European childcare regimes beyond the ‘average family’ model: A fuzzy clustering analysis

Abstract

The sustained expansion of early childhood care policies across Europe has altered the landscape of work-family reconciliation, yet this transformation remains unevenly distributed across social groups. This study utilises an intersectional (de)familialisation framework that attends to inequalities of access to parental leave and childcare provision. Drawing on comprehensive policy data across European countries, we expand the care regime typology beyond the average workers’ perspective to offer insight into care regimes from the perspective of the unemployed. By analysing three separate periods of a child’s life until school age, we trace the shifting policy driven incentives for care as the child grows. Our sample also begins to close the gap in knowledge about family policy in Southeastern Europe. Finally, we present and utilise fuzzy clustering, a method that combines the strengths of traditional clustering and fuzzy set ideal type analysis, while addressing their limitations. Combining eligibility considerations alongside the established dimensions of leave generosity, childcare availability, and gender concentration of rights, we aim to contribute to a growing intersectionality-led policy scholarship that puts the question of who is excluded from policy goals at the heart of the analysis. Our findings examine the class asymmetries in access to leave rights, while also offering a perspective in which class and gender inequalities are not separate axes of (dis)advantage.

Introduction

The sustained expansion of early childhood care policies is providing many families with enhanced opportunities to combine paid work and care. In Europe, childcare service provision and parenting leaves were expanded (Eurydice, 2025; Ferragina, 2023; Kang and Meyers, 2018), even amid broader welfare retrenchment (Ferragina et al., 2012). A flurry of reform activity in the last couple of decades has seen parenting leaves becoming more generous and gender-equal (Earle et al., 2025). In 2019, the European Union agreed on new minimum standards for parental leave, increasing earmarked leave entitlements (De la Porte et al., 2020, 2023). The Barcelona targets on early child education and care were also recently revised and raised to ensure upward convergence in childcare participation (Council of the European Union, 2022). Notably, support for family-orientated redistributive efforts also grew among right-wing populist movements across Europe pushing for a generous yet exclusionary vision of family policy that privileges certain segments of the population (Graff and Korolczuk, 2022b; Orenstein and Bugarič, 2022; Szelewa and Polakowski, 2020).

Much scholarly focus has already been placed on cross-country variations in leave rights and childcare provision (for example, Szelewa and Polakowski, 2023; Finch, 2021; Ciccia and Verloo, 2012; Javornik, 2014; Lohmann and Zagel, 2016; Ciccia, 2017; An and Peng, 2016). Yet, childcare policies have often been examined from the perspective of a relatively privileged majority (Daly, 2020) and much less attention has been dedicated to the issue of stratification (Saxonberg, 2024). Recently, authors have called for a more nuanced consideration of the inequality aspect of parental leave designs along the lines of social class, highlighting the role of complex eligibility criteria (Dobrotić, 2022; Dobrotić and Blum, 2020; Javornik and Kurowska, 2017; Mathieu et al., 2020; McKay et al., 2016). As shown by Dobrotić and Blum (2020), most European countries predicate access to parenting leave benefits on sufficient employment history; parents with atypical employment paths are excluded from provision or relegated to a less-generous entitlement. Eligibility criteria can systematically exclude certain parents, mirroring and amplifying existing inequalities (Dobrotić, 2022; Mathieu et al., 2020; McKay et al., 2016).

Additionally, scholars have long highlighted the inequality of childcare service usage (Abrassart and Bonoli, 2015; Ferragina and Magalini, 2023; Pavolini and Van Lancker, 2020; Van Lancker and Ghysels, 2016; Wood et al., 2023). As participation rates in early childcare services has increased in Europe, so has the inequality of access (Välimäki et al., 2024). The importance of access to leave and childcare services for vulnerable parents is difficult to understate; for example, Bartova et al. (2022) finds unemployed single mothers with access to generous leave have higher chance of future paid work, while Scherer and Pavolini (2023) show a positive association between childcare provision and employment outcomes for mothers with low educational attainment. At the same time, Marynissen et al. (2021) finds more than a quarter of mothers in Belgium cannot access leave, while the eligibility of a further third is precarious, and this accounts for a large part of educational and age gradients in leave use. The story of childcare policy transformation writ large can be one of bifurcation along the lines of social class (Ferragina, 2023).

Despite increased discussion on social inequalities in access to childcare and parental leave rights, there is paucity of studies that would approach childcare regime typologies from this perspective. The present article seeks to address this gap. We integrate inequalities of access alongside the (de)familialisation perspective into typology building, focusing simultaneously on the scope of rights and on their inclusiveness. We additionally incorporate the gendered dimension of parenting leave rights thus attending comprehensively to what entitlements are on offer and who gets to use them. For the first time, we expand the childcare regime typology beyond an ‘average family model’, a privileged, double earner couple, to offer insight into care regimes from the perspective of a more vulnerable group – the unemployed couple.

Our study also begins to close the gap in knowledge about childcare policies in Southeastern Europe, presenting a more comprehensive mapping of European provision today. While Szelewa and Polakowski (2008) and Javornik (2014) have notably challenged the gap in the literature regarding CEE family policy by comparing country cases in the region, knowledge about CEE in comparative perspective remains largely based on the Visegrád states and knowledge accumulation about less studied parts of Europe, particularly the Balkans, has stalled. Including these countries alongside the rest of Europe allows us to observe distinct patterns of post-Yugoslavian childcare policy (see Dobrotić, 2022) and attend to different trajectories of care policy development across Europe (Dobrotić and Stropnik, 2020; Hašková and Saxonberg, 2016; Yerkes et al., 2022).

Finally, we present and utilise fuzzy clustering, a method that to the best of our knowledge has not been used before in the comparative family/childcare policy literature. Ferragina and Deeming (2022) have noted the increasing popularity of both clustering and fuzzy-set techniques in comparative research. The family policy literature provides abundant examples of both clustering algorithms (Bambra, 2007; Cho, 2014, among others) and QCA/fuzzy-set techniques, particularly fuzzy-set ideal type analysis (Ciccia and Verloo, 2012; Lee, 2025; Szelewa and Polakowski, 2008, 2023). In their 2010 effort to identify intergenerational policy regimes, Saraceno and Keck (2010) utilise both methods and decide against reporting the results of either. Traditional cluster analysis has been criticised as a poor fit for modelling the vagueness and gradual variation in family policy data (see Ciccia and Verloo, 2012). At the same time, as Saraceno and Keck (2010) and Cooper and Glaesser (2011) have noted, QCA techniques are too sensitive to cut-off points, while calibrating inclusion into ideal-typical sets can also be challenging in a fast-changing field (see Javornik, 2014). The fuzzy clustering approach we adopt combines strengths of both traditional clustering and fuzzy set ideal type analysis at the same time addressing some of their limitations.

Background

A prominent concept used to explore the nexus of parenting leave policies and childcare provision is (de)familialisation (see e.g., Ciccia, 2022; Javornik, 2014; Kurowska, 2022). Defamilialisation contends with, on the one hand, intergenerational care dependencies and corresponding obligations, and, on the other, gendered patterns of care provision. Originally conceptualised with respect to broader comparative perspective on welfare and care regimes (Ciccia, 2022), Leitner’s (2003) “Varieties of familialism” article has served as a foundation for the next two decades of early childcare scholarship (Daly, 2022; Sirén et al., 2020). Leitner’s (2003) work encompasses four ideal typical constellations of care policies, characterised by the strength or weakness of familialisation and defamilialisation efforts on display. With respect to childcare policies, Leitner’s (2003) defamilialisation is exemplified by the provision of childcare services and the relative absence of familialising measures, such as generous leave benefits. It aims to attenuate the familial obligation to care by offering alternatives, but does not aid the family in providing care themselves. In contrast, explicit familialism honours the right to provide care by offering financial support, while equating familial care with an obligation, as no alternative to familial care is offered. The final two types of familialism, implicit and optional, are also antipodal: one is characterised by a lack of policy effort in any direction – effectively leaving responsibility for childcare on the family, the other exhibits strong support in both directions, offering universal access to childcare alongside financial support for familial care.

Early on, defamilialisation would often be positioned as the goal of childcare policy in so far as it facilitates female (re-)commodification (Bambra, 2007; Cho, 2014; Szelewa and Polakowski, 2008). This approach, however, has increasingly come under scrutiny for narrowing the analytical lens (Lohmann and Zagel, 2016) and failing to accommodate for a plurality of purposeful activities (Yerkes and Javornik, 2019). More recent works, particularly those that set (de)familialisation in the capability approach framework (e.g., Javornik and Yerkes, 2020; Kurowska, 2018; Sirén et al., 2020; Yerkes and Javornik, 2019), stress policy focus should rather be on increasing the capabilities of individuals and families to choose the type of care arrangements they “have a reason to value” (Sen, 1999). From this perspective, optional familialism seems a singularly desirable policy approach, as it supports a broader choice of care arrangements compared to other regime types.

However, several authors have long expressed doubts about the real-world existence of optionality in childcare and leave system architectures (Javornik, 2014; Lohmann and Zagel, 2016; Saraceno and Keck, 2010). Furthermore, noting years of sustained expansion in family policy provision also raises the question whether we can find a real typical group of European countries that exhibit implicit familialism in relation to the ‘average worker’ today. At the same time, inequalities of access (Dobrotić and Blum, 2020) and the resulting uneven distribution of gains could suggest the supposed retreat of implicit familialism excludes less privileged groups.

For the past two decades, defamilialisation of childcare policies has been almost exclusively measured from the point of view of a standard dual earner family, with both parents on stable employment contracts, where access to full rights is assumed (for an overview of previous studies in this respect, see Appendix 1, third column). Under this assumption, the question of defamilalisation is not necessarily one of access to employment but of maintaining economic status (see Ferragina and Magalini, 2023). In addition, parenting leaves portray institutional norms that delineate expected ‘standard behaviour’ regarding care and work after childbirth (Yerkes et al., 2022). Recent scholarship suggests what we know about leave generosity (and thus opportunities for familial care) in different countries can shift considerably when we consider eligibility criteria (Dobrotić and Blum, 2020; Kurowska, 2021). Which parents get to access a ‘standard’ of behaviour is a question intersected by social class.

Dobrotić and Blum (2020) compared the inclusiveness of parental leave benefits in 21 European countries in 2006 and 2017 and found that eligibility criteria vary widely between cases, but change in inclusiveness over time had been limited. Later, Dobrotić & Iveković Martinis (2023) expanded the analysis to the other two main parenting leave types (maternity and co-parent leaves) across six countries showing inclusivity can depend on the type of right and whether it targets mothers or co-parents. Furthermore, Dobrotić (2022) has evidenced how gender and class inequalities in parental leaves intersect and compound in post-Yugoslavian countries and so did McKay et al. (2016) with respect to Canadian leave entitlements. Finally, Yerkes and Javornik (2019) and Westhoff et al. (2022) have shown that different approaches to formal childcare policy provision have different gender and class consequences as well.

Nevertheless, no study has yet approached the analysis of inequalities in access to parenting leave and childcare policies together and incorporate them into a typology. Such comprehensive perspective is crucial. First, the (de)familialising potential of early childcare policies can only be fully captured through a combination of parenting leave and childcare provision (e.g., Javornik, 2014; Leitner, 2003). Second, just as inclusivity in leave rights and childcare services differ across countries (Dobrotić and Blum, 2020; Yerkes and Javornik, 2019), it may also shift between the two policy areas within the same country.

Alongside the question of who gets to be (de)familialised through childcare policies, another relevant consideration is when this occurs with respect to the child’s age. The distinction between services for children under and over the age of three is often foundational to existing scholarship (e.g. Gornick et al., 1997; Korpi, 2000; Saraceno and Keck, 2010). While a more gradual age segregation has been so far hindered by the availability of comparative data, the first year represents a distinct period of policy intersection and should therefore be treated separately.

Finally, the analysis of childcare policies needs a gender-sensitive lens (Bambra, 2007; Saxonberg, 2013; Szelewa and Polakowski, 2023). Some authors have even advocated for replacing the term defamilialisation with degenderisation, arguing it better reflects the goals of contemporary childcare/family policies (Saxonberg, 2013). Others, however, underscore the two terms intersect but are not mutually replaceable (Kurowska, 2018). Notably, the recent European Union Directive 2019/1158 on work life balance can be considered a familialising and degenderising effort simultaneously: by putting a floor on leave rights duration, the Directive facilitates the parents’ right to provide care, but by ear-marking equal rights for fathers it exercises degenderising regulatory pressure on member states (see De la Porte et al., 2020, 2023). This highlights the need for comparative studies of childcare policies to cover both perspectives to adequately represent the landscape of childcare regimes in Europe.

All the above considerations guide our comparative analysis of childcare policies in Europe. First, the (de)familialising perspective we adopt is intersected with both social class and gender, enabling us to analyse who is supported in their choice to opt for familial or institutional care. Second, exploring childcare related entitlements by several age groups allows us to shed light on when (de)familialisation occurs. In this study, we present a novel dataset which allows us to stratify our analysis by three discrete temporalities (provisions for children under one, between two and three, and over 3 years of age). In doing so, we attend more closely to childcare gaps (Eurydice, 2019; Finch, 2021): the junction where leave retreats while childcare services are yet to fully develop. Third, by looking at childcare policies through the lens of the capability approach, we focus on conceptualising policies as means which contribute to capabilities (Sen, 1999).

Research design

Two family models

In order to extend the childcare regime analysis beyond the ‘average, dual earning family model’, we propose two empirical models corresponding to two ‘model families’, characterised by their different attachment to the labour market. The first family reflects the model couple that has been used as an implicit model in previous research. Both parents have a strong attachment to the labour market with previous earnings at the average wage for their country and would fulfil any employment history requirements. We are going to refer to this couple as family A. The second family (B), whose perspective has not been hitherto accounted for, is a couple where both members are currently unemployed and have been for at least 9 months prior to birth. Both are citizens of their country with previous employment history where they both earned two thirds of the national median wage. ¹ The second family is less privileged that the ‘average’ one, but we did not set out to explore ‘a worst-case scenario’: being unemployed for less than 1 year, the parents would not be classified as long-term unemployed (Dromundo et al., 2023) and would enter parenthood while they still qualify for most unemployment benefits (OECD, 2023a). Hence, this family’s situation makes explicit the overlap between unemployment and family policies. It is worth noting that as family policy has expanded, unemployment policies have seen retrenchment (Ferragina et al., 2012). Our goal however was to keep the two systems separate, focusing on only rights bestowed by the leave provision. We consulted OECD and MISSOC data on unemployment compensation where relevant and recorded only benefit levels separate from the unemployment support.

We acknowledge that the question of who effectively can make an unconstrained choice regarding whether, when and how to provide childcare across social class lines may not be limited solely to individuals’ labour market attachment, but may also extend to broader characteristics of their socio-economic status (Ferragina, 2023). Nevertheless, from a policy-analysis perspective, the principal criterion governing eligibility for both parental leave and in some cases equal access to institutional childcare is employment status.

Dimensions of childcare policies and data collection

We collected data along the three major dimensions discussed in childcare policy analysis: availability of childcare; generosity and gendered composition of leave entitlements; and inclusivity of leave design.

When it comes to childcare, availability and affordability constitute the foundation for capabilities; quality and flexibility become issues only once places are accessible (Yerkes and Javornik, 2019).

Generosity of leave entitlements encompasses duration, benefit, and the gendered concentration of post-birth leave rights. The data was collected twice, once from the point of view of an average earning family qualifying for full entitlements and then regarding an unemployed couple, reflecting the vast differences between the two ‘model families’.

Variables reflecting the inclusivity of leave design were collected following Dobrotić and Blum’s (2020) indicators for employment-based eligibility criteria.

Data on childcare availability

A long-standing issue in the existing literature is the lack of reliable data on childcare availability leading to the need to often use enrolment rates as a stand-in for the availability of formal childcare (e.g., Saxonberg, 2013; Szelewa and Polakowski, 2023, but see Javornik, 2014; Finch, 2021). While this solution is widely used, the reliability and data harmonisation of existing figures for enrolment rates has been questioned (Keck and Saraceno, 2011; Sirén et al., 2020). More importantly, the use of enrolment rates conflates the opportunities that policies create and the manifested outcomes (Javornik, 2014; Korpi, 2000; Yerkes and Javornik, 2019).

To address this issue, we designed a survey aimed at gathering information on the availability of affordable full time formal childcare. ² (full survey in Appendix 2) and distributed it to experts across Europe. The majority of our experts were full members of the International Leave & Policy Researchers Network.

Experts were asked to consider provision separately for different age groups and were invited to share comments and resources in a free text form. The survey was distributed to 98 researchers across 30 European countries in April and May of 2022. The chief benefit of surveying experts is in allowing aggregation of local data for a multifaceted issue where there is a dearth of existing data (Lindstädt et al., 2020; Von Soest, 2023).

We employed a rigorous validation framework ensuring internal and external triangulation (Von Soest, 2023). We collated the data and checked the responses against previous studies, international databases, as well as the additional resources and explanations provided in the survey itself and against other respondents for the same country. We consulted available data on enrolment rates (Eurostat, 2021, Eurydice, 2019, Eurydice, 2025, OECD, 2023b), cost of provision (OECD, 2020), existing childcare guarantees, and the age at which formal provision starts (Eurydice, 2019). We considered any discrepancies and disregarded survey data in cases where it cannot be reconciled with a second expert estimation or other official sources. This process reduced our final sample to 25 countries. The full data is detailed in Appendix 4.

When it comes to the unemployed couple, we also needed to check whether their access to formal childcare is not hindered further by their lack of employment. Many European countries use various criteria to ration places for children under three, including giving priority access to parents already in employment (Eurydice, 2019). We used Eurydice’s (2019) report on ECEC provision in Europe to trace top-level regulations governing priority admissions for parents in employment and children living in poverty.

Data on generosity of leave

Our primary source for leave design was the 17th International Review of Leave Policies and Related Research (Koslowski et al., 2021). The resulting data tracks post-birth leave duration at every payment rate and type of entitlement. Collecting such data required several decisions both on the nature of any employment history and on parents’ potential behaviour and circumstances. The two ‘model families’ differ on their employment situation and previous income but share other features. Both families are constructed as heterosexual couples who are welcoming a first child through birth as biological parents. Both families work, or have worked, in the private sector but are, or have been, a part of a labour union only when the country note indicates this is the case for the vast majority of the working population. For both types of families we chose policy options which maximise port-birth duration including at the expanse of payment levels.

Data on leave inclusivity

The LP&R review was additionally used to reconstruct the eligibility index proposed by Dobrotić and Blum (2020). The data collected was limited to the employment-based criteria in their work as citizenship is assumed for our model families. We also elected to produce two separate indices corresponding to the two dimensions comprising Dobrotić and Blum’s (2020) employment-based criteria. ³ The first one concerns employment history and includes the employment period needed for receiving the full benefit, alongside rules surrounding job interruptions and employer changes. The second index measures exclusion based on different employment forms and sectors. The reason to separate the two aspects was the observation that countries included in Dobrotić and Blum’s (2020) work can differ significantly in their inclusiveness scores between the two dimensions.

Construction of indicators

While space does not permit a more nuanced discussion on leave terminology, a brief explanation of the terms we have chosen is needed. Our data makes no formal distinction between the periods of leave legislation or their colloquial names, such as ‘maternity’, ‘paternity’, or ‘childcare’ leaves. Rather, the distinction strictly follows the type of rights based on who can access them. Leave reserved for the Birth Parent (BP) refers to the entitlements that are only available to the parent giving birth barring extraordinary circumstances. Leave reserved for the Second Parent (SP) likewise means non-transferable rights for a second parent. The final type is transferable leave (T). This is an entitlement that could be shared with the other parent, either through the agreement of a default holder of the right or designated as a family right. Often, shared leave is coded as an entitlement solely for the mother. On the one hand, this approach is compelling, as gender norms or gender pay gaps often convert shared leave policies into illusory opportunities for fathers (Javornik and Kurowska, 2017; Kurowska and Javornik, 2019). On the other, in policy-oriented comparative research it remains important to consider what institutional norms and incentive structures are conveyed in a regime that allows sharing and one that categorically reserves rights for the mother. ⁴

The table in Figure 1 summarises which variables are used in the analysis from the perspective of each family model.

Figure 1.

Family models operationalisation. The table indicates which variables are used in the regime analysis from the perspective of each of the two family models: entitlements for familial care and the opportunities to shift care to formal provision (defamilialisation), the gender concentration of leave rights (degenderisation), and socioeconomic stratification (eligibility).

Both models consider three age groups separately when it comes to (de)familialisation: from birth to 1 year of age, until the child reaches its third birthday, and until 6 years of age. The two datasets explore the shifting levels of (de)familialisation as children grow as the type of care the state is willing to support changes. This is represented by the proportion of the period covered by job-protected leave duration (family A), the replacement rate for that period (both models), and the ease of access to affordable childcare for the relevant age group (both models). All variables are recorded from 0 to 1, with 1 indicating full coverage for the period. ⁵

This operationalisation assumes a bifurcation of available outcomes examined simultaneously – parents can choose both to provide familial care for the whole age period while we record the totality of resources they could claim and attempt to make use of formal childcare which in practice might be less accessible to them as childcare provision is not always available to parents while they claim leave benefits (Koslowski et al., 2021). Access to childcare might also be hindered for the unemployed as many countries prioritise the access to services for parents who are already in employment (Eurydice, 2019), which is reflected in the final variable we add only for family B.

The interval answers given in the childcare survey were converted to a single number according to the following rules. If a country case received only one response corroborated by external triangulation, or more than one response that concur an interval (internal triangulation), e.g., $0\%-20\%$ , the recorded value is the average from the interval. In cases where two researchers have responded with neighbouring intervals, e.g., $0\%-20\%$ and $21\%-40\%$ , we recorded the low end of the higher one. This means the variable always takes values between 0.1 and 0.91.

The gendered access to familial care is measured by three variables showing the proportion of total FTE leave reserved for the BP, the SP, or transferable.

The two eligibility indices we standardised against the maximum points achievable in Dobrotić and Blum’s (2020) article so that they also take values between 0 and 1.

The analytical method: An overview of fuzzy clustering

The well-known deterministic, or crisp, clustering method is an exploratory data analysis tool used to discover groupings and patterns in data. Although various techniques exist to that end, the most salient property of the method for the present discussion is that the traditional clustering only allows for binary memberships where cases can belong to exactly one cluster. As a result, crisp clustering algorithms portray a dichotomous reality. Such methods are generally unequivocal in that they contain no ambiguity when the real world, and the way we evaluate it, does (Zimmermann, 2010).

When the source of imprecision emerges from the lack of sharp criteria, such as precisely defining benefit generosity, fuzzy logic offers a natural way to model phenomena (Zadeh, 1965). Accommodating fuzziness suits family policy particularly well; a palimpsest of layered development (Daly and Ferragina, 2018; Inglot, 2008), recent reforms continuously (re-)assemble hybrid systems (Ciccia, 2017; De la Porte et al., 2023).

The key feature of fuzzy clustering is the production of prototypes and gradient membership degrees. A membership degree other than 1 indicates a case lying between prototypes, in other words, a hybrid case (Ruspini et al., 2019). Cluster prototypes can be usefully conceptualised as ‘centres of gravity’. They do not map onto any real-world cases; rather, they represent distinct patterns of policy entitlements that cases within the sample gravitate towards. Prototypes are akin to ‘real types’ as for example developed by Arthur Spiethoff (Redlich, 1970). Spiethoff’s types are historically grounded series of typical configurations; ‘corresponding images’ that underline essential features of concrete reality (Cahman, 1965). In that, they are different from the ideal types at the heart of fuzzy set ideal type analysis (FSITA). FSITA uses a priori theoretical dimensions and cut-off points as deductive lenses with which to group cases (Ebbinghaus, 2012). The researcher here is heavily involved in the creation of a priori meaning (Kvist, 2007). Since our goal is to interrogate ideal types, we forgo the ideal typical aspect of FSITA, while maintaining the fuzzy-set theoretical. In that, the present effort is not dissimilar to ‘empirical robustness’ tests of Esping-Andersen’s (1990) three worlds of welfare (Ragin, 1994; Shalev, 1996).

Finally, FSITA requires the construction of cut-off points, thresholds at which a vague concept becomes more like its negation. Aspects that have been insufficiently investigated previously need to be omitted. For instance, Javornik (2014) points to a limited theoretical knowledge regarding the role of leave eligibility in ensuring continuous female employment necessitating its exclusion. The advantage of fuzzy clustering is clear in providing an analysis free of a priori thresholds, buttressing novel variable inclusion.

As an approach to unsupervised machine learning, fuzzy clustering has been utilised for a long time in various fields of study (Yang, 1993). More recently, fuzzy clustering is a prominent method in data mining and image segmentation (Naz et al., 2010; Ngo et al., 2018). The use of fuzzy clustering in social sciences is just emerging. Martín-Quintana (2022) used a fuzzy clustering approach to examine how families in Spain navigated the COVID-19 lockdown. Piccarreta and Struffolino (2024) recently made the case for fuzzy clustering on sequence data in demographics research. Helske et al. (2024) showed fuzzy clustering outperforms crisp partitioning when used as predictors in regression models.

Fuzzy clustering nonetheless shares some pitfalls with its crisp counterpart. Clustering is susceptible to outliers and variation in one aspect can overshadow others, i.e., it can have an undue weight in the model not intended by the researchers but shown in the results (Hudson and Kühner, 2010). Preparing the data in a way that reflects the intended conceptual weight of dimensions would minimize this effect. In our case, since we want all variables to contribute equally, they all take values between 0 and 1. While each dimension is sent to the algorithm on the same scale, its feature importance depends on dispersion, i.e. how much it differs across cases. In practice, this means that a low variance dimension, such as childcare provision for children over 3, exerts less influence over the output than childcare provision for children under 3.

A final precaution was taken regarding outlier cases which can affect the performance of clustering (Ruspini et al., 2019). Noise clustering algorithms can alleviate this by considering an additional ‘noise cluster’ where outliers are partially assigned. A version of noise clustering was performed on our data. We used ‘fclust’ – a toolbox for fuzzy clustering in the R programming language (Ferraro and Giordani, 2015).

Another issue we need to address is that fuzzy clustering assumes a priori knowledge about the number of clusters which we do not possess. Fortunately, Ferraro and Giordani (2015) have implemented a variety of indices that allow researchers to not specify the number of clusters in advance and for the algorithm to return only the optimal number according to the chosen validity indicator. We utilized the Fuzzy Silhouette validity measure: a fuzzy counterpart to the well-known Average Silhouette Width Criterion for crisp clustering proposed by Campello and Hruschka (2006).

Finally, we implement an improvement to the objective function proposed by Klawonn and Hoppner (2003) (for technical details, see Appendix 3). A polynomial fuzzifier is introduced ensuring zero and full membership are attainable, and outlier prototypes are not inappropriately drawn in by the gravity of the centre.

Results

The family A: The ‘average couple’

While we do find four real-world policy prototypes, they do not map neatly into the ideal typical varieties of familialism. Optional (de)familialisation is not an observable reality in any prototype the countries in our sample gravitate towards. In other words, while separate cases might provide elements of optionality in care provision, those are fleeting and truncated after the first year. According to our findings, it is difficult to substantiate attaching a single measure of conformity to one ideal type across all age groups. Ideal types can be more easily found in specific age groups than clusters of countries, underscoring the importance of analysing childcare policy by separate age groups.

This particularly applies to implicit familialism. Countries tend to be explicitly familialising in the first year. The most implicit period is contained in the second and third years as leave generosity is truncated while childcare is yet to fully develop and a childcare gap opens. Similarly, defamilialisation is the period after that, when childcare tends to be available with no parental leave alternative. There is no prototype in which the ‘average family’ model today experiences, from birth to school age, what Leitner described as lack of “any kind of familialistic policy” (Leitner, 2003: 359).

When it comes to (de)genderisation, explicitly genderising leaves are rare and reserved for relatively short entitlements. We find one degenderising prototype and two that are conspicuously implicit. In the latter case, conversion factors, such as ingrained social norms, are likely to play an outsized role in outcomes (see Pfau-Effinger, 2023).

Prototype centres

The fuzzy silhouette criterion determined the data is optimally partitioned into four clusters. ⁶

The prototype centres for family A are shown in Figure 2 in a modified violin chart. This method of displaying data maintains many advantages from earlier works utilising radar charts to analyse family policies (Finch, 2021; Javornik, 2014; Javornik and Kurowska, 2017), while condensing the information spatially.

Figure 2.

Family A prototype centers. Each shape represents one cluster prototype. The thickness of the shape at each dotted line corresponds to the value of the respective variable.

The value of each variable is represented by the thickness of the shape across its row. All variables are positively coded; the contribution of policy elements to the family’s capability set expands with the area of the shape. For example, the first three variables correspond to the first year of a child’s life. All prototypes cover this period with job-protected leave, but parents face varying degrees of financial support. At the same time, formal childcare is limited in every prototype. Prototype two, termed early optional familialism (EgFam) comes the closest to expanding the capability of new parents to include choosing employment or home care in the first year of a child’s life. It is also inclusive of a ‘non-standard’ previous employment and distributes the opportunity for home care in the most gender equal way found in our results. Juxtaposing this prototype with the fourth one underscores the fruitfulness of separating eligibility criteria: self-employment and precariousness in employment history are distinct axes of exclusion across countries.

Considering the number of countries with any degree of membership, the four clusters are of roughly similar size. EgFam has one case fully in its orbit – Iceland – but it is not otherwise small. Every Scandinavian country has some membership to this type. Prototype four, or exclusive implicitly gendered familialism (ExFam), is smaller but membership degrees tend to be higher, gravitating around a core of three highly conforming countries – Croatia, Slovenia, and Serbia. The prototype broadly conforms to the historical and present-day characteristic of leave in the post-Yugoslavian space described by Dobrotić (2022). The leave design is particularly classing, excluding parents with precarious employment, but is not explicitly genderising. Norms around parenting would then most likely ensure the birth parent stays home (Dobrotić, 2022; Grönlund and Javornik, 2014).

Prototype one, or restricted explicitly gendered familialism (ExGen), accommodates numerous lesser memberships contrasted by the very high conformity of Greece and Italy. Lastly, inclusive prolonged gendered familialism (InFam) is the most populous prototype in our sample with six CEE countries at its centre. Some support for home care is available until the child reaches 3 years of age and the leave design is inclusive with little sectoral differentiation. While the capability to provide care at home is protected beyond what other prototypes provide, the capability to work is hindered until the child reaches 3 years, at which point the access to formal childcare is drastically improved. This sudden jump in provision is characteristic of the full members of InFam in CEE (Hašková and Saxonberg, 2016).

Membership degrees

The chart in Figure 3 shows a heatmap of the membership degrees by prototype; the darker the colour, the higher the membership of a country to a given prototype.

Figure 3.

Family A membership degrees. For each country on the left, the heatmap shows its membership degree to the corresponding cluster on the top.

The inclusion of a separate noise cluster leads to membership degrees that do not necessarily sum up to 1. Cases with membership in the noise cluster, while outliers in relation to the sample, do not necessarily resemble each other. There are four countries with high membership to the noise cluster in our model – Germany, the United Kingdom, Belgium, and France.

Germany and to a lesser extend Denmark, Finland and Sweden are partially assigned to the noise cluster due to a relatively high childcare availability early on, combined with a generous leave benefit for the first year. With no independent private childcare sector and integrated governance under one authority, these countries are four cases out of the six EU member states with no childcare gap between well-paid leave and childcare entitlement and they also ensure a period of overlap between provisions (Eurydice, 2025). Despite diverging on other dimensions, these four countries best approximate Leitner’s (2003) optional (de)familialism, at least for very young children.

The UK shares some characteristics with the first prototype (low explicit familialisation, low inclusivity of sectoral differences), but combines them with the transferable leave and low inclusivity of employment history characteristic of ExFam. However, none of the original prototypes can accommodate a lack of available affordable childcare so severe and spread across all age groups. The UK best approximates Leitner’s (2003) implicit familialism.

While Germany and the UK are outliers that do not resemble each other, France and Belgium share a remarkable amount of characteristics and move together between iterations of $k$ -solutions until they escape the noise cluster completely at a five-cluster solution. They appear to represent a true hybrid regime, layering characteristics from three prototypes to enforce a distinct policy structure. Navigating this system with a stable and well-paid previous employment allows one not only to purchase formal care and achieve defamilialisation through the market, as is the case in more implicit regimes, but also to leverage the long job protection and attain some level of familialisation while maintaining employment status (see Bister et al., 2024). Hook (2015) has previously noted this optionality in France and similarly relates it to deep polarisation based on class. These results also underscore a bidirectional movement in family policy characterised by commodification for lower-class women but increasing freedom for middle- and upper-class ones (Ferragina, 2023).

The family B: The unemployed couple

Known groupings break down when we switch to the second (unemployed) model family. We find six prototypes for family B and the countries that approximate them do not tend to be grouped together when it comes to an ‘average family’ model. From the unemployed couple’s point of view, implicitness in family policy is widespread. In the absence of leave benefits, childcare becomes the only option but even if a place is available and affordable, the unemployed couple might find priority is given to employed parents.

However, parental leave is not a right that is necessarily linked to current employment, and we find that unemployed people can still access it in many countries. There is one prototype that approximates explicit familialism. Notably, it also explicitly genderises – the leave rights are concentrated in the birth parent, a characteristic that is not there for the employed, showing the spread of degenderisation is unequal and a fall back to maternalism (see also Duvander and Koslowski, 2023).

Prototype centres

The chart in Figure 4 shows the prototypical shapes found for the unemployed family.

Figure 4.

Family B cluster prototypes.

Countries shift not only their levels of generosity but also their legislated concentration of rights between an average earning couple and an unemployed one. Overall, leaves for the ‘average couple’ are more degenderising. This is because the second parent’s rights are more tightly reliant on secure employment than those of the birth parent. Dobrotić (2022) also finds that, in most post-Yugoslavian countries, gender equalising reforms only benefit employed fathers. We can corroborate this and extend it beyond these countries.

Membership degrees

The chart in Figure 5 shows the heatmap of membership for family B.

Figure 5.

Family B membership degrees.

Scholarship about how leave policies towards the unemployed develop in comparison with the average earner is limited. However, our findings corroborate those of Dobrotić (2022) when it comes to the post-Yugoslavian countries in our sample. Our post-Yugoslav cluster is scattered across the new prototypes. Post-1990 reforms in Croatia have favoured leave benefits for unemployed mothers, while Serbia privileges middle and high income earners (Dobrotić, 2022). Slovenia, on the other hand, provides a rare example of little difference in the treatment of employed and unemployed parents, alongside Czechia. Czechia is consistent in portraying familialisation until the child is three as an ideal which scholars have traced back to the 1970s (Hašková and Dudová, 2017). By providing flat rate benefits, Czechia gives more incentives for the unemployed to provide informal care relative to any previous incomes.

Conclusion

Our findings show that optional (de)familialism is still mostly an ideal, while implicit familialism for the ‘average parent’ is rare in Europe today. It is from the point of view of unemployed parents that we can see the retreat of implicit familialism has been uneven. In addition, concrete ideal types can more readily be found by subdividing the first stages of a child’s life into distinct temporalities where institutional norms are clearly conveyed. Incorporating eligibility criteria enriches the analysis of (de)familialisation and contributes to showing how access can be stratified based on ‘nonstandard’ employment, thus helping to contextualise leave generosity.

We can also note many similarities to previous classification attempts, which were, however, confined to the perspective of the more privileged, employed parents. Our third family A prototype is populated by CEE countries showing that, considered on a European level, they maintain similarities (Chybalski and Marcinkiewicz, 2021; Saraceno and Keck, 2010, see also Dobrotić and Stropnik, 2020; Szelewa and Polakowski, 2020). The inclusion of previously underexplored cases underlines how similar historical trajectories distinct from Western European patterns of development (see Inglot, 2008) and present-day family policy learning within CEE (Buzogány and Varga, 2018; Orenstein and Bugarič, 2022, see also Graff and Korolczuk, 2022a) play out across the region. Our findings are consistent with previous works when it comes to France and Belgium which are most often grouped together (Korpi, 2000; Leitner, 2003; Misra et al., 2007; Saxonberg, 2013). The position of Ireland in the gravitational orbits of both Southern countries like Italy and the UK also echoes previous research (Bettio and Plantenga, 2004; Ciccia and Verloo, 2012). The position of Germany as an outlier that also straddles two clusters is congruent with recent typologies that offer diametrically opposing labels: degenderising (Finch, 2021) and genderising (Szelewa and Polakowski, 2023). Lastly, our results serve to confirm those previous typologies that find the Nordic countries not to be a monolith (Kröger, 2011; Saraceno and Keck, 2010; Saxonberg, 2013, among others). At the same time, the addition of eligibility considerations serves to separate Sweden as the most exclusive of the Scandinavian countries (Dobrotić and Blum, 2020; Kurowska, 2021). So, while Sweden is often presented as an ideal in previous typologies (Szelewa and Polakowski, 2023; Saxonberg, 2013; Leitner, 2003, among others), this vision of defamilialisation can also be highly exclusive.

Our empirical findings also serve to demonstrate the advantages of fuzzy clustering in capturing real-world complexity. Gradual membership allowed us the nuance of classification absent in crisp partitioning. Fuzzy classification algorithms are also well suited for comparison across time and for revealing the evolution of prototypical policies. Furthermore, the existing variety of algorithms can handle different types and sizes of data (e.g., for a textual data Lazhar, 2019; Jiang et al., 2010). Fuzzy clustering algorithms thus hold a vast potential for future social science applications.

There are some limitations to our approach that bear mentioning. Regarding the data for the availability of childcare, while we attempted to corroborate every survey response we coded, this was not always possible. Additionally, sometimes the respondent would indicate that availability differs significantly according to where parents live; an important spatial inequality (see Scherer and Pavolini, 2023) we could not incorporate.

Cash transfers to families outside of the parenting leave system is a policy area often neglected in typologies of childcare (Leitner, 2003; Javornik, 2014; Szelewa and Polakowski, 2008; Szelewa and Polakowski, 2023, but see Lohmann and Zagel, 2016; Korpi, 2000), an omission repeated here. On the one hand, our goal was to closely follow well-established ways of partitioning childcare provision for the ‘average family’ to show how inequalities of access disrupt known country groupings. On the other, child benefits, often distributed as they are via the tax system (De Deken, 2017; Ferrarini et al., 2012), present a challenge for large-N comparative research (Clarke et al., 2022), placing them outside the scope of the present effort but offering a fruitful line of inquiry for future research, particularly pertinent when it comes to the ways in which fiscal welfare enforces existing inequality (see Sinfield, 2018). Incorporating data on intergenerational care dependencies (e.g., Le Bihan et al., 2019; Leitner, 2003) represents another avenue for enriching the analysis of (de)familialization.

Furthermore, the relative lack of data on how unemployment and family measures intersect means the present effort lacks consideration about how activation requirements might be relaxed (and unemployment benefits prolonged) as a direct result of welcoming a child (but see Kowalewska, 2017). As data gaps fill, future research might overcome this limitation and extend the analysis further.

Finally, there are additional axes of exclusion that we could not address in the present article. The perspective of leave and childcare entitlements of same-sex couples (Wong et al., 2019), non-citizens (Dobrotić and Blum, 2020; Duvander and Koslowski, 2023; Eugster, 2018), or single parents (Bartova et al., 2022) would likely lead to identification of other regime prototypes and country clusters. We thus encourage future studies to extend our analysis using fuzzy clustering methods, to other types of less-privileged groups of parents and further deepen the understanding of inequalities in the way parenting leave and childcare architectures in Europe are (de)familialising and (de)gendering care.

Supplemental Material

Supplemental Material - European childcare regimes beyond the ‘average family’ model: A fuzzy clustering analysis

Supplemental Material for European childcare regimes beyond the ‘average family’ model: A fuzzy clustering analysis by Dimitrina Ivanova and Anna Kurowska in Journal of European Social Policy.

Supplemental Material

Supplemental Material - European childcare regimes beyond the ‘average family’ model: A fuzzy clustering analysis

Supplemental Material for European childcare regimes beyond the ‘average family’ model: A fuzzy clustering analysis by Dimitrina Ivanova and Anna Kurowska in Journal of European Social Policy.

Supplemental Material

Supplemental Material - European childcare regimes beyond the ‘average family’ model: A fuzzy clustering analysis

Supplemental Material for European childcare regimes beyond the ‘average family’ model: A fuzzy clustering analysis by Dimitrina Ivanova and Anna Kurowska in Journal of European Social Policy.

Supplemental Material

Supplemental Material - European childcare regimes beyond the ‘average family’ model: A fuzzy clustering analysis

Supplemental Material for European childcare regimes beyond the ‘average family’ model: A fuzzy clustering analysis by Dimitrina Ivanova and Anna Kurowska in Journal of European Social Policy.