo and A. Pratt (2010), The ruling class: management and politics in modern Italy, Oxford University Press.
43. [43] Niederle, M., Segal, C. and Vesterlund, L. (2008). "How costly is diversity? Afirmative action in competitive environments", NBER Working Paper No. 13923.
44. [44] Rehavi, M. M. (2007). "Sex and politics: Do female legislators afect state spending?", mimeo, Berkeley.
45. [45] Rosenson, B.A. (2006). "The Impact of Ethics Laws on Legislative Recruitment and the Occupational Composition of State Legislatures", Political Research Quarterly, 59(4), 619-627.
46. [46] Stevens, A. (2007). "Women, Power and Politics", Palgrave Macmillan, Houndmills.

Figure 1: Elections’ timing

Figure 2: Councilors’ education

(a) Male politicians

(b) All politicians Figure 3: Kernel density plots for the treatment group

Table 1: Baseline covariates

Note. Pre-treatment baseline covariates in treatment and control municipalities and their diference. Covariates are: population over age 15 (Population), employment rate (Employment rate), female employment rate (Female employment rate), share of university graduates (Local education level), share of municipalities located in the South (Located in the South), share of municipalities with center-right majority (Center-right majority), with left-wing majority (Left-wing majority), with civic list majority (Civic list majority) and with coalition majority (Coalition majority), political competition (Political competition) defined as the diference between the seat shares received by the majority and by the second largest minority group. Standard errors in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent.

Table 2: Descriptive analysis on the average years of education

Note. The average number of years of education of municipal councilors in treatment and control group municipalities at the last election before the adoption of gender quotas (Before period) and the first election after the adoption of this provision (After period). For treated (control) municipalities, After elections are held when the gender quota is in force (no longer in force). Panel A shows the results for all politicians, Panel B - for females and Panel C - for males. Standard errors in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent. 29 29

Table 3: Average years of education of council members Panel A: all politicians

Note. OLS regressions of average years of education of council members. Panel A shows the results for all politicians, Panel B - for females and Panel C - for males. The sample contains the last election before the adoption of gender quotas and the first election after it for every Italian municipality. Columns 2-4 include province dummies; columns 3-4 control for the municipality employment rate, the share of university graduates, population over age 15 and its square (coeficients of the control variables are not reported); column 4 includes interactions between province dummies and the dummy After. Standard errors clustered at province level in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent.

Table 4: All politicians

Note. The table reports the Diference in diferences coeficient from OLS regressions of variance of years of education of council members. Treatment and After dummies are included in the regression, but their coeficients are not reported. The sample contains the last election before the adoption of gender quotas and the first election after it for every Italian municipality. Columns 2-4 include province dummies; columns 3-4 control for the municipality employment rate, the share of university graduates, population over age 15 and its square (coeficients of the control variables are not reported); column 4 includes interactions between province dummies and the dummy After. Standard errors clustered at province level in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent.

Table 5: Placebo Diference in diferences

Note. The table reports the Diference in diferences coeficient from OLS regressions of average years of education of council members. Treatment and After dummies are included in the regression, but their coeficients are not reported. The sample contains elections in control and treatment group municipalities during a five year period before the "placebo reform" on January 1, 1988 and a five year period after it. Columns 2-4 include province dummies; columns 3-4 control for the municipality employment rate, the share of university graduates, population over age 15 and its square (coeficients of the control variables are not reported); column 4 includes interactions between province dummies and the dummy After. Standard errors clustered at province level in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent.

Table 6: Heterogenous efects of gender quotas on average years of education

Note. Panel A reports the Diference in diferences coeficient from OLS regressions of average years of education of council members for every Italian municipality where the pre-treatment share of female councilors was lower than the one prescribed by the quota. Treatment and After dummies are included in the regression, but their coeficients are not reported. Panel B reports the Diference in diferences coeficient and the Triple diference in diferences coeficient from OLS regressions of average years of education of council members. FemaleEmployment95 is a dummy for the municipalities with the female employment rate above the 95th percentile. Treatment, After and the pairwise interactions with the dummy FemaleEmployment95 are included in the regression, but their coeficients are not reported. In both panels we use the last election before the adoption of gender quotas and the first election after it. Columns 2-4 include province dummies; columns 3-4 control for the municipality employment rate, the share of university graduates, population over age 15 and its square (coeficients of the control variables are not reported); column 4 includes interactions between province dummies and the dummy After. Standard errors clustered at province level in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent.

Table 7 Other outcome variables

Note The dependent variable in columns 1 to 3 is the share of councilors with previous high-skill j ob (see Table A 3 in Appendix for a detailed list of occupations) ; in columns 4 to 6 - the share of councilors that are reelected in subsequent municipal elections ; in columns 7 to 9 - the turnover rate . Column 1 4 and 7 show the results for all councilors columns 2 5 and 8 - for female councilors and columns 3 6 and 9 - for male councilors . Columns 4 to 6 report the Diference in diferences coeficient and the Triple diference in diferences coeficient from OL S regressions of the share of reelected politicians . Compete is the diference between the seats share of the maj ority and the second largest party group . Treatment After and the pairwise interactions with the variable Compete are included in the regression but their coeficients are not reported . The sample contains the last election before the adoption of gender quotas and the first election after it for every Italian municipality. All columns include province dummies and interactions between province dummies and the dummy After. All columns control for the municipality employment rate the share of university graduates population over age 1 5 and its square (coeficients of the control variables are not reported) . Standard errors clustered at province level in parenthesis . The following symbols indicate diferent significance levels : * * * - significant at 1 percent * * - significant at 5 percent * - significant at 1 0 percent .

Note. The table reports the Diference in diferences coeficient from OLS regressions of average years of education of council members. Treatment and After dummies are included in the regression, but their coeficients are not reported. The sample contains all elections during the period 1985-2009 for every Italian municipality. Columns 1-3 include province dummies and interactions between province dummies and the dummy After. Columns 4-6 include municipality dummies and dummies for 5-year periods. All columns control for the municipality employment rate, the share of university graduates, population over age 15 and its square (coeficients of the control variables are not reported). Columns 1 and 4 show the results for all councilors, columns 2 and 5 - for female councilors, column 3 and 6 - for male councilors. Standard errors clustered at province level in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent.

Table 9: Political variables

Note. OLS regressions of average years of education of council members. Treatment and After dummies are included in the regression, but their coeficients are not reported. The sample contains the last election before the adoption of gender quotas and the first election after it for every Italian municipality. All columns include province dummies and interactions between province dummies and the dummy After. All columns control for the municipality employment rate, the share of university graduates, population over age 15 and its square (coeficients of the control variables are not reported). Columns 1-3 include dummies for left-wing majority (Left-wing majority), civic list majority (Civic list majority) and coalition majority (Coalition majority). Center-right majority is used as excluded category. Columns 4-6 control fo electoral competition (Political competition) defined as the diference between the seat shares received by the majority and by the second largest party group. Columns 1 and 4 show the results for all councilors, columns 2 and 5 - for female councilors and columns 3 and 6 - for male councilors. Standard errors clustered at province level in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent.

Appendix

Table A.1: Female presence

Note. OLS regressions of share of females in municipal councils. The sample contains the last election before the adoption of gender quotas and the first election after it for every Italian municipality. Columns 2-4 include province dummies; columns 3-4 control for the municipality employment rate, the share of university graduates, population over age 15 and its square (coeficients of the control variables are not reported); column 4 includes interactions between province dummies and the dummy After. Standard errors clustered at province level in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent.

Table A.2: Share of municipal councilors with a high school diploma All politicians

Note. OLS regressions of share of municipal councilors that have achieved at least a high school diploma (or equivalent). The sample contains the last election before the adoption of gender quotas and the first election after it for every Italian municipality. Columns 2-4 include province dummies; columns 3-4 control for the municipality employment rate, the share of university graduates, population over age 15 and its square (coeficients of the control variables are not reported); column 4 includes interactions between province dummies and the dummy After. Standard errors clustered at province level in parenthesis. The following symbols indicate diferent significance levels: *** - significant at 1 percent, ** - significant at 5 percent, * - significant at 10 percent.

Table A.3: Skill-intensive occupations

Note. Data from Italian Ministry of the Interior, Department for Territorial and Internal Afairs.

References

1. 2013-11: “Gender quotas and the quality of politicians”, Audinga Baltrunaite, Piera Bello, Alessandra Casarico y Paola Profeta.

References

1. 2013-10: “Brechas de Género en los Resultados de PISA :El Impacto de las Normas Sociales y la Transmisión Intergeneracional de las Actitudes de Género”, Sara de la Rica y Ainara González de San Román.

References

1. 2013-09: “¿Cómo escogen los padres la escuela de sus hijos? Teoría y evidencia para España”, Caterina Calsamiglia, Maia Güell.

References

1. 2013-08: “Evaluación de un programa de educación bilingüe en España: El impacto más allá del aprendizaje del idioma extranjero”, Brindusa Anghel, Antonio Cabrales y Jesús M. Carro.

References

1. 2013-07: “Publicación de los resultados de las pruebas estandarizadas externas: ¿Tiene ello un efecto sobre los resultados escolares?”, Brindusa Anghel, Antonio Cabrales, Jorge Sainz e Ismael Sanz.

References

1. 2013-06: “DYPES: A Microsimulation model for the Spanish retirement pension system”, F. J. Fernández-Díaz, C. Patxot y G. Souto.

References

1. 2013-05: “Vertical differentiation, schedule delay and entry deterrence: Low cost vs. full service airlines”, Jorge Validoa, M. Pilar Socorroa y Francesca Medda.

References

1. 2013-04: “Dropout Trends and Educational Reforms: The Role of the LOGSE in Spain”, Florentino Felgueroso, María Gutiérrez‐Domènech y Sergi Jiménez‐Martín.

References

1. 2013-03: “Understanding Different Migrant Selection Patterns in Rural and Urban Mexico”, Simone Bertoli, Herbert Brücker y Jesús Fernández-Huertas Moraga.

References

1. 2013-02: “Understanding Different Migrant Selection Patterns in Rural and Urban Mexico”, Jesús Fernández-Huertas Moraga.

References

1. 2013-01: “Publicizing the results of standardized external tests: Does it have an effect on school outcomes?, Brindusa Anghel, Antonio Cabrales, Jorge Sainz y Ismael Sanz.

References

1. 2012-12: “Visa Policies, Networks and the Cliff at the Border”, Simone Bertoli, Jesús Fernández-Huertas Moraga.

References

1. 2012-11: “Intergenerational and Socioeconomic Gradients of Child Obesity”, Joan Costa-Fonta y Joan Gil.

References

1. 2012-10: “Subsidies for resident passengers in air transport markets”, Jorge Valido, M. Pilar Socorro, Aday Hernández y Ofelia Betancor.

References

1. 2012-09: “Dual Labour Markets and the Tenure Distribution: Reducing Severance Pay or Introducing a Single Contract?”, J. Ignacio García Pérez y Victoria Osuna.

References

1. 2012-08: “The Influence of BMI, Obesity and Overweight on Medical Costs: A Panel Data Approach”, Toni Mora, Joan Gil y Antoni Sicras-Mainar.

References

1. 2012-07: “Strategic behavior in regressions: an experimental”, Javier Perote, Juan Perote-Peña y Marc Vorsatz.

References

1. 2012-06: “Access pricing, infrastructure investment and intermodal competition”, Ginés de Rus y M. Pilar Socorro.

References

1. 2012-05: “Trade-offs between environmental regulation and market competition: airlines, emission trading systems and entry deterrence”, Cristina Barbot, Ofelia Betancor, M. Pilar Socorro y M. Fernanda Viecens.

References

1. 2012-04: “Labor Income and the Design of Default Portfolios in Mandatory Pension Systems: An Application to Chile”, A. Sánchez Martín, S. Jiménez Martín, D. Robalino y F. Todeschini.

References

1. 2012-03: “Spain 2011 Pension Reform”, J. Ignacio Conde-Ruiz y Clara I. Gonzalez.

References

1. 2012-02: “Study Time and Scholarly Achievement in PISA”, Zöe Kuehn y Pedro Landeras.

References

1. 2012-01: “Reforming an Insider-Outsider Labor Market: The Spanish Experience”, Samuel Bentolila, Juan J. Dolado y Juan F. Jimeno.

References

1. 2011-13: “Infrastructure investment and incentives with supranational funding”, Ginés de Rus y M. Pilar Socorro.

References

1. 2011-12: “The BCA of HSR. Should the Government Invest in High Speed Rail Infrastructure?”, Ginés de Rus.

References

1. 2011-11: “La rentabilidad privada y fiscal de la educación en España y sus regiones”, Angel de la Fuente y Juan Francisco Jimeno.

References

1. 2011-10: “Tradable Immigration Quotas”, Jesús Fernández-Huertas Moraga y Hillel Rapoport.

References

1. 2011-09: “The Effects of Employment Uncertainty and Wealth Shocks on the Labor Supply and Claiming Behavior of Older American Workers”, Hugo Benítez-Silva, J. Ignacio García-Pérez y Sergi Jiménez-Martín.

References

1. 2011-08: “The Effect of Public Sector Employment on Women’s Labour Martket Outcomes”, Brindusa Anghel, Sara de la Rica y Juan J. Dolado.

References

1. 2011-07: “The peer group effect and the optimality properties of head and income taxes”, Francisco Martínez-Mora.

References

1. 2011-06: “Public Preferences for Climate Change Policies: Evidence from Spain”, Michael Hanemann, Xavier Labandeira y María L. Loureiro.

References

1. 2011-05: “A Matter of Weight? Hours of Work of Married Men and Women and Their Relative Physical Attractiveness”, Sonia Oreffice y Climent Quintana-Domeque.

References

1. 2011-04: “Multilateral Resistance to Migration”, Simone Bertoli y Jesús Fernández-Huertas Moraga.