Do women and female politicians have different policy priorities?

It is generally assumed that the differences in policy priorities and preferences between men and women legislators mirror the differences observed amongst the citizenry, as women politicians share experience with women in general (Ferreira & Gyourko, Reference Ferreira and Gyourko2014). Various surveys of general populations document the divergent policy preferences of men and women, even if other background characteristics are controlled (Applegate, Cullen & Fisher, Reference Applegate, Cullen and Fisher2002; Finseraas, Jakobsson & Kotsadam, Reference Finseraas, Jakobsson and Kotsadam2012; Gottlieb, Grossman & Robinson, Reference Gottlieb, Grossman and Robinson2018; Jaffee & Hyde, Reference Jaffee and Hyde2000; Shapiro and Mahajan, Reference Shapiro and Mahajan1986). These differences are typically explained by two mechanisms: one refers to the socialisation into gender roles and stereotypes, and the other assumes the rational calculation and promotion of gender-based interests (Shorrocks & Grasso, Reference Shorrocks and Grasso2020).

Socialisation of gender roles has long-lasting influences on social perceptions, which translate into political priorities and preferences. For a long time, women have been stereotyped as communal, compassionate, collaborative and family oriented, while the masculine stereotypes have been centred around agentic, goal-oriented and instrumental traits (Eagly & Steffen, Reference Eagly and Steffen1984). This stereotypical division between ‘caregivers’ and ‘breadwinners’ corresponds to the division of policy areas (portfolios), popularly labelled as either more feminine, such as childcare or education, or more masculine, such as entrepreneurship, roads or defence. In consequence, men and women tend to support policies that are related to the interests stereotypically aligned with their gender.

The observed differences in policy preferences might also stem from rational calculations. As long as women are disproportionately burdened with caring for children, the elderly and the ill (and they are indeed, even in the most gender-equal societies), they should have a real interest in higher redistribution and expanded welfare services. Women use education, health and social care services either for themselves or for their children and parents to a greater extent than men. At the same time, men, typically less entrenched in the welfare system, may prefer limited public expenditures and low taxation (Iversen & Rosenbluth, Reference Iversen and Rosenbluth2006). Although women are not more likely to support leftist parties in all contexts, they tend to prefer active labour market policies, particularly in countries with higher divorce rates and higher female labour market participation (Iversen & Rosenbluth, Reference Iversen and Rosenbluth2006, pp.17–18). As Morgan (Reference Morgan2013) demonstrated, political parties, particularly when facing electoral instability and dealignment of the core constituencies, often attempted to strategically respond to women’s preferences by expanding work–family policies. Shorrocks and Grasso (Reference Shorrocks and Grasso2020) additionally argue that the gender gap in policy preferences is context-dependent: larger and more stable across generations in contexts characterised by low welfare provision, such as the USA, and weaker in the more welfare-oriented settings, such as the UK.

Gender differences in policy preferences amongst the electorate do not translate automatically into the differences in legislators’ policy choices, but we may expect that the officeholders’ gender contributes to the explanations of policy change. Under the classical median voter theorem (Downs, Reference Downs1957), the gender of an elected decision-maker is irrelevant to policy choices, as all politicians are incentivised to converge towards the policy position of a median voter. However, when we take into account more realistic premises, such as personalisation of elections, political leadership based on individual traits and imperfect electoral accountability, a politician’s identity becomes a plausible factor in explaining their policy choices and outcomes. Under the citizen–candidate model (Alesina, Reference Alesina1988; Besley & Coate, Reference Besley and Coate1997; Osborne & Slivinski, Reference Osborne and Slivinski1996), politicians can inject their preferences into public policies. In addition, the ‘politics of presence’ model of political representation assumes that female politicians bring different interests and issues to politics (Mansbridge, Reference Mansbridge1999). It pertains particularly to ‘women’s issues’, defined as those that directly and disproportionately affect women, or that due to traditional gender roles mostly influence women (Besley & Case, Reference Besley and Case2003). In such theoretical frameworks, it is valid to expect that an increasing representation of a particular category (let it be gender, class, race or age cohort) amongst elected officials leads to a systematic policy change, modification of spending patterns or a procedural change in the sheer decision-making process.

The question concerning the causal relationship between the officeholders’ gender and the implemented policies has already been posed concerning the policies determined at the national level, such as reproductive rights (Berkman & O’Connor, Reference Berkman and O’Connor1993), maternity leaves (Kittilson, Reference Kittilson2008), family benefits (Ennser-Jedenastik, Reference Ennser-Jedenastik2017), social investment policies (Garritzmann & Schwander, Reference Garritzmann and Schwander2021), developmental aid (Hicks, Hicks, & Maldonado, Reference Hicks, Hicks and Maldonado2016) and defence spending (Koch & Fulton, Reference Koch and Fulton2011).

At the local level, several surveys of local politicians indicate that women and men in office have different policy preferences, generally aligning with the stereotypes of masculine and feminine issues (Slegten, Geys & Heyndels, Reference Slegten, Geys and Heyndels2019; Slegten & Heyndels, Reference Slegten and Heyndels2020; Svaleryd, Reference Svaleryd2009). A growing body of scholarly literature directly investigates policy outcomes rather than policymakers’ preferences. The composition of local councils and the mayor’s gender are used to explain the observed differences between jurisdictions. Numerous local government units with the local authorities’ discretion over the distribution of basic public goods, particularly in decentralised states, offer a very good analytical leverage to use experimental and quasi-experimental designs. In particular, difference-in-differences (DiD) or regression discontinuity (RD) designs are used to pinpoint the causal relationship between descriptive and substantive representation using observational data.

The pioneering study of Chattopadhyay and Duflo (Reference Chattopadhyay and Duflo2004) exploited a natural experiment with reserved seats in India, which demonstrated that local leaders invest more in infrastructure that is relevant to the needs of their gender. Ferreira and Gyourko (Reference Ferreira and Gyourko2014) used RDD to analyse the results of mixed-gender mayoral races in the USA between 1950 and 2005 and did not find systematic differences in total public expenditures, taxation, employment in local administration, health, welfare, leisure, infrastructure or various crime rates. These results held both in the short and the long run. Similar null results, yet with slightly different sets of outcome indicators, have been obtained in the analyses of local governments in Spain (Campa, Reference Campa2011; Gago & Carozzi, Reference Gago and Carozzi2021), Italy (Casarico, Lattanzio & Profeta, Reference Casarico, Lattanzio and Profeta2022; Gagliarducci & Paserman, Reference Gagliarducci and Paserman2012; Rigon & Tanzi, Reference Rigon and Tanzi2012), Czechia (Kuliomina, Reference Kuliomina2021), Korea (Jung, Reference Jung2022) and Norway (Geys & Sørensen, Reference Geys and Sørensen2019).

Systematic differences in spending or policy outcomes attributed to the officeholder’s gender are found less often. Funk and Philips (Reference Funk and Philips2019) showed that newly elected female mayors in Brazil systematically modified the distribution of public funds in accordance with the interests more important for their gender: they allocated more for education, healthcare and social assistance, and less on transportation and urban development, relative to their male counterparts. Ordine et al. (Reference Ordine, Rose and Giacobbe2023) demonstrated that the increased presence of women in local politics leads to slightly reduced public spending, except for childcare and social care. This finding is generally in line with another study of cuts in local budgets: Cabaleiro Casal and Buch Gómez (Reference Cabaleiro Casal and Buch Gómez2018) found that forced to reduce spending, female mayors are more likely to preserve social spending than their male counterparts. Earlier, Holman (Reference Holman2014) reported that US cities with female mayors were more likely to participate in funding social welfare programs. Bratton and Ray (Reference Bratton and Ray2002) found that the local childcare systems in Norway developed quicker in municipalities with higher shares of women local councillors. They argue, however, that this effect was more evident in the period of policy innovation and faded out over time. This observation led the authors to a broader generalisation that descriptive representation is more closely related to policy innovation or early adoption than expansion or maintenance of an existing policy (Bratton & Ray, Reference Bratton and Ray2002, p.436). Several other studies confirmed the relationship between female descriptive representation and the development of local childcare or education, for example, in Sweden (Svaleryd, Reference Svaleryd2009), Germany (Hessami & Baskaran, Reference Hessami and Baskaran2019; Yamada, Reference Yamada2024) and Austria (Walenta-Bergmann, Reference Walenta-Bergmann2023).

The tentative conclusion from this literature is that the effects of the local officeholders’ gender on substantive policy are not universal. Hessami and da Fonseca (Reference Hessami and Lopes da Fonseca2020), in their recent meta-analysis, argue that the increase in female descriptive representation causes a better provision of public goods in developing countries. In developed countries, the officeholder’s gender tends to matter less, and the differences in spending patterns are not systematic. Yet, the picture is even more complex. Many studies highlight the contingency of gender effects on spending or policy outcomes, stressing the importance of the institutional context (Koch & Fulton, Reference Koch and Fulton2011), the political leaning of the whole council (Walenta-Bergmann, Reference Walenta-Bergmann2023), the extent of women’s representation in other offices (Funk & Philips, Reference Funk and Philips2019; Holman, Reference Holman2014; Koch & Fulton, Reference Koch and Fulton2011) and intersectional identities (Clots-Figueras, Reference Clots-Figueras2011).

In this paper, we focus on female representation amongst local executives and search for causal evidence on whether the election of women leads to changes in the provision of childcare and spending on this service. The advantage of our approach is that we are able to compare the possible impact of the mayor’s gender on both service coverage and public spending. With two different types of indicators from different data sources and fine-grained yearly data covering four consecutive terms, the analyses should lead to more robust conclusions. Noteworthy, within the period of our analyses, we are able to capture a dynamic development of childcare services in Poland that occurred in the recent decade. The subsequent section provides more contextual information on this process.

Institutional context: mayors as policymakers

Polish local governments offer a favourable context to investigate the linkage between the descriptive and substantive representation of women. Due to the concentration of power in the local executives, instead of the proportion of women amongst councillors, we may look at the mayor’s gender as an explanation of local policies. It reflects the fact that mayors are the key policymakers when it comes to local services. Heinelt et al. (Reference Heinelt, Hlepas, Kuhlmann, Swianiewicz, Heinelt, Magnier, Cabria and Reynaert2018) place Poland amongst the European countries with the highest values of ‘mayoral strength index’. Mayors in Poland are elected directly, under a majoritarian two-round system, simultaneously with the municipal councils (elected under the First-past-the-post (FPTP) rule in municipalities below 20,000 inhabitants and under the open-list proportional representation system (OLPR) in larger municipalities). Mayors currently serve a 5-year term (4-year term before 2018). Having a direct mandate, they can be recalled only in a popular referendum. In such cases, a minimal turnout threshold is required, which makes the potential recall very unlikely (Gendźwiłł & Swianiewicz, Reference Gendźwiłł, Swianiewicz and Sweeting2018). A majority of mayors in Poland are not formally affiliated with any national party, usually running as candidates representing local independent lists.

Mayors in Poland typically control the council majority, although the electoral rules allow for the formation of a divided local government. However, even with an unstable or opposing council majority, executive mayors can still pursue their agenda as the key decision-makers. They do not preside over the council, but co-define the council agenda, enjoying the exclusive prerogative to submit the budget proposal. In practice, councillors can introduce only small amendments and influence local policies. As the independent civil service at the local level is almost non-existent, mayors practically control the local administration, appoint the heads of administrative departments, and have a strong influence on municipal institutions (schools, kindergartens, libraries) and agencies delivering basic services.

Poland has been experiencing a slow increase in the share of women amongst local councillors and mayors (Table 1). While the percentage of women amongst local councillors exceeded 30 per cent after the 2018 elections, and at that time was comparable to the level of female descriptive representation in the national parliament, there is still a huge legislative–executive gender gap, with only 12 per cent of women amongst mayors elected for the term 2018–2023.

Table 1.

Women amongst local elected officials in Poland, 2002–2018

Local childcare is one of the policy areas that underwent a substantial and dynamic change in recent decades. Incentivised initially by the EU structural funds and subsequently by the central government programs (with a flagship Toddler [Maluch] program established in 2011), local governments have been developing infrastructure to secure preschool education and early childcare. Many local governments supported subsidised private childcare facilities to meet the increased local needs. It was in response to an increasing demand related to social modernisation, changes in the model of family and changes in the labour market (Szelewa, Reference Szelewa2012; Szelewa & Polakowski, Reference Szelewa and Polakowski2008). Figure 1 demonstrates that from 2005 to 2020 the share of children enrolled in various childcare facilities grew from around 55 per cent to 90 per cent for the cohort between 3 and 6 years old and from less than 5 per cent to 15 per cent for the cohort between 0 and 3 years old.Footnote ¹ Thus, Poland serves as a very good example of a dynamic change in public service that is strategically important for women’s interests.

Figure 1.

Development of local childcare services in Poland, 2002–2022.

Source: Statistics Poland, Ministry of Finance.

Note: Data on children enrolled in private nurseries unavailable before 2012.

Both tasks – provision of childcare for children up to the age of 3 years and preschool care for the cohort aged 3–6 years – are part of a municipality’s own tasks in Poland. In preschool services, the public sector plays a dominant role, when one takes into account both coverage and financing: 73 per cent of children enrolled in preschool education attended kindergartens run by municipalities. In addition, preschool services are predominantly financed by public funds. Private kindergartens receive a grant from the local budget (for each child enrolled), which is equal to the basic maintenance cost in a public kindergarten (this rule has been in force since 2013; previously, since 2007, the grant was at least 75 per cent of the basic cost). The funds transferred from municipal budgets to private entities for the current operation of kindergartens in 2022 accounted for 20 per cent of all current expenditures on preschool education.

The situation is slightly different in the case of care for younger children (under 3 years). Most nurseries are private, and more than half of the available places operate in this sector. In 2021, 57 per cent of children under the age of 3 years attended private nurseries. However, local governments have some influence on the development of this sector. Local authorities keep a register of nurseries, supervise and verify compliance with sanitary requirements. They often provide subsidies to private nurseries (in 2022, these grants constituted almost 20 per cent of the current expenditures allocated to the nursery care system).

Data and methods

In the subsequent analyses, we rely on the difference-in-differences (DiD) design as our basic specification. DiD is a quasi-experimental scheme using panel regression (Meyer, Reference Meyer1995; Roth et al., Reference Roth, Sant’Anna, Bilinski and Poe2023) to trace the causal relationship between the treatment and the outcome. It relies on a counterfactual assumption that in the absence of intervention, the treatment group and the control group would have followed parallel trends over time. To estimate the treatment effect, the difference between the pre-treatment and post-treatment period observed in the treatment group is subtracted from the analogous difference in the control group. DiD scheme is well suited to analyse time series, potential delays and durability of the effects.

In our case, the binary treatment variable represents the election of a female mayor who replaces a male incumbent (regardless of whether he retired or was defeated). The outcomes are four indicators measuring the scope (coverage) of local childcare policies: two related to the nurseries (and similar forms of institutionalised care for children below the age of 3 years) and two related to the preschools (and similar forms dedicated for children between 3 and 6 years old). For the sake of comparability, we investigate the number of children covered by the service in the particular municipality divided by the number of children from the target cohort (0–2 for nurseries, 3–6 for kindergartens). We analyse the total coverage rates, including both public and private facilities, as the latter were typically subsidised by local governments (see Section 3). Similarly, current municipal budget expenditures are expressed in PLN per capita.Footnote ² We focus on the current expenditures reported post hoc in the standardised budgetary reports, as they account for the vast majority of municipal expenditures on childcare (more than 80 per cent). They are also more stable, as investment expenditures are characterised by greater fluctuations between years and greater dependence on external factors, such as the availability of EU funds. Table 2 presents definitions of the variables used as outcomes in our analyses.

Table 2.

Outcome variables

If the mayor’s gender influences local childcare policy in the way stipulated by the literature, we should observe a systematic change of these indicators, more precisely, positive treatment effects, which could be attributed to the election of a woman in place of a man.

The group of treated municipalities with newly elected female mayors is contrasted with the control group composed of the municipalities where a male mayor replaced another man in office. By excluding (quite numerous) municipalities in which incumbents were re-elected, we make treatment and control groups more similar and single out potentially confounding effects of novelty or incumbency. We also verified whether the election of a female mayor could be an indicator of a more general preference of local electorates for female politicians, which would imply problems with endogeneity. We found that in each term, the percentage of women amongst municipal councillors increased, and the magnitude of these increases was never larger amongst the municipalities in which a woman was elected for a mayoral position.

We estimate our models separately for each 4-year term, treating the previous term (t, t−1, t−2, t−3) as the pre-treatment period and four consecutive years (t+1, t+2, t+3, t+4) as post-treatment periods. By doing so, we can trace the persistence of the assumed treatment effects over time and their potential interferences with the political budget cycle. As in each electoral cycle a different set of municipalities receives treatment (i.e. a newly elected female mayor), and some of the municipalities treated in one term may be part of the control group constructed for the subsequent term, we do not pool our observations and thus are not able to employ a staggered design (Roth et al., Reference Roth, Sant’Anna, Bilinski and Poe2023), which would allow for multiple treatments occurring in different timepoints and a longer time horizon to observe potential policy change. We argue that the current expenditures for childcare and enrolment indicators (including places both in public facilities and publicly funded private facilities) are able to capture policy change, if it occurs, quicker than the investment expenditures related to longer planning processes.

The decision to include the election year (t) in the pre-treatment period is motivated by the timing of local elections in Poland. During the whole analysed period, local elections were held either in October or November, with mayors assuming their duties in November or December, which means that the potential influence of newly elected local authorities on local budgets and policy outcomes was rather negligible.

We compiled the official election results for almost 2,500 municipalities starting from the 2006 elections. We excluded from our sample cities with county status – 66 large urban municipalities which are simultaneously counties, as the scope of their responsibilities differs significantly from other municipalities. Additionally, in our analysis, we excluded municipalities that significantly changed their administrative borders (thus, also their area and population), as it might be assumed that these territorial changes profoundly changed municipalities’ financial situation and local demand for public services. This case concerned only twenty-eight municipalities, which make up about 1 per cent of the whole sample. Separately for each outcome variable, we also excluded municipalities for which the time series of the outcome variables were incomplete or included anomalies, most likely attributable to one-off events or the quality of reporting.Footnote ³

On the basis of the electoral records, we selected municipalities in which a woman replaced an incumbent man (313 cases in three electoral cycles) and municipalities in which a man was elected to replace an incumbent man (1,749 cases in three electoral cycles). In fact, for each term, control and treatment groups are composed of different municipalities. Taking into account the aforementioned low level of female representation, it is not surprising that the control group is about six times larger than the treatment group. The descriptive statistics of the outcome variables are presented in the Appendix (Table A1). We verified the pre-treatment balance of the relevant covariates (population size, operating surplus per capita, the share of female councillors and the mayor’s age; the results are reported in Table A2 in the Appendix). In four out of twelve balance tests, the differences between the treatment and control groups were statistically significant (p-values below 0.1), thus we decided to include these variables as additional controls in our DiD models. In the subsequent section, we present models with and without covariates.

We estimated average treatment effects on the treated (ATET) separately for each of the four outcomes and for each term by fitting a linear regression model with time and municipality fixed effects. For each model, we tested the parallel trends assumption (PTA) verifying whether, in the pre-treatment period, the (linear) trends in the outcome variables were parallel between the control and treatment groups. In the second step of the analysis, by including leads and lags of the variable indicating treatment, we tested whether the treatment effects were time specific. This is done for two purposes. First, to verify whether the effects can be observed in anticipation of the treatment – it could be the case if, for example, male incumbents had changed the policy being challenged by a female contender. Second, to verify whether the effects of the treatment appear delayed, as the establishment or expansion of the service may take some time.

Subsequently, to test the robustness of our results, we report the analyses using generalised synthetic control (GSC; Xu, Reference Xu2017), a more complex research design that allows for the presence of unobserved time-varying confounders affecting units differently. GSC builds upon earlier synthetic control approaches (Abadie, Diamond & Hainmueller, Reference Abadie, Diamond and Hainmueller2010). While both GSC and DiD assume strict exogeneity, GSC offers an important advantage in handling the parallel trends assumption, as it allows for the presence of unobserved time-varying confounders. This approach estimates latent factors and factor loadings, enabling unit-specific responses to unobserved confounders. We used so-called matrix completion (MC) techniques, as proposed by Athey et al. (Reference Athey, Bayati, Doudchenko, Imbens and Khosravi2021), and a data-driven approach to construct the synthetic control unit and impute counterfactuals. The basic idea of the MC method is to view the panel data as a matrix where rows represent units (e.g. countries, regions) and columns represent periods. Some entries in this matrix are observed, while others (the counterfactuals) are missing. The goal is to ‘complete’ the matrix by estimating these missing values. This is done by assuming the matrix has a low-rank structure, meaning the product of two lower-dimensional matrices can approximate it; this implies that a small number of underlying factors explain most of the variation in the data.

Although we believe that the GSC may further improve the reliability and validity of our estimation, one should notice that GSC relies on sufficient pre-treatment data to accurately estimate latent factors and factor loadings. The fewer the pre-treatment observations, the less reliable synthetic control units.