Determinants of a Silent (R)evolution: Understanding the Expansion of Family Policy in Rich OECD Countries

This paper contributes to the comparative social policy literature in two ways. First, we use multiple correspondence analysis in order to assess the different directions and the degree of (employment-oriented) family policy change over the past three decades in 18 rich Organisation for Economic Co-operation and Development (OECD) countries. Second, we perform a series of correlations to identify the core drivers of these developments. Our main findings—based on five international datasets—are: (i) we have been witnesses of a significant expansion of family policies over the past three decades in almost all countries analysed, although the degree of change (distinguished by first-, secondand third-order change) differs across the OECD area; and (ii) whilst in the 1980s and 1990s social democracy and organised women were key drivers of family policy expansion, during the 2000s public opinion, that increasingly seems to support a “modernised” family lifestyle in which mothers are employed, seems to have played an essential role in explaining policy change.


Introduction
According to many welfare state analysts we entered an "era of permanent austerity" some time ago (see, e.g. Pierson 2001); the economic crisis since 2008 has significantly exacerbated the pressures on public social policies, with the consequence that the expansion and even maintenance of many programmes have been made more difficult. However, during this era of austerity family policies oriented to promoting maternal employment have been significantly expanded in rich Organisation for Economic Co-operation and Development (OECD) countries (Ferrarini 2006;Lewis 2009;OECD 2007). Despite cross-country variation in the degree of "generosity" and duration of parental leaves (Ray, Gornick and Schmitt 2010) and childcare services (Daly 2002(Daly , 2005Hegewisch and Gornick 2008;OECD 2007), women's increased labour force participation has been facilitated by improved public support, 2 even in those countries that have previously been identified as featuring a "strong male breadwinner model" (Fleckenstein and Lee 2014;Lewis 1992).
Echoing Goldin's (2006) notion of a "quiet revolution" that transformed women's employment, education, and family, Esping-Andersen (2009) argues that a successful societal and political completion of this process-which he terms an "incomplete revolution" in women's roles-requires further social policy development to avoid deleterious equilibria and the widening of other forms of inequality. 3 In addition, a number of scholars have explored the socio-economic, political, and societal drivers and functional underpinnings of this expansion (Abrahamson 2010;Bleijenbergh and Roggeband 2007;Bolzendahl 2009;Daly 2010;Fleckenstein 2011;Lewis and Campbell 2007;Morgan 2006Morgan , 2013Seeleib-Kaiser 2010). However, there have been few attempts to systematically analyse the developments in maternal employment supporting policies across a large group of rich OECD countries (but see Ferrarini 2006;Korpi, Ferrarini, and Englund 2013;Lambert 2008;Montanari 2000;Morgan 2013).
In this respect we contribute to the literature in two ways. First, we use multiple correspondence analysis (MCA) in order to assess the direction of change, and how this might be characterised using Hall's (1993) concept of first-, secondand third-order change in analyzing maternal employment-oriented family policy, 4 over the past three decades in 18 rich OECD countries. Second, we perform a series of correlations to identify the potential core drivers of these developments. In accordance with comparative social policy literature we hypothesise a strong effect of social democracy and organised women and a growing importance of cultural values in the explanation of family policy expansion.

State of the Art
Family policy 5 is said to have a multiplicity of functions: horizontal redistribution, the enhancement of individual choices, increasing fertility rates, supporting economic growth and productivity, as well as reducing gender inequalities. 6 From a societal perspective family policies can contribute to "horizontal redistribution," between generations as well as between households with, and without, children (Kaufmann 2012;Saraceno 2007); to favour individual choices by supporting the reconciliation between care and paid work (Beck 1992;Lewis 2009); and to reduce the costs of having children and child poverty (Letablier et al. 2009). From an economic perspective, employmentoriented family policy is part of an overall redesign of welfare states geared to foster "active citizenship", also among mothers who were formerly not employed (see Lewis 2009), through the development of an "enabling state" (Gilbert 1989). More generous family policies are said to lead to higher employment rates for women (Huber and Stephens 2001;Letablier et al. 2009), In this respect, we provide a comprehensive map of change in several key family policy programmes over time (1980 -2008) and we separately investigate the drivers of this expansion for each decade. In our analysis we use "generosity" as a multi-dimensional measure of various explicit family policies, including statutory entitlements relating to the duration and benefit level of maternal, parental, and childcare leaves, as well as child allowances, and the spending on childcare. "Modernisation" is used as a concept to capture the development towards an employment-oriented family policy, especially through the expansion of childcare services. 11 By using the terms "generosity" and "modernisation" we do not intend to suggest that the change in family policy led to greater de facto gender equality. 12 Scholars discussed the expansion of welfare states using a very wide range of explanations: functional approaches ("the logic of industrialism", "modernisation", "neo-Marxist explanations"), arguments based on the claim that "politics matter" (via "democracy", "popular protest", "the social-democratic model", "influence of other parties", or "power resources"), the evolution of the transnational context ("international trade", "international competition", "international cultural modelling"), state-centred arguments ("state as an independent actor", "path dependency") and ideational approaches (for a review see Amenta 2003;Flora and Heidenheimer 1976;Nullmeier and Kaufmann 2010;Skocpol and Amenta 1986). These theories have also been reflected in family policy research, differentiating between socio-economic, political, ideational, and international factors as potential drivers for policy change (see for an overview Mätzke andOstner 2010a, 2010b). Furthermore, the emergence of "new social risks" (NSR) (Bonoli 2005;Taylor-Gooby 2004), low fertility rates (Esping-Andersen 2009), changing skill "requirements" (Fleckenstein, Saunders and Seeleib-Kaiser 2011), and the demand for gender equality (Bolzendhal 2009) seem to underpin the political "necessity" to adapt welfare state arrangements in de-industrializing economies, including an expansion of (employmentoriented) family policies.
Although from a functional perspective one might expect the adaptation of welfare state arrangements to happen irrespective of the political system and power constellations, power resources theory, among others, has highlighted the importance of organised labour and political parties in the development of welfare states (Korpi 1983;Stephens 1979). While Iversen and Stephens (2008) argue that social democracy cum organised women's movements can explain the expansion of family policy in OECD countries, a number of scholars (Huber and Stephens 2001;Kittel and Obinger 2003;Seeleib-Kaiser 2002) have emphasised that the role played by partisanship in explaining social policy developments has been declining significantly. For instance, Seeleib-Kaiser (2010) and Fleckenstein (2011) have shown how electoral competition and political entrepreneurship have played a crucial role in changing the policy positions of the German Christian Democrats vis-à-vis family policy in the 2000s, from being a staunch supporter of a strong male breadwinner model towards being in favour of an adult worker model. In addition, a number of researchers have argued that political culture plays an important role in understanding cross-national differences in welfare state development (Lipset 1996);recently, Pfau-Effinger (2005) has highlighted the importance of culture and ideational change in understanding welfare state developments, while Brooks and Manza (2007) and Wendt, Mischke and Pfeifer (2011) have emphasised the importance of public opinion and policy preferences.
Despite important qualitative work, which has identified policy drivers for certain countries 13 (e.g. Daly 2010) and in some small n-comparisons Korpi, Ferrarini and Englund 2013;Morgan 2013;Orloff 2006), research aiming at generalisable explanations continues to be relatively scant. Based on recent socio-economic changes, such as rapid deindustrialisation and unsettling of dominant political cleavages (Emmenegger et al. 2012;Thelen 2014), we hypothesise potential period effects explaining family policy change, i.e. drivers of family policy expansion that have been dominant in the past might be less relevant in subsequent decades. 14

Methods and Data
MCA was pioneered by Bourdieu (1979) to map different types of individual consumption preferences onto a continuous space without resorting to predictive techniques (Le Roux and Rouanet 2010). We use MCA to assess the direction and degree of change in family policy (also see Ferragina, Seeleib-Kaiser, and Tomlinson 2013) and a series of correlations to assess the determinants of change over three decades among 18 OECD countries. Our analysis is based on five international comparative datasets: 15 1. the "Comparative Family Policy Database" (Gauthier 2011); 2. the "OECD Social Expenditure Database" (OECD, 2010); 3. the "Family and Changing Gender Roles Survey" (ISSP 1988(ISSP , 1994(ISSP , 2002; 4. the "LABORSTA dataset" (ILO 1997(ILO -2008; and 5. the "Comparative Welfare States Data set" . MCA visualises change in a Cartesian space and can deal with non-normal distributions of macro-institutional data (Le Roux and Rouanet 2010). To identify the main drivers of expansion in family policy we propose a series of correlations over three decades (the 1980s, the 1990s, and the 2000s), thereby avoiding the conceptual problem of the repeated consideration of data without variance 16 (for a detailed explanation, see Shalev 2007).
Describing the Expansion of Family Policy with MCA MCA simultaneously associates "variables" (categorical variables that capture the "generosity" of family policy) and "cases" (18 rich OECD Understanding the Expansion of Family Policy countries). 17 Hence, if countries within a dataset have particular characteristics represented by categorical variables A, B, and C coded into low, medium, and high, then the corresponding categories within A, B, and C can be plotted onto a two-dimensional space (figure 1). Countries and categories are said to occupy a joint space; thus it is possible to interpret the "category plot" (figure 2) and position of each country onto the same space (figure 3). For example, assume a dataset has three categorical variables: 18 A (the total number of weeks for various leaves associated with maternity and child rearing), B (the average replacement rate for various leaves associated with maternity and child rearing), and C (the child allowance) are coded "low", "medium", or "high" (the categories for each variable). MCA jointly considers countries and categorical variables in the space. Now, suppose that a country within the data scores as follows: High A, High B, and Medium C (figure 1). This means the country displays a "long" duration for leave, a "high" replacement rate for leave, and a medium level for the child allowance. By triangulation the country can be placed within the category plot as shown in figure 1. Furthermore, by using longitudinal data we are also able to plot countries over time as the categorical variables shift temporally. Finally, a measure of how well the data fit the model is given by the percentage of variance explained. Because the method is not probabilistic, there is no requirement for a large dataset or a random sample (Le Roux and Rouanet 2010).    , 1980s, 1990s, and 2000s. Notes: (1) Ost, Austria; Ger, Germany; Bel, Belgium; Fra, France; Nor, Norway; Den, Denmark; Net, The Netherlands; Swe, Sweden; Fin, Finland; Nzl, New Zealand; Aus, Australia; Ire, Ireland; US, United States; UK, United Kingdom; Jpn, Japan; Can, Canada; Ita, Italy; Swi, Switzerland. (2) Finland 2 represents the position of Finland in the 1980s, Finland 3 in the 1990s and Finland 4 in the 2000s and so forth for the other countries. Source: Authors' elaboration based on Ferragina et al. (2013).

Understanding the Expansion of Family Policy
On this methodological basis, we adapted MCA to the study of family policy in three steps. (i) We recoded all continuous, count, and ordinal variables into quartiles before estimating the model in order to avoid the problems associated with poorly distributed variables. (ii) We interpreted the Cartesian spaces emerging from the cloud of indicators and the position of each country, rather than the axes (see figure 2). (iii) Finally, we plotted average scores for three decades (1980s, 1990s, and 2000s) and 18 rich OECD countries. In this way, the trajectory of each country is easy to interpret, because the characteristics of the spaces identified by the category plot remain the same over time.
Although not ideal, our temporal analysis is based on averages for decades in accordance with Sabatier (1988), who suggested that one can more realistically capture policy change by analysing a period of 10 years or more rather than simply focusing on one year data points.
Following Esping-Andersen's widely-used typology of welfare regimes, countries can be characterised as belonging to the liberal, Christian-democratic, or social-democratic regime type 19 (Esping-Andersen 1990). In addition, the critiques of scholars of gender and welfare, such as Lewis (1992), Saraceno (1994), and others (e.g . Fraser 1994;Korpi 2000;Mahon 2002;Orloff 1993;Sainsbury 1999), have emphasised the necessity of a "fruitful dialogue" between proponents of welfare regime theory and gender typologies (Mahon 2001). Building on his own work on partisanship and power resources and feminist work on welfare regimes, Korpi (2000) introduced a typology of gendered welfare state institutions and family policy with models supporting traditional or dual-earner families, or refraining from offering any families explicit state support (general family support, dual-earner support, or market-oriented policies). These types are linked to distinctive configurations of partisan support. Hence, we bridge the categorisation of Esping-Andersen (1990, 1999 and Korpi (2000), to define the space and position of each country in the MCA, by also considering the care dimension. It is important to note, that although our approach is informed by welfare regime theory, we do not simply assign countries to clusters. Rather, their position in the space is defined by the correspondence of the indicators chosen to measure the "generosity" of family policy. Thereby, taking into account the multi-dimensionality of family policy (benefit levels, duration of leaves, and care services), our MCA allows us to analyse the evolution of family "policy packages" (Kamerman and Kahn 1994) over three decades.
The Indicators. Building on our definition of explicit family policy we have included the following variables in the model: (1) the weeks of leave (maternity, parental, and childcare), (2) the average replacement rate of three types of leave (maternity, parental, and childcare), (3) the child allowance (for the first child), and (4) the expenditure on family services standardised by the number of children. For the first three variables we relied on the updated "Comparative Family Policy Database" . In this database maternity/parental leave refers to leave granted in connection with childbirth and includes a period of leave prior to and after childbirth. Childcare leave is defined as an optional leave taken after maternity/parental leave. There is a strong inconsistency across countries on how these leaves are defined and this has of course an impact on our operational measure (Moss 2012). After the 1990s the distinction between these two types of leaves has become less clear for several reasons (Gauthier 2011, 2-3). (i) Some countries have made part of the "maternal" leave available to fathers, therefore it is harder to distinguish between maternal and parental leave. (ii) In some countries parental leave can be used in a flexible way until the child is 8 years old. This means that parental leave is no longer tightly linked with childbirth and is closer to the more comprehensive definition of childcare leave. 20 (iii) The term parental leave has been extended to encompass leave for family reasons (i.e. reasons related to the care and upbringing of children).
Hence, we integrate the duration and replacement rates of maternal, parental, and childcare leave, into two single indicators: "length of leave" and "replacement rate of leave". This means that we will not be able to investigate, whether, and to what extent, leave is shared between mothers and fathers, which feminist scholars have identified as critical to the construction of gender relations (e.g. Gornick and Meyers 2009), but we will assess the overall "generosity" of the institutional arrangements for families. 21 According to our definition, family policy can be "quite generous" but still insufficient to effectively address the impact of gender inequalities (here echoing Korpi's [2000] claims about the policy model of "general family support").
Due to the lack of internationally comparable longitudinal datasets for childcare provision, we have created a proxy variable, derived from the OECD public social expenditure data for family services (OECD 2010). The indicator is calculated as the expenditure for family services per child aged five and below (as a percentage of GDP per capita in PPP). Although not perfect, this measure allows us to include the service dimension (often disregarded in longitudinal comparative research, but see Lambert 2008). Tables 1 and 2 show that OECD countries over the period 1980 -2009 have increased expenditure for services, in particular childcare and family services, more significantly than outlays for cash benefits. We discuss the implications of this finding in the "Results" section.

Assessing the Drivers of Family Policy Expansion
In order to identify the determinants of family policy expansion, we propose a series of correlations (for the 1980s, 1990s, and 2000s) between our dependent variables, measuring the "generosity" of family policy, and the usual independent variables considered to explain welfare state development (Huber and Stephens 2001). Building on Shalev's argument for simple and descriptive techniques (such as MCA) and Sabatier's (1988) suggestion to consider longer time periods to capture policy change, we include the average    decadal value for the independent variables using simple correlations rather than multiple regression models. 22 The Indicators. The dependent variables to explain the "generosity" of family policy over time are those previously described and used in the MCA, plus public expenditure for family policy as a percentage of GDP (OECD 2010). The independent variables of the model are derived from three general sets of factors determining policy change: (i) socio-economic factors and conditions; (ii) political and institutional explanations; and (iii) ideas. The socio-economic variables included are: "female labour market participation rate", 23 "dependency ratio", "total fertility rates", "de-industrialisation index", and "the percentage of people/women that are employed in jobs that require high general skills". 24 These variables capture the most important socio-demographic factors identified in the literature to explain the shift of family policies (Gauthier 1996. However, their impact on the expansion of family policies might be mediated or delayed by other factors. For example, demographic changes might have delayed effects on the expansion of family policy due to the temporal dissonance between the fertility drop and the public recognition of this decline as a social problem (Seeleib-Kaiser and Toivonen 2011). The de-industrialisation index is a way to capture the emergence and intensification of NSR (Bonoli 2005). An increase in the proportion of service sector employment might contribute to an expansion of employment-oriented family policy, building on a functionalist logic, whereby welfare states aim to minimise social risk. Following on the work of Estevez-Abe (2005) and Fleckenstein and Seeleib-Kaiser (2011), which highlights that employers requiring workers with high general skills are more likely to support an expansion of employment-oriented family policies, and assuming rational actors, we have used the variable "percentage of women employed in occupations requiring high general skills" as a proxy measure for the impact of "organised employers" on family policy. 25 We hypothesise that family policy will expand as a result of "organised employers interests" to protect their investment in human capital.
Political variables included in our model are: "union density", "per cent of the vote for the left, the non-religious centre, Christian-democratic, or right parties 26 ", and "women elected to parliament as a per cent of the total number of MPs". These variables evaluate the traditional "politics matter" hypotheses. The power of the left and "organised labour" are considered the main drivers explaining the expansion of welfare states by a number of influential analysts (e.g. Huber and Stephens 2001). However, over the last decades partisanship seems to have lost importance in explaining variation in welfare states (see Seeleib-Kaiser 2002;Seeleib-Kaiser, Dyk and Roggenkamp 2008). Feminists have identified a number of factors associated with gender relations as significant in the expansion and character of family policy. Various indicators have been used to evaluate the effect of gender in politics: women's representation Understanding the Expansion of Family Policy in parliament, women's employment rate, active presence of women's equality movements, and the gender machineries (or equality units) in government administrative bureaucracies (Bleijenbergh and Roggeband 2007;O'Connor 1993;Orloff 2009;Walby 2004;Zippel 2006). According to Bolzendhal's (2009) empirical study in 12 OECD countries, political participation of women and women's employment rate are the most important factors in explaining the expansion of family policy and welfare states more generally.
Societal "modernisation" is captured through our public opinion variable. The variable measures the preferences of people relating to the employment of mothers with pre-school children. 27 Crompton and Lyonette (2006) and Pfau-Effinger (2005) have highlighted the persistence of different attitudes in OECD countries in relation to mothers' employment. We hypothesise that in countries where mothers' employment is not perceived as detrimental for the development of young children, family policy-especially public childcare provision-is likely to be more generous and supportive of women's employment. Finally, we could also be witnesses of a virtuous circle, as generous family policy might foster increased public support, which leads in turn to a further policy expansion. Budig, Misra, and Boeckmann (2012) have argued that cultural attitudes toward women's employment might be seen as a mediating factor between family policy and women's earning. A higher degree of "generosity" of parental leaves and public childcare is associated with higher earnings for mothers when cultural support for maternal employment is high, but have less positive or even negative relationships with earnings where cultural attitudes support the male breadwinner/female-caregiver model (Keck and Saraceno 2013, 5).

An "Incomplete Revolution?"
Our MCA 28 allows the definition of the characteristics (in accordance to the indicators used to measure family policy generosity) of each "geographical space" presented in figure 2. When we talk about liberal, Christian-democratic and social-democratic spaces we do not refer to specific countries but to spaces in which the 18 countries move over the period analysed. For example, if Germany moves to the "social-democratic" space, it does not mean that Germany has become a social-democratic country from an outcome perspective, but rather that public provision of childcare as well as the generosity of parental leave has increased over time to a similar level found in countries that originally occupied the "social-democratic" space.
In this respect, the "cloud of indicators" in the four quadrants can be interpreted as follows (figure 2): the top right quadrant can be defined as a liberal space, characterised by residual public family policy. It is "a liberal space" which is characterised largely by the absence of explicit family policy (what Korpi [2000] has defined a "market-oriented" family policy orientation). The bottom right quadrant is characterised by low to medium comprehensive childcare services and medium to low child allowance-this quadrant seems to represent a "hybrid space" (hybrid because it contains elements of the Liberal and the Christian-democratic spaces, even if closer to the latter). The bottom left quadrant is characterised by a medium provision of childcare services, a medium to long duration of and medium level of replacement rates for leaves, but high child allowances-we can clearly speak of "a Christian-democratic space" which is transfer heavy (what Korpi [2000] has defined "general" family policy orientation). Finally, the top left quadrant is characterised by comprehensive childcare provision, long leave arrangements with high replacement rates, and medium high levels of child allowances-an approach that is often associated with a social-democratic family policy package; this is a space geared towards a dual-earner model (Korpi 2000).
A dynamic MCA, using average values for each of the decades, unveils significant policy changes, with countries moving from one quadrant to another over time ( figure 3). The data suggest that we are indeed witnesses to a (r)evolution, although perhaps incomplete (Esping-Andersen 2009). This whole process of transformation can be characterised as a "socialisation" of family policy, featuring an expansion of childcare services and leave entitlements (for the interpretations of the spaces, see figure 2). However, the degree of change is quite different among the 18 countries analysed. 29 In table 1 we classify countries in three groups according to the degree of change of their employment-oriented family policy, following Hall's (1993) categorisation of first-, second-, and third-order change. A first-order change implies that the country did not substantially change the family policy package, remaining in the same space over time. A second-order change implies a stronger movement from one quadrant close to another or from the border of a quadrant to another quadrant, largely indicating a change of various instruments within the policy package. Finally, a third-order change implies the movement from one space to another, suggesting not only a change of a policy instrument, but a change of the paradigm underlying the policy package.
The only country that has not undergone a significant change in "explicit" family policy is the United States (Moss 2012). Here we find no statutory paid maternal, parental, and childcare leaves at the federal level and no sustained policy initiative over time to introduce these (although there are some relevant developments at state level). Furthermore, investment in public childcare services has continuously been at the lowest level among the 18 OECD countries OECD 2010). However, we note that the United States relies on rather different policy instruments from most of the other countries analysed in order to support families (Morgan 2006;Orloff 2006). In particular, the expansion of the Earned Income Tax Credit and the elimination of the Aid to Families with Dependent Children have been tied to increasing mothers'especially single, poor, but not only-employment levels. Childcare tax credits, Understanding the Expansion of Family Policy aimed at better-off families, might generate similar effects to publicly provided services by offsetting some of the costs of private provision. Moreover, the enactment of the Family and Medical Leave Act in 1993 was quite significant politically, even though it is an unpaid leave.
New Zealand, Switzerland, and the United Kingdom 30 are characterised by a first-order change in their employment-oriented family policy and have remained within the liberal space over the past three decades. These countries are still characterised by relatively short leaves and low public investment in childcare. Also Belgium and Italy, while undergoing a first-order change over the last three decades, remained within the Christian-democratic space (Morgan 2013). They are characterised by a medium level of childcare service provision and a short to medium length of family-related leaves. Finally, the other three countries that are characterised by a first-order change in employment-oriented family policies are France, 31 Finland, and Sweden. These countries occupy the social-democratic space based on their generous leave (long leave with high replacement rates) and comprehensive childcare arrangements. 32 Australia, Canada, and Austria are characterised by a second-order change with regard to their employment-oriented family policy package. Australia moved from the liberal space closer to the Christian-democratic space due to more generous leaves and a higher investment in childcare. Similarly, Canada moved into the Christian-democratic space, starting from a position located between the liberal and the Christian-democratic space in the 1980s, due to more generous leaves and a higher investment in childcare. Finally according to the measurements employed in our MCA Austria moved into the socialdemocratic space, especially because of a higher investment in childcare (see tables 1 and 2).
The four countries that have undergone third-order changes are Germany, Norway, Ireland, and Japan. The move of Germany to and the positioning of Austria in the social-democratic space would seem rather surprising according to common welfare regime typologies and their earlier support of male breadwinner models. However, our results confirm Leitner's (2010) findings, whereby Austria began a process of rapid family policy expansion, which was subsequently outpaced by the developments in Germany during the past decade. Starting from the late 1990s, Germany developed a "modernised" approach to family policy and since 2002 pursued what has been defined a "sustainable family policy model" (Nachhaltige Familienpolitik) (Ostner 2010;Rürup and Gruescu 2003;Seeleib-Kaiser 2010). This model considers children as society's future assets, aims to increase the fertility rate by supporting parents, and tries to reduce child poverty by boosting mothers' employment. These policy changes led to a tripling of spending for maternity and parental leave benefits and a more than doubling of spending for day care/home services (see table 2). Although both Austria and Germany are located in the social-democratic space for the 2000s due to their increased "generosity" of childcare services and leave entitlements, this does not mean they have eliminated all tax incentives for the male breadwinner model-indeed Germany continues to have a joint taxation system. Moreover, in terms of outcomes, significant gender inequalities persist compared to countries which have been characterised by social-democratic policies for longer. For example the gender wage gap continues to be much higher, women's (full-time) employment rates much lower, and representation of women managers on company boards is almost non-existent (see OECD 2011).
Norway also moved into the social-democratic space, largely due to the fact that spending for maternity and parental leave tripled, spending for day care/ home services quadrupled, and other benefits tripled (see table 2). Although the policy structure along these dimensions is now similar to those in Sweden (a dual-earner model), there is a consistent difference concerning gender relations due to the higher provision of the so-called "cash for care" schemes, which allows one parent to stay at home for childrearing 33 (Duvander, Lappegård, and Andersson 2010). This finding supports the claims of Olk (2009, 48), highlighting that Norway can be characterised as a latecomer regarding the expansion of public childcare, when compared to the other Scandinavian countries (Sweden and Denmark in particular).
Ireland and Japan moved from the liberal to the Christian-democratic space. During the period 1980 -2009 expenditure for child allowance and maternity/parental leave in Ireland doubled, other cash benefits including childcare leave increased more than five times and day care/home services, including public childcare, increased from a very low level to 0.8% of GDP (see table 2). The positive state of public finances and the unprecedented employment growth during the 1990s largely contributed to this change (Daly and Clavero 2002). In Japan the expansion is certainly less remarkable than in Ireland, but family policy has clearly departed from the very low levels of "generosity" experienced during the 1980s. Spending on child allowances tripled and spending on maternity/parental leave and day care/home services has doubled, although from a very low base (see table 2). However, these policy changes have not unsettled the Japanese male breadwinner model, which remains deeply rooted in public policy (especially the tax system), cultural norms and business practices (Schoppa 2010;Seeleib-Kaiser and Toivonen 2011). In addition, although our multi-dimensional family policy analysis locates Japan in the Christian-democratic space, we acknowledge the absence of a Christian-democratic party and the overall hybrid nature of the welfare state (see Esping-Andersen 1997).
Prima facie Denmark and the Netherlands seem to counter the overall trend of family policy expansion. This counterintuitive "finding", however, needs to be largely understood as an artefact due to the fact that Gauthier's comparative family policy data set only includes data for publicly funded leave benefits not taking into account the considerable degree of occupational and sectoral arrangements, which are often quite comprehensive as they are regulated through collective bargaining agreements (for Denmark see Abrahamson [2010] and for the Netherlands, see Knijn and Saraceno [2010]). Denmark and the Netherlands have relatively high "real" replacement rates, if these occupational leave policies are included (Groenendijk and Keuzenkamp 2010;Rostgaard 2010).
Let us summarise our argument thus far. According to some scholars (e.g. Bleijenbergh, Bussemaker, and de Bruijn 2006;Bruning and Plantenga 1999;Morgan and Zippel 2003), we would expect countries classified as "male breadwinner" or "caregiver parity" models to mainly focus on leave (and working time) arrangements as well as child allowances. However, even in these countries social care policies have moved in the direction of providing more services that allow mothers to work (Bleijenbergh and Roggeband 2007, 439). If we consider the starting points of the various countries in the 1980s, the changes in family policy have been remarkable and are grosso modo in line with some of the demands feminist scholars made many decades ago (see, e.g. Sassoon 1987). 34 This development is even more remarkable, if we take into account the parallel retrenchment that is taking place in other social policy domains (see., e.g. Ferragina, Seeleib-Kaiser, and Tomlinson 2013;Fleckenstein, Saunders and Seeleib-Kaiser 2011;Korpi and Palme 2003).
However, there are two important caveats to this general argument. The first concerns the different principles that have led to an extension of family policy in different countries and the second is about the significance of this expansion in relation to other social policies. First, Mahon (2002, 344) identified three ideal-typical models of family policy: "the neo-familialist", "the third way", and "the egalitarian horizon". The "neo-familialist" model modernises child-care provision but does not attempt to substantially modify gender differences. France and Finland, high spenders (table 2) in public childcare provision, come closest to this ideal type. The Netherlands and the United Kingdom come closest to the "third way" model, which seeks to universalise the male breadwinner model, rather than to address the unequal distribution of the care work. "The egalitarian horizon model" aspires to foster an equitable sharing of domestic care work between genders, among others by providing the most comprehensive public childcare services. Sweden and Denmark come closest to this ideal type. Second, one also needs to acknowledge that family policy expenditure, despite significant expansions, remains relatively marginal compared to the expenditure for pensions and health, constituting on average between 8% and 12% of overall public social expenditure (OECD 2010), and that the degree of change varies significantly.

Explaining the Expansion
The 1980s. Our correlation analysis for the 1980s confirms the findings of Iversen and Stephens (2008), whereby social democracy and politically-organised women (per cent of women in parliament) are closely associated with family policies encompassing (long) leaves with high replacement rates, and high expenditures for childcare services (see table 3). Furthermore, a higher level of  Understanding the Expansion of Family Policy union density seems to be an important factor explaining the variation of the overall public expenditure on family policies. The share of the centre-right vote does not seem to affect significantly the variation of "generosity" in the domain of family policy, with the exception of its negative impact on child allowance payments. Socio-economic factors, as well as political factors, play a crucial role in explaining the different levels of "generosity" across countries. In particular, a higher female labour force participation and a higher old age dependency ratio are positively correlated with the level of "generosity", i.e. long leaves with high replacement rates, generous child allowances, and higher spending for childcare services (see table 3). Confirming whereby the finding of Seeleib-Kaiser and Toivonen (2011) for Germany and Japan, a low fertility rate per se does not explain the expansion of family policy among rich OECD countries. In addition, the level of deindustrialisation does not have a significant effect.
The 1990s. In the 1990s, the impact of political and socio-economic variables on policy changes remained substantially unchanged (see table 4). Also, public opinion did not have a significant effect, which might be explained by the delayed effect of this factor within the policy-making process. Finally, the proportion of (women) workers employed in jobs that require high general skills does not seem to explain the change in the "generosity" of employmentoriented family policy. 35 The 2000s. During the 2000s we observe a declining effect of partisanship and socio-economic factors and an increased impact of societal preferences (see table 5). In contrast to the 1980s, partisanship no longer has a significant effect on the level of child allowances. The continued predominance of traditional family values 36 acts as a brake on the overall process of family policy expansion, moderating the degree of expansionary change of leave entitlements as well as spending on childcare services. This finding may be a consequence of a number of socio-economic and political developments. Policies supporting households "headed" by breadwinning men may be perceived to be out of sync with the realities of life, as the families supported by single earners (almost all men) are far less common, and household forms and life styles have become more diversified (Beck-Gernsheim 2002). A stronger commitment to gender equality and changing preferences among women plays an important role in encouraging political parties to develop and promote "modernised" family policies. For example, the German Christian Democrats have adopted a "modernised" family policy, which was partially driven by the aim to maximise their votes among the "modern" urban female electorate (Fleckenstein 2011;Seeleib-Kaiser 2010). This example highlights that policies do not operate in a vacuum, but are constructed in relation to changing cultural norms. Culture plays a role not only through policy, but also in its interaction with policy 37 (Budig, Misra, and Boeckmann 2012, 188 -189). Finally, family policy   expansion might also have a positive feed-back effect, reinforcing "modernised" public opinion (Sjoberg 2004). The percentage of women employed in jobs requiring high general skills does not correlate with an expansion of employment-oriented family policy. However, we would like to see further explorations of the connections between the levels of women in low-skilled or high-skilled occupations and the provision of public versus occupational or individually negotiated parental leave arrangements.

Understanding the Expansion of Family Policy
It is important to note two data limitations of our model: (i) the lack of data availability for skills during the 1980s, and (ii) the exclusion of public opinion data for the 1980s, due to the fact that only Austria, Germany, Ireland, Italy, The Netherlands, the United Kingdom and the United States were included in the ISSP (1988) survey. However, in all seven countries public opinion supporting the employment of mothers with pre-school children has dramatically increased over the decades (see table 6), which is in line with our overall argument.
To sum up, in the 1980s and the 1990s political and socio-economic factors were highly significant in explaining the different family policy approaches in rich OECD countries; however, developments during the 2000s confirm Carter's (1998) argument, whereby the future of welfare states seem to be more dependent on public opinion and less influenced by traditional political ideologies of political parties or other organised groups.

Conclusion
One of the most important findings of our analysis is that in an era of perceived permanent austerity (Pierson 2001) and overall welfare state retrenchment (Korpi and Palme 2003), rich OECD countries have not been prevented from expanding family policies. Au contraire! In many countries, we are witnessing the expansion of family policy, leading to a socialisation of family care responsibilities, traditionally disproportionately performed by women (Daly and Lewis 2000). Although at the institutional policy level the expansion of family policy might be characterised as a "silent revolution" relevant for gender equality, we think a cautious interpretation is necessary: gender inequalities in income, opportunities, leisure and other significant outcomes remain and are sometimes sustained by policy, even if we observe a shift in their character towards support for women's employment.
The policy shift has been particularly significant in countries that had previously emphasised more conservative approaches to family policies, such as Germany, Ireland, Japan, and Norway. Hence, it can no longer be assumed that in the majority of rich OECD countries care for young children will be mainly provided through unpaid work within the family (Mahon 2002, 35). Nevertheless, a certain number of countries still fail to provide adequate  Source: Authors' elaboration using ISSP (1988,1994,2002).

24
E. Ferragina and M. Seeleib-Kaiser childcare arrangements, constituting a barrier for full-time maternal employment (Hegewisch and Gornick 2011, 128). Furthermore, in some countries, such as the United States and the United Kingdom, gender discrimination continues to strongly intersect with class (Mandel and Shalev 2009); high childcare costs constitute a disincentive to labour force participation, especially among less educated and unskilled women (Esping-Andersen 2009). This means that higher-class and more educated women tend to have better opportunities than women belonging to a lower social class (Evertsson et al., 2009). Our analysis clearly confirms partisanship and women's political agency as the main drivers for family policy change during the 1980s and 1990s (cf. Iversen and Stephens 2008). For the 2000s, however, the importance of these drivers has significantly declined. As societal preferences have undergone profound changes-to some extent driven by the activities of women's equality movements as well as by the experience of women's employment-the policy preferences of voters have also changed. Electorates in western democracies increasingly want policies supporting "modern" family lifestyles which depend on women's employment. As political parties react to these changed policy preferences, the traditional differences in family policy positions between political parties decline. The extent to which this translates into support for "gender equality", and how such equality might be defined, is as yet not decided. However, we want to emphasise that societal policy preferences, for too long believed to be set in stone, are undergoing profound changes; public opinion increasingly matters for changing policies. The changed policy preferences are also mirrored in new political discourses that prioritise social investment and the preservation of the human capital of women, especially of those who are highly skilled (Knijn and Smit 2009;Mahon 2006;Morel, Palier, and Palme 2012; for a critical assessment of the literature see Nolan 2013). The expansion of family policies geared to supporting women's employment and investment in children is very likely to continue.

Funding
Funding to pay the Open Access publication charges for this article was provided by the University of Oxford.
Understanding the Expansion of Family Policy input in developing the MCA used in this paper and Mary Daly, Ann Orloff, and two anonymous reviewers for their helpful comments of earlier drafts.
2. For a review, see Hegewisch and Gornick (2011). 3. The notion of a "quiet" revolution in gender is contested because the rapid expansion of maternal employment-oriented family policies has not yet done away with traditional gender inequalities even in Scandinavian countries (Duvander and Ferrarini 2013;Lundqvist 2011). For this reason, England (2010England ( , 2011 argued that the "gender revolution" is "uneven and stalled" and Gerson (2010) claims that it is "unfinished". Furthermore, Morgan (2008) suggested that governments in Western Europe often promote family policy expansion to increase fertility rates and improve economic efficiency rather than gender equality.
4. Our measurement of family policy rests on a limited number of indicators (that are further discussed in the "Methods and Data" section); for this reason we acknowledge that other significant implicit and explicit measures not included in our work due to the absence of comparable data (i.e. tax breaks, housing allowances, and derived benefits in social insurance) might play an important role in shaping family policy.
5. On the contested and multivalent character of family policy, see Fink and Lundquist (2010).
6. Political actors are likely to follow different orientations in the broad field of family policy, and even within a common political orientation, policies may have conflicting effects. For example, extensive care leaves, which are certainly "generous", may undermine gender equality if they are not accompanied by greater participation of fathers in care work and greater involvement of mothers in paid employment. Conservative political forces have often championed such long leaves as part of their efforts to win over women voters (Morel, Palier, and Palme 2012;Morgan 2006Morgan , 2008Morgan , 2009Morgan , 2012Morgan , 2013Morgan and Zippel 2003). 7. It has been estimated that the cost of children account for 20 -30% of the household budget (Letablier et al. 2009).
8. Gender analyses of the welfare state have greatly contributed to a better understanding of family policy (Daly and Rake 2003;Lewis 1992;Sainsbury 1994) and welfare regime theory (Ciccia and Verloo 2012;Korpi 2000;Lewis 1997;Mahon 2001;Orloff 1997;Sainsbury 1999). They have argued that "generosity" is ungendered and have highlighted the fact that there can be very different effects of programs folded into the general rubric of family policy.
9. According to Annesley (2010), the "Adult Worker Model" assumes three configurations: (i) the "gender neutral model" in which women have strong incentives to work but there is no institutional support to combine work and care; (ii) the "gender difference approach" in which family policy fosters women's employment, reducing the burden of care for women, but not aiming to promote equal sharing of care between men and women; (iii) a model-so far only an unrealized ideal-that tries to equalize the distribution of paid work and care between men and women. (Iceland is the closest case, see Ciccia and Verloo 2012).
10. Our analysis also builds on the conceptualization introduced by Lewis (2006) whereby family policy can best be understood as a cross-cutting policy area. 11. Inglehart and Norris (2003) and Jackson (1998) have argued that "modernization" might lead to greater gender equality by promoting cultural changes and Understanding the Expansion of Family Policy women's employment and undermining the legitimacy of different treatment for men and women.
13. Mahon (2006) and Jenson (2008) have highlighted the role that international organizations, such as the OECD and the EU, have played in identifying family policies as central to the strategy of achieving higher employment rates, countering demographic trends, contributing to social cohesion and gender equality. Although these developments at the international level have certainly contributed to the diffusion of new ideas, it seems very difficult to model these exogenous factors and we have therefore limited our analysis to domestic determinants.
15. See Appendix Table A1 for the description of all variables. 16. In contrast to much of the welfare state literature (i.e. Huber and Stephens 2001), we do not rely on multiple regression (MR) or time series cross sectional (TSCS) analyses to explain the variation of family policy. Although more sophisticated than simple correlations, MR and TSCS analyses present some conceptual problems in relation to comparative analysis (Shalev 2007). Imagine one wants to test the effect of union power, measured by union density, on the generosity of family policy: by including the variable "union density" for every year, even if there is no significant change in the value, which is quite common in these datasets, one violates the underlying assumptions of regression analysis, as MR and TSCS analyses assume institutional variables to significantly vary.
18. In our case there are four variables but we use a simplified example with three variables to illustrate the technicalities of MCA.
19. For a review of welfare regimes literature, see Ferragina and Seeleib-Kaiser (2011). 20. For instance in Sweden families are entitled to 450 days of leave that can be shared between parents, and that can be taken on a full-time or part-time basis until the child is 8 years old. This scheme is called parental leave, but is similar to the concept of a childcare leave (Gauthier 2011, 2 -3).
21. See Ferrarini (2006) for an analysis that distinguishes these types of leaves and the implications for gender equality.
22. Hence, we can only identify general trends and associations, rather than provide robust causal explanations including the size of the effects. In addition, we scrutinized the variation of family policy change in five-year/ten-year intervals to capture the dynamic of change. More specifically, we have correlated the variation of political and socio-economic variables over five years (ten years) and family policy change over the same period of time (lagged by one year). The additional empirical models confirm the findings we present in 'Results' section.
23. The sources of the data are listed in Appendix Table A1. 24. Unfortunately we have only data for the 1990s and 2000s for these last two indicators.
25. This variable is only a proxy; for a qualitative study of the influence of employers on family policy in Britain and Germany, see  26. We have also run alternative models by using different variables to capture the influence of political parties, i.e. "the percentage of MPs", "the share of cabinet ministers", "the cumulative share of cabinet ministers"; the outcomes are quite similar to those presented in our analysis.
27. See Appendix Table A1. Unfortunately, we can only include this variable into our models for the 1990s and the 2000s, due to a lack of data availability for a large number of countries during the earlier period.
28. Our MCA explains 85% of the variation of the variables used to measure the "generosity" of family policy. This goes well beyond the 60% threshold required for a robust analysis (Le Roux and Rouanet 2010).
29. For a detailed description of the policy change in parental leaves across the countries analysed, see Moss (2012). 30. Although maternity leaves in the United Kingdom have been extended during the New Labour governments, there have been only marginal changes with regard to paternity and parental leaves and the replacement rates continue to be rather low (Daly 2010). Furthermore, our model does not consider the potentially significant changes resulting from the expansion of tax credits (for these, see Bradshaw 2011). 31. Mahon (2001) characterised France as undergoing shifts in a "familistic direction" due to change in various policy incentives (for a general overview about the familistic nature of French family policy, see . 32. Mahon (2002), however, argues that in gendered terms, Finland and France are moving toward a "new familialism" and away from gender-egalitarian concerns.
33. Some scholar also claimed that "cash for care" programs have had a limited impact on the level of family policy generosity (see Eydal and Rostgaard 2011). 34. "Gender perspectives have had a growing influence on policy paradigms, both within social-science debates around the role of ideas and culture in policies, and on new social policy paradigms themselves" (Orloff and Palier 2009, 406).
35. Fleckenstein and Lee (2014) show the existence of a correlation between these two variables for a selected group of countries (Britain, Germany, South Korea, and Sweden). 36. Measured by looking at the reported opinions about employment and motherhood provided by the International Social Survey Programme, we consider the percentage of people who answered "stay at home" to the question "when there is a child under school age a women should. . .".
37. This means, for example, that the effect of family policy change on mother's employment and earnings is not universal, but it is mediated by different cultural contexts. As shown by Budig, Misra and Boeckmann (2012) and Keck and Saraceno (2013), an increased "generosity" of explicit family policy is associated with higher earnings for mothers when cultural support for maternal employment is high, but have less positive or even negative relationships with earnings where cultural attitudes support the male breadwinner model.