Gender Ideology and Women’s Labor Market Transitions Within Couples in the Netherlands

Objective: This study examines the role of women’s and their partners’ gender ideology in shaping women’s labor market entries, exits, and changes in hours of employment. Background: Recent research argues that women’s gender ideology is crucial for understanding women’s contemporary labor market participation. However, the role of male partners’ gender ideology for partnered women’s labor market participation has received less attention. Method: The analysis uses three waves of a large-scale household panel survey based on a random sample of individuals within Dutch households. Random-effect models are applied to study whether women’s and their partners’ gender ideology are associated with women’s labor market transitions and whether relevant household characteristics’ associations with women’s labor market transitions are conditional on both partners’ gender ideology. Results: Women’s gender ideology is associated with the probability of women’s labor


market entries and exits, but not with changes in women's hours worked, whereas their male partners' ideology is related only to the probability of women's labor market exits. Furthermore, the negative association of having children with changes in women's hours worked is stronger for traditional compared to egalitarian women.
There is no clear evidence that gender ideology moderates the association of the male partner's labor market resources with women's labor market transitions. Conclusion: Women's labor market transitions are not only reactions to economic pressure and institutional constraints but also women's and marginally their partners' gender attitudes.
The increase in female labor force participation since the 1950s is often considered one of the great successes of female emancipation. Women in paid work continued to be an exception in the first part of the 20th century, but dual-earner families and women in full-time employment are now widely accepted in all Western countries (Blossfeld & Drobnič, 2001). However, gender inequalities persist in labor force participation rates and number of hours worked, and many women continue to be economically dependent on their partners (Merens, Hartgers, & Van Den Brakel, 2012). Recent research even suggests a "stalled" gender revolution as the development toward gender equality has slowed or even stopped (England, 2010;Hochschild, 1989;Thébaud & Pedulla, 2016).
Some scholars have argued that women's own preferences and attitudes for the household division of paid and unpaid work-shaped by their gender ideology-are central to understanding the contemporary labor market behavior of women (Charles, 2011;Hakim, 2000). However, most women do not make decisions about their involvement in child care, household tasks, and paid work within a social vacuum. Instead, these decisions are negotiated with partners whose allocation of time and resources are also affected. Because partnerships take a greater toll on women's labor market careers than men's (Bianchi & Milkie, 2010), and men are more likely than women to seek a traditionally gendered division of paid and domestic work (Davis & Greenstein, 2009), the role of male partners' gender ideology should not be neglected in studying women's labor market outcomes. We, therefore, examine the effect of gender ideology on women's labor market transitions at the couple level, also taking into account male partners' gender ideology.
We make two main contributions to the literature. First, ours is among the first studies to test whether male partners' gender ideology has a direct effect on women's labor force transitions net of women's own gender ideology and other relevant factors, including the presence of children and the partners' labor market resources. Although extreme divergence in gender attitudes within couples is rare and partners' attitudes become more similar over time (without achieving complete convergence, on average; Johnson & Huston, 1998;Kalmijn, 2005), it is clear that important differences exist. On average, men have more traditional gender attitudes than women (Kalmijn, 2005) even within couples (Poortman & Van Der Lippe, 2009), and the partner's attitudes are likely an important, although neglected, influence on women's labor market decisions. Our study advances research by using direct measures of gender ideology for both partners. This approach allows us to more precisely assess the influence of gender ideology on women's labor market transitions compared to earlier research that has relied on proxies such as education or religion.
Second, we consider how women's labor market transitions are shaped by the interaction of both partners' gender ideology with the following two important couple-level conditions: the male partner's labor market resources and the presence of young children in the household. Previous research on the effect of male partners' labor market resources on women's labor market participation has been contradictory as some studies find negative effects (Baker & Benjamin, 1997;Bernardi, 1999;Bernasco, de Graaf, & Ultee, 1998;Henz & Sundström, 2001;Long, 1980;Verbakel & de Graaf, 2009) and other positive effects (Brekke, 2013;Van Tubergen, 2008), and again others do not find much evidence for any of the two (Verbakel, 2010). Predictions of household specialization and exchange theory are often confirmed in older studies, whereas social capital theory is mostly supported in analyses with more recent data. Moreover, studies with a longitudinal design often find that an originally negative relation between the male partner's labor market resources and women's labor market participation weakens over time (Henz & Sundström, 2001;Verbakel, 2010;Verbakel & de Graaf, 2009). This change in the relation of the male partner's labor market resources and women's labor market participation could be attributed to the increasing support for women's employment during the past decades (Bolzendahl & Myers, 2004;Cunningham, 2008a;Kraaykamp, 2012)-a factor that has rarely been incorporated directly in these studies on partner effects. We therefore argue that considering both partners' gender ideology as moderators of the association between the male partner's labor market resources and women's labor market involvement may reconcile the opposing positions.
Gender ideology may also shape the well-established influence of children on women's labor force participation. Prior research has shown that the consequences of having children for women's labor market activity depend on institutional arrangements, including maternal leave and child-care supply (Begall & Grunow, 2015;Sainsbury, 1996;Van Der Lippe & Van Dijk, 2002). However, women's decisions to reduce their involvement in the labor market after giving birth might also be shaped by their gender ideology and that of their partners.
We analyze the first three waves of the Netherlands Kinship Panel Study (NKPS), which allows us to measure three outcomes related to women's labor market participation: labor market entries, labor market exits, and changes in the number of hours worked. Women's labor force participation in the Netherlands has increased slowly but steadily during the past 2 decades, from 43. 5% in 1990to 55.0% in 2000and 63.4% in 2010(Statistics Netherlands, 2016a. Employment policies make it relatively easy for employees to change their number of working hours, and part-time jobs are generally similar to full-time jobs in status, terms, and condition. This results in substantial variation in women's working hours. In 2010, about 75% of all employed women worked part-time (i.e., less than 35 hours a week) and of these about 40% worked less than 20 hours a week (Statistics Netherlands, 2016b). Heterogeneity in the degree of attachment to the labor market makes the Netherlands a good setting in which to study the effect of various individual, partner, and couple characteristics on women's labor market participation.

The Direct Effects of Gender Ideology
Gender ideology is commonly defined as "individuals' level of support for a division of paid work and family responsibilities that is based on the belief in gendered separate spheres" (Davis & Greenstein, 2009, p. 87). Individuals who endorse a traditional gender ideology believe that women should be responsible for domestic tasks and childrearing, whereas men should be responsible for providing an income for the family. Traditional gender ideology is further based on alleged innate gender dissimilarities and often accompanied by support for male primacy, that is, believing that men are better suited than women for work outside the home. In contrast, individuals supporting an egalitarian gender ideology believe in equal responsibility and capability of men and women for work within and outside the household. Current debates on gender ideology challenge the view that this unidimensional concept captures the full range of gender ideologies (Grunow, Begall, & Buchler, 2018;Knight & Brinton, 2017). Some have observed that, for example, the belief that men and women have innately different interests and motivations is a distinct dimension of gender ideology, which individuals may support or reject regardless of their belief in different spheres (Cotter, Hermsen, & Vanneman, 2011). However, such heterogeneity of gender ideologies may be less relevant when studying only women's degree of involvement in paid employment as an outcome for which belief in gender essentialism or other potential dimensions of gender ideology may be less central. Hence, although acknowledging the multidimensionality and wide-ranging scope of attitudes and beliefs that fall under the large umbrella of the term gender ideology, our conceptualization and measurement of gender ideology is unidimensional and focuses on individuals' attitudes toward women's involvement in the labor market, in line with comparable earlier studies (Corrigall & Konrad, 2007;Cunningham, 2008b).
An important aim in the study of gender ideology and female labor market activity is to identify the direction of any causal association between the two. Cognitive dissonance theory and the theory of planned behavior lead to different predictions about this causal order. According to cognitive dissonance theory, individuals adapt their attitudes in response to life course events to reduce discrepancies between their experienced life and their self-image (Festinger, 1957). Cognitive dissonance theory has been supported by studies showing that women in employment increase their own as well as their partner's support for egalitarian gender attitudes and develop stronger work commitment over time (Corrigall & Konrad, 2007;Cunningham, 2008a;Gangl & Ziefle, 2015;Kraaykamp, 2012;Thornton, Alwin, & Camburn, 1983;Alwin, Braun, & Scott, 1992;Bolzendahl & Myers, 2004;Schober & Scott, 2012).
In contrast, the theory of planned behavior proposes that intentions precede behavior (Ajzen, 1991). Within this framework, gender ideology can influence women's labor market involvement in several ways. Women's gender ideology can influence their degree of involvement in the labor market at an early stage of their career, and this initial decision may shape their labor market status for years to come (Corrigall & Konrad, 2007;Cunningham, 2008b), but people often have many conflicting interests and beliefs (e.g., maximizing their financial income vs. adhering to norms and values), which can lead to a divergence between women's preferred involvement in the labor market (or their partners' preferred division of paid and domestic work) and their actual employment conditions at a specific time-point (Janus, 2012). This divergence may be a driver for change in women's labor market behavior during their entire working age.
Compromises in terms of labor market involvement may be most prevalent among women in partnerships. Within couples, negotiations about the division of paid and domestic work can be an enduring process that results in different outcomes under varying life course conditions or different stages of the partnership (Hochschild, 1989). Renegotiations of couples' division of paid and domestic work may also be sparked by various life course events such as the birth of a child or a change in the household's income situation (which will both be discussed in more detail later), an aging family member requiring care, children leaving the household, moving to a new city, or one partner retiring while the other is still of working age. These (re-)negotiations are more likely to be initiated if attitudes and behavior diverge (among any of the partners), for example, if women have egalitarian (traditional) attitudes but do not (do) participate in the labor market. Moreover, both partners' gender attitudes may shape the outcome of these negotiations. Accordingly, women's or their partner's gender ideology may induce change in women's labor market involvement during their entire working age.
Moreover, several developments during the past 2 decades may have contributed to an increased role of gender ideology. The first refers to the increasing normative dominance of individualism in which behavior is expected to conform to one's "'true [self]'" (Charles, 2011, p.365). In addition, changes in family and labor market policies in the Netherlands (and many other countries) facilitating the reconciliation of paid and domestic work for women by increasing child-care supply and the availability of part-time employment may also have allowed women to give a greater weight to their attitudes in deciding whether to focus on paid or domestic work (Hakim, 2000).
Recent studies have already shown that women with more traditional gender attitudes are less likely to enter and more likely to exit the labor market and decrease their working hours using up to 2-year differences in the measurement of the attitudes and subsequent transitions (Stam, Verbakel, & de Graaf, 2013;Khoudja & Platt, 2018). We can expect this relation to hold even over a somewhat longer observation window under the assumption that the attitudes will not have changed substantially in the meantime due to life course events that are not captured by our longitudinal design.
Longitudinal research examining whether their male partners' gender ideology also affects women's labor market participation over time is rare. Male partners may, implicitly or explicitly, expect women to follow traditional female life trajectories, and women with high career ambitions and a partner with traditional attitudes may forego their career ambitions despite high opportunity costs to avoid conflict and secure the stability of their families (McRae, 2003). Male partners with more traditional gender attitudes also spend less time on housework (Bianchi, Milkie, Sayer, & Robinson, 2000;Blair & Lichter, 1991;Cunningham, 2008b;Poortman & Van Der Lippe, 2009), leaving women with a greater share of such work. Thus, men's traditional attitudes and corresponding behavior can leave women with less time and energy to devote to labor market participation, regardless of women's own gender ideology and career ambitions. Overall, the male partner's gender ideology may either deter women from changing their labor market status over time or encourage them to do so each time a potential change in women's labor market involvement is debated in the household.
We therefore hypothesize that women who have male partners with traditional gender attitudes are less likely to enter the labor market and more likely to exit the labor market or to reduce their hours of work per week compared to women whose male partners have more egalitarian gender attitudes, even when controlling for women's own gender attitudes (Hypothesis 1).

Moderating Effects of Gender Ideology: The Partner's Labor Market Resources
Most prior research on partner effects has focused on partners' labor market resources as central explanatory factors of female labor force participation. As gender attitudes, the male partner's labor market resources can be framed as a condition that affects the likelihood of renegotiations of the division of paid and domestic work to (re-)occur and as a direct influence on the outcome of these negotiations. At the same time, the partner's labor market resources are more prone to suddenly change (e.g., due to a partner's job loss) than gender attitudes, which may more directly affect the need or opportunity for changes in women's labor market involvement. The following three theoretical frameworks are relevant to this research: household specialization, exchange theory, and social capital theory. The household specialization framework extends human capital theory's focus on the individual maximization of financial resources to the couple level: married or cohabiting couples are assumed to pool their resources to maximize utility at the household level instead of the individual level (Becker, 1985). The theory further claims that the most efficient way to maximize household utility is to divide domestic tasks and paid work between partners due to the economic gains inherent in specialization. In contrast, exchange theory argues that specialization is the consequence of power-based negotiations between partners (Lundberg & Pollak, 1996). The partner with more profitable labor market resources has a more powerful bargaining position, allowing him or her to minimize the contribution to domestic tasks and maximize the time spent in paid labor. Despite these differing suggested mechanisms, both theories predict that increases in a partner's income or improvements in his labor market status will increase the probability that a woman will exit the labor market or reduce her work hours. Decreases in partner income or labor market exits should, in turn, increase women's labor market entries or hours worked. Social capital theory proposes an opposing mechanism to household specialization and exchange theory by arguing that cohabiting partners use their partner's resources to improve their own labor market situation (Bernardi, 1999;Bernasco et al., 1998). Partners can improve each other's skills and competencies and provide each other with useful information, guidance, and training in several steps of the employment process, including advice on job searches and job interviews as well as access to employment in some cases (Bernardi, 1999;Bernasco et al., 1998;Van Tubergen, 2008;Jacob & Kleinert, 2014). As a result, the opportunity costs for nonparticipation increase for those who have a higher skilled partner. Hence, through mutual support and spillover effects, partners help each other with their labor market endeavors, and the partner with lower resources might profit more from this exchange.
Some studies have found empirical evidence for household specialization and exchange theory, showing that with a high income (when compared with a lower income) partner, women are more likely to exit and less likely to (re)-enter paid employment (Bernasco et al., 1998), more likely to transition from full-time to part-time work (Henz & Sundström, 2001), and more likely to exit the labor market, that is, leaving paid employment and stop searching for paid work (Shafer, 2011). Furthermore, male partners' education, employment, and occupational attainment have been negatively related to women's number of hours worked per week (Verbakel, 2010;Verbakel & de Graaf, 2009) and women's labor market participation (Bernardi, 1999).
Other studies more strongly support the predictions of social capital theory. For example, some evidence indicates that women's labor market participation is positively influenced by a male partner's employment status and earnings (Brekke, 2013) and that the partner's education is positively related to women's employment (Van Tubergen, 2008) and wage (Brynin & Schupp, 2000). Verbakel (2010) showed that women with a partner who works many hours per week are at lower risk to reduce their own working hours than women with a partner who works fewer hours and that a male partner's occupational status is negatively related to women's risk of exiting employment. However, in that same study, she found overall little evidence for an effect of male partners' resources (education, employment, number of hours worked, occupational status) on women's employment entries and exits or increased working hours and ultimately rejected the hypothesis that there is a clear relation between male partners' labor market resources and women's labor market activity.
We argue that discrepancies in empirical findings and support for varied theoretical perspectives arise in part from the assumption that couples' household utility is defined objectively primarily in economic terms. This objective utility expectation predicts a household strategy that promises the highest economic utility for the household, but negative effects of the male partner's labor market resources on women's labor force participation have weakened during the past decades (Henz & Sundström, 2001;Verbakel & de Graaf, 2009), suggesting that objective economic utility maximization may be an incomplete explanation for couple behavior.
The fact that gender attitudes became increasingly egalitarian (Bolzendahl & Myers, 2004;Cunningham, 2008a) across the same time period that the link between partners' labor market resources and women's labor force participation weakened suggests that subjective utility expectations (Esser, 1999) may also be important. In other words, household utility is likely not limited to economic factors but also requires an alignment of the couple's division of paid and domestic work with their attitudes and preferences. The couple's household strategy is, therefore, the outcome of a complex equation of economic considerations and normative attitudes (Wallace, 2002). Women with traditional gender attitudes can focus on domestic work without imposing financial strain if their partners provide sufficient funds for the household. In contrast, women with more egalitarian attitudes may use their partners' resources to find better employment. The gender attitudes of the male partner might influence whether he uses his labor market resources to assist his partner in the labor market or whether he uses them to discourage her career ambitions. Variations of this idea have been tested earlier, but either used a national-rather than an individual-or couple-level unit of analysis (Abendroth, 2014) or were limited by cross-sectional data and a relatively small sample of urban areas only (Khoudja & Fleischmann, 2017).
Our hypothesis is that both partners' gender attitudes moderate the effect of the male partners' labor market resources on women's labor force transitions. Specifically, we expect that if women or their partners have more traditional gender attitudes, the labor market resources of the male partner make it more likely for women to exit the labor market and to reduce their number of hours work per week and less likely to enter the labor market. Correspondingly, if women or their partner have more egalitarian gender attitudes, the labor market resources of the male partner make it more likely for women to enter the labor market and increase their number of hours worked and less likely to exit the labor market (Hypothesis 2).

Moderating Effects of Gender Ideology: Young Children in the Household
Children in the household are strongly associated with decreased labor market participation or working hours among women (Uunk, Kalmijn, & Muffels, 2005;Van Der Lippe & Van Dijk, 2002). The time demands associated with caring for children, especially preschool aged children, reduce the time available to women to devote to paid work. The demands related to child care decrease after children enter kindergarten, and children become more cost-intensive with age (Banks & Johnson, 1993). Thus, women's availability and economic motivation for labor force involvement should increase as children age. Nonetheless, unless arrangements are made to reconcile child care with employment, having young children often implies that one of the parents will exit the labor market-and this parent tends to be the mother (Sanchez & Thomson, 1997;Verbakel, 2010). Although the time demands required for child care probably do not add up neatly with every additional young child (i.e., two young children probably will not require exactly twice as much time than one young child), we can still expect an increase in time demands with every additional young child and, correspondingly, a decrease in time demands with one (or more) young children reaching school age while there has been no new childbirth.
The way in which couples reconcile child-care demands with paid work depends to some degree on the institutional context of their country of residence (Van Der Lippe & Van Dijk, 2002). Dutch family policies in the 2000s are characterized by a relatively low scope and subsidization of parental leave and by giving employees more rights to change their working hours from full-time to part-time and vice versa (Lewis, Knijn, Martin, & Ostner, 2008). Empirical evidence suggests that during the past decade women in the Netherlands became more likely to reduce their working hours rather than to exit the labor market completely after giving birth to a first child (Begall & Grunow, 2015). All in all, the most important means to reconcile work and family in the Netherlands is the 1.5 breadwinner model with men in full-time and women in part-time employment (Portegijs & Keuzenkamp, 2008).
In this article, we argue that the degree to which women invest time in child care at the expense of their time in paid employment depends to some degree on their own gender attitudes and those of their partners. Parents in the Netherlands usually have reservations against sending their children to formal child care 5 days a week (Portegijs, Hermans, & Lalta, 2006). This might result in the mother taking care of the child at least 1 day per week and perhaps more frequently for traditional parents. More traditional male partners might also be less willing to invest in external child care than egalitarian partners, making it more difficult for women to engage in the labor market after giving birth. Furthermore, when compared with their more egalitarian counterparts, women with traditional attitudes spend more time on child care, whereas men with traditional attitudes spend less (Davis & Greenstein, 2009;Poortman & Van Der Lippe, 2009). This may also decrease the time mothers spend in paid work (Sanchez & Thomson, 1997;Schober, 2013).
Hence, we expect that an increase in the number of young children in the household compared to no change in the number of young children increases the probability that women will exit the labor market and decreases their number of hours worked. In turn, experiencing a decrease in the number of children aged younger than 4 years (which is the school starting age in the Netherlands) compared to experiencing no change in the number of young children should increase women's likelihood of entering the labor market and their number of hours worked. Our hypothesis is that all of these effects are moderated by gender ideology: The role of children aged younger than 4 in decreasing women's labor market activity and of children reaching school age in increasing women's labor market activity should be stronger among couples with more traditional gender attitudes when compared with those with more egalitarian attitudes (Hypothesis 3).

Data
We use the NKPS, a large-scale longitudinal survey that focuses on family life in the Netherlands (http://www.nkps.nl/). The NKPS is well suited for our analysis for a number of reasons. First, it has a large sample at Wave 1 of 8,161 individuals, aged between 18 and 79 years, who were chosen by use of random sampling of addresses of private households in the Netherlands. However, women younger than 30 years and single women without children are somewhat underrepresented and couples with children are somewhat overrepresented. Second, it collects information of several household members including partners and other family members in the household. Third, it contains data on women's and their partner's work, income, education, and gender attitudes. Longitudinal household surveys that contain questions about the gender attitudes of respondents and their partners are still relatively rare. Fourth, respondent and partner data are collected across three waves with 3-to 4-year intervals, allowing for an analysis of change over time.
For the first wave, data were collected between 2002 and 2004 (Dykstra et al., 2005), for the second wave in 2006 and 2007 , and for the third wave in 2010 and 2011 (Merz et al., 2012). For the main respondent (anchor), the NKPS used a computer-assisted face-to-face questionnaire and a self-completion questionnaire. The partner of the main respondent (alter) received a shorter version of the self-completion questionnaire. In the third wave, a variety of data collection methods were employed in efforts to increase the response rate (and for financial reasons): computer-assisted telephone interviews (27% of all third wave interviews), computer-assisted web interviews (55%), and computer-assisted face-to-face questionnaire (18%). The response rate of anchors in the first wave was 44.7% after a nonresponse follow-up in 2004, which is comparable to earlier family surveys in the Netherlands, and 76% for the coresident partners of these anchors (Dykstra et al., 2005). The response rate of partners varied to some degree by relationship quality, but less so compared to parents, siblings, and children. Moreover, only a few relationships qualified as poor (Dykstra et al., 2005). Based on the anchors that participated in the preceding wave, there was a 74.6% response rate in Wave 2 and 72.1% in Wave 3, whereas their partners' response rate was 60% in Wave 2 and 55.4% in Wave 3 Merz et al., 2012).
The initial pooled sample of all three waves consists of 18,642 person-wave observations across 8,161 individual respondents. To estimate the effect of characteristics of the partner at a given wave on women's labor market transitions at a subsequent wave, we only focus on long-term couples. Hence, we remove observations of single respondents (n = 4,537 person-waves across n = 2,595 individual respondents), those who are not cohabiting with their partner or spouse unless they are going to cohabit with this partner in the following wave (n = 488 person-waves across n = 411 individual respondents), those for which we have no information about the partner due to a refusal to participate (n = 2,631 person-waves across n = 2,082 individual respondents) or other reasons (n = 1,921 person-waves across n = 1,620 individual respondents), and observations of respondents who have a new partner since the preceding wave (n = 699 person-waves across n = 672 individual respondents). We also exclude the observations of respondents in the waves preceding or following one of the just mentioned exclusion criteria with the one exception already mentioned previously, which are cohabiting couples who were not cohabiting in the earlier wave (n = 1,708). Furthermore, we remove observations of respondents who were only present in one wave (n = 674) and the first observation of any valid respondents (i.e., those who are cohabiting with their partner and for which we have information about this partner across at least two consecutive waves; n = 2,242) although the latter still enter the sample indirectly as a basis for a transition that will be measured in the second observation of the analyzed respondent (see later for a more detailed explanation of our analysis). Then, we exclude observations of homosexual couples (n = 82 person-waves across n = 67 individual respondents), women who are aged younger than 18 and older than 64 years (n = 557 person-waves across n = 427 individual respondents), and respondents whose age was inconsistently indicated across waves (n = 6 person-waves across n = 6 individual respondents). Finally, women who transition into early retirement, disabled women, and full-time students are also excluded from the analysis (n = 314 person-waves across n = 266 individual respondents).
The remaining analytic sample consists of 2,783 person-waves and covers 1,899 women and their partners. The number of person-waves is not twice as high as the number of respondents as the maximum of three observations of a respondent amount to a maximum of two observed transitions (one between Wave 1 and Wave 2 and a second one between Wave 2 and Wave 3), meaning that for some respondents, we only observed one transition. Compared to couples in the Dutch population in 2002 (at the beginning of the data collection), our analytic sample underrepresents unmarried cohabiting couples without children (6% of all [cohabiting] couples in the analytic sample compared to 12% in the population) and married cohabiting couples without children (30% in the analytic sample compared to 37% in the population), whereas married couples with children are overrepresented (58% in the analytic sample compared to 46% in the population). Unmarried cohabiting couples with children are about evenly represented (6% in the analytic sample compared to 5% in the population; own calculation based on Statistics Netherlands, 2002).

Measures
Dependent variables. We construct the following two variables: labor market entry, which indicates that women were inactive at t k and active at t k + 1 , and labor market exit, which indicates that women were active at t k and inactive at t k + 1 . With this approach, we have two independent samples, one for estimating entry probabilities, with women who are inactive in two consecutive waves as the reference group and one for estimating exit probabilities, with women who are active in two consecutive waves as the reference group (cf. Jeon, 2008). Due to the relatively high level of female labor force participation in the Netherlands, we can expect the sample for labor market entries to be smaller than the sample for labor market exits. We also use changes in the number of hours worked between two waves as an outcome variable. We truncated the number of hours worked per week at 60 hours and subtracted the number of hours worked at t k from the number of hours worked at t k + 1 . For the analysis of this outcome, we focused only on respondents who were active between t k and t k + 1 so that the estimated coefficients are not mainly driven by observations of labor market entries and exits again. Independent variables. We use the following five items to measure gender ideology of women and their partners: (a) "A woman must quit her job when she becomes a mother," (b) "It's unnatural if men in a business are supervised or managed by women," (c) "It's more important for boys than for girls to be able to earn a living later in life," (d) "Working mothers put themselves first rather than their families," and (e) "It's best to divide tasks and responsibilities in a relationship according to the customs, traditions, and rules that have always been in force." Respondents could indicate their (dis)agreement with this statement on a 5-point scale ranging from 1 (strongly agree) to 5 (strongly disagree). The items were recoded so that a higher value stands for more traditional gender attitudes. Notwithstanding that items 2 and 5 do not specifically refer to respondents' attitudes about women's employment, factor analysis showed loadings of .61 to .65 on one factor for women and loadings of .58 to .68 for men supporting our unidimensional conceptualization of gender ideology. Cronbach's of the index was .79 for women and.77 for men. We construct one gender ideology factor for women and one for men by averaging across the items. The two scales are, expectedly, correlated (r = .45; p = .00), but a variance inflation factor of 1.3 for both variables provides no indication for problematic levels of multicollinearity. We use the measure for gender attitudes in the wave preceding potential labor market transitions (so either Wave 1 or Wave 2). We decided to use this lagged-variable approach to minimize the problem of expost rationalization. It may take some time until attitudes result in changes in women's labor market behavior as women might wait for the right opportunity to align their attitudes with their behavior. We assume that the 3-to 4-year time span between each wave is long enough to expect relevant (observed or unobserved) life course events to happen and short enough to capture significant changes in attitudes.
We use four measures to capture the male partner's labor market resources. We constructed the relative education of partners in the following way: First, we measured education for each of the partners by distinguishing between the following four levels: 0 = "primary education," 1 = "lower secondary vocational education," 2 = "upper secondary education," and 3 = "tertiary education." We then constructed a categorical variable with the following three values: couples in which the man is in a higher education category than his female partner, couples in which the man is in a lower education category than his female partner, and couples who are equally educated as the reference. We chose this approach as it captures the ideas of household specialization and exchange theory more closely than the male partner's absolute educational level would have. Second, we include a dummy variable that indicates whether the male partner is not working. This includes male partners who are unemployed, a student, disabled, or retired. Third, to measure the income of the male partner, we add his salary with his welfare benefits (e.g., social security, pension, student grant), if applicable. We truncate the income distribution at 10,000 Euros a month to avoid bias through income outliers. For the 278 respondents who provided only an income range, we used the midpoint. We then created a three-category income change variable that distinguishes between a male partner who had an income increase of more than 20% since the preceding wave, a male partner who had an income decrease of more than 20% since the preceding wave, and a male with no substantial income change as the reference category. We use 20% as a cut-off point to maximize statistical power while making sure that the income change is substantial enough to be felt by the household. More extreme income changes may have a greater impact on women' labor market behavior but are also less common and, therefore, not observed often enough for the more demanding analysis of the interaction effects. The 20% threshold has also been used in earlier studies (e.g., Jeon, 2008).
We also construct measures that indicate changes in the number of children younger than the age of 4 in the household. Even though education in the Netherlands becomes compulsory after the age of 5, almost all children attend primary school by age 4 (Luijkx & de Heus, 2008). Children aged younger than 4, therefore, place the greatest time demands on parents. We construct a three-value categorical variable that measures whether there has been an increase in the number of children younger than the age of 4 in the household between two waves, a decrease in the number of children younger than the age of 4 (meaning one or more children younger than age 4 in the previous wave have reached school age and there has been no new childbirth), or no changes in the number of children younger than the age of 4 in the household between two waves, which we use as the reference category. Although transitioning from having no young child to at least one young child in the household (and from one or more young children to no young children) may cause changes in time demands greater than for any additional young child, the time demands still increase with every additional young child and therefore may lead to labor market transitions. Of the families with young children in our sample, 76% do not have more than one child at a time younger than the age of 4, meaning that in practice our measure still mostly captures transitions from one young child to no young child or vice versa.
We control for the lagged total number of children younger than the age of 4 in the household, for women's highest educational degree (measured as described previously), as well as for the time variant characteristics general health of the women, measured in every wave on a 5-point scale, age and age squared of the women as well as wave.

Analytic Strategy
Given our interest in labor market transitions, we pool pairs of waves from the three waves of the survey and model changes in the labor force participation and number of hours worked between t k and t k + 1 . We estimate separate logistic regressions models for entering and exiting the labor market (compared to remaining inactive and remaining active, respectively) and linear regressions to model changes in the number of hours worked between two consecutive waves. We introduced individual random effects and robust standard errors in the logistic as well as in the linear regression models to account for the potential nonindependence of repeated observations over time within respondents (Allison, 2009). Although using random effects might lead to biased results through unobserved time-invariant variables, it has the advantage of allowing us to model between-individual and within-individual differences simultaneously. We did not use fixed effect models as some of our variables of interest do not have much within-individual variation (e.g., relative education). For modeling changes in the number of hours worked, a Hausman test suggested that a random effects model does not estimate significantly different coefficients compared to a fixed effects model, supporting our choice, 2 (18) = 22.49 (p = .21).
With about 10% missing values on the income variables, 1% missing on each partners' gender role attitudes, and two missing cases on the dependent variables (n = 334 in total), listwise deletion might affect the results (Acock, 2005). We therefore use multiple imputation with chained equations to multiply impute the missing values on the income and gender attitude variables (and the interaction they are part of) based on 10 imputed data sets, following the rule of thumb that the number of imputed data sets should correspond to the share of missing observations (White, Royston, & Wood, 2011). Our equation models consist of the variables used in the analytical model, plus the auxiliary variables household size, household type, and region. We include estimates of the models with missing values listwise deleted in the online supplement (Table A1).
We first estimate the direct effects of our explanatory variables, including controls, on women's labor market entries and exits as well as the changes in the number of hours worked. Then we estimate two-way interaction effects that examine whether male partners' labor market resources and changes in the number of young children interact with each of the partners' gender attitudes.

Robustness of Results
In the case of the labor market entry and exit models, we introduce a potential selection bias into our estimates as the first observed measurement of respondents might reflect unobserved characteristics that lead to a greater propensity to remain in the first given state (Cappellari & Jenkins, 2008;Stewart & Swaffield, 1999). We incorporate generalized residuals in our models to account for these initial conditions (Jeon, 2008;Orme, 2001). We find that the results from this robustness test do not diverge substantially from the results presented (see Table A2 in the online supplement). Table 1 indicates that 37% of all women who are inactive in the labor market in a given year enter the labor market within the next 3 to 4 years. Of all the women who were active in the labor market in a given year, about 7% exited it. This does not mean that much more labor market entries than exits happen. As more than half of the Dutch women are active in the labor market, the sample that uses active women as the reference group (n = 2,033) is much larger than the sample that uses women who are inactive as the reference group (n = 748). Furthermore, the women in our sample were more often less educated when compared with their partner than the other way around. Unsurprisingly, men had on average more traditional gender attitudes (M = 1.98; SD = 0.68) than women (M = 1.74; SD = 0.63); t(2717) = −17.86, 2-sided p < .001.

Explanatory Results
We now present the results of our explanatory analysis of labor market entries, exits, and changes in the number of hours worked. As we apply the same explanatory framework for the outcomes, we first discuss results of the direct estimated effects for all three outcomes (depicted in Table 2) and subsequently continue with reporting the results of the interaction analyses (depicted in Table 3). To improve readability, we only discuss the significant interaction effects in the text (Figures 1 and 2). Providing only partial support for Hypothesis 1, partners' traditional gender attitudes were positively related to women's labor market exits, but not significantly related to women's labor market entries or to changes in their number of hours worked (see Table 2). There was more evidence for an influence of women's gender attitudes on changes in their own labor market activity. Women with traditional attitudes were less likely to enter the labor market and more likely to exit the labor market than women with less traditional attitudes. Contrary to our expectations, there was no evidence for an effect of women's gender attitudes with changes in their number of hours worked per week.
Consistent with the predictions of household specialization theory, we found a positive association of the partner's absolute income and women's labor market exits, but overall, support for household specialization theory was rather weak as there was no other indication for an association between male partners' labor market resources and women's labor market entries, exits, or changes in their numbers of hours worked. To test whether larger income changes of the partner may lead to transitions in women' labor market status, we estimated effects of increases and decreases in the partners' income of at least 40% (results available on request). Indeed, these estimations showed an odds ratio of 2.98 for the coefficient of partner's income increases on women's labor market exit with a 95% confidence interval (CI) of [1.31, 6.75]. This suggests that women whose partner raised his income with at least 40% were more likely to exit the labor market than women whose Note. The random effects logit estimator of labor market entries does not differ significantly from the pooled logit estimator as a likelihood ratio test could not reject the null hypothesis that rho = 0. A likelihood-ratio test comparing the random-effect linear model with a one-level ordinary linear regression is also not significant. The coefficients in brackets should be read with caution due to relatively small cell sizes (see Table A3 in online supplement). ref. = reference. HH = household. * p < .05; ** p < .01; *** p < .001.
partner had a relatively stable income, which fits the prediction of household specialization theory. As the number of observations with such an extreme income change is rather low, however, we decided to keep the 20% cut-off for the interaction analysis to increase statistical power. Evidence for social capital theory was even weaker as we found no evidence that male partners' labor market resources increased women' involvement in paid employment.
Another finding was that the likelihood of a woman exiting the labor market increased after an increase in the number of young children. Although the bracketed coefficient suggested that a decrease in the number of young children in the household was negatively related to the probability of a labor market exit, this association was not a focus of our hypotheses, and the cell sizes in this category were too small to justify confidence in these estimates. The Table 3.

Random-Effects Models of Women's Labor Market Entries, Labor Market Exits, and Changes in the Number of Hours Worked With Interaction Effects
Labor market entry (logit) Note. The coefficients in brackets should be read with caution due to relatively small cell sizes (see Table A3 in online supplement). ref. = reference, HH = household, FGRA = Female gender role attitudes, MGRA = Male gender role attitudes. * p < .05; ** p < .01; *** p < .001. Note. Shaded areas indicate the 95% confidence interval. We show the conditional estimated effect of a lower educated male partner only in the range of −1 to 2 on the scale of men's attitudes due to the relatively small number of observation (n = 23) with scores higher than 2 on the scale (corresponding to about 4 on the original, not centered, scale). AME = Average marginal effect. same applied to the bracketed coefficient that suggested an increase in the number of young children in the household was negatively related to women's likelihood of entering the labor market. Employed women who experienced an increase in the number of children aged younger than 4 reduced their paid work per week by 8.63 hours when compared with women who experience no change in the number of children aged younger than 4. Experiencing a decrease of one or more children younger than the age of 4 increased women's hours in paid work by about 2.24 hours when compared with women in households with no change in the number of children younger than age 4. Furthermore, each additional young child in the household was associated with a decrease in women's working hours 3 to 4 years later.
In Table 3, we examined whether both partners' gender attitudes moderated the effects of the male partner's labor market resources and young children in the household on changes in women's labor market activity. We only found one significant interaction of women's or their partners' gender attitudes with male partners' labor market resources. Women with a lower educated male partner compared to women with Note. Shaded areas indicate the 95% confidence interval. We show the conditional estimated effect of an increase and decrease in the number of young children only in the range of −1 to 2 on the scale of women's gender attitudes due to the relatively small number of observations (n = 18) with scores higher than 2 on the scale (corresponding to about 4 on the original, not centered, scale). AME = Average marginal effect.
an equally educated male partner were more likely to decrease their number of hours worked if the male partner had more egalitarian attitudes. A closer examination of the interaction effects (as depicted in Figure 1) showed that among women whose partner had highly egalitarian attitudes, there was a 2.56 hour (95% CI [−4.40, −0.72]) difference in the changes of working hours between women with a lower educated partner (−1.70 hours total decrease in working hours, 95% CI [−3.27, −0.14]) and women with an equally educated partner (0.86 hours total increase in working hours, 95% CI [−0.01, 1.73]). With decreasing endorsement of egalitarian attitudes by male partners, the estimated effect of having a lower educated partner compared to an equally educated partner approached zero and lost significance. This finding could be interpreted as partly consistent with Hypothesis 2 because it suggested that the predictions of social capital theory that a partner with lower resources decreases women's number of hours worked applied more to couples with egalitarian attitudes. However, the lack of other evidence that gender ideology moderated the effect of the male partners' resources on women's labor market transitions led us to reject Hypothesis 2.
The results provided more evidence for a moderating role of gender ideology on the estimated effect of changes in the number of young children in the household. Consistent with our expectations, the results showed that with increasingly traditional attitudes of women, an increase in young children in the household led to a stronger decrease in women's number of hours worked. The 95% CI of the estimated effect of having one or more additional young children on changes in women's number of hours worked per week was between 2.9 and 8.6 hours of less work for women with egalitarian gender attitudes and between 12 and 32 hours ofless work for traditional women (see squares in Figure 2). Moreover, the estimated effect of a decrease in the number of young children in the household was less positive the more women supported a traditional gender ideology. Whereas women with highly egalitarian attitudes increased their number of hours worked by about 5 hours (95% CI [1.88, 8.10]), after a decrease in the number of young children in the household, the women with less egalitarian attitudes did not significantly change their working hours (see circles in Figure 2). Both of these findings were consistent with Hypothesis 3. However, as there was no evidence for a moderating role of the male partners' gender attitudes, and only little indication for an interaction effect between having young children and gender attitudes when it comes to labor market entries and exits, we could only partly confirm Hypothesis 3.
We made a number of additional analyses to check how far our results were sensitive to smaller changes in the model specifications (available on request). First, due to the large number of interactions in Table 3, we checked whether the order or number of interactions that were added simultaneously in the model mattered. Second, to test whether we might have to account for both partners' gender attitudes simultaneously to detect any moderating effects, we estimated three-way interactions of both partners' gender attitudes and the male partners' labor market resources and changes in the number of young children in the household. Third, to check how far our results were sensitive for alternative specifications of the income effects, we estimated models that included the natural logarithm of the partner's lagged income and a control for male partners without income instead of dummies for an income increase and decrease of the partner and the partner's absolute income. Finally, we reestimated the models limiting the age range of the sample to 18 to 55 years old for women. None of these robustness checks came to substantially different results as the main analysis presented previously.

Discussion
In this article, we examined whether male partners' gender ideology affects women's labor market transitions over and above women's gender ideology and, additionally, whether the gender ideology of both partners moderates the effects of the partner's labor market resources and of changes in the number of young children in the household.
Adding to findings of earlier cross-sectional research on the relation between male partners' gender ideology and women's labor force participation (Chuang & Lee, 2003;Khoudja & Fleischmann, 2017), we found that a male partner with traditional gender attitudes increases women's risk to exit the labor market, but has no effect on women's labor market entries or changes in the number of hours worked. This finding is particularly remarkable considering that until the early 1980s legislation in several Western countries gave husbands legally a stronger voice than their partner. For instance, in Germany, the husband still had to consent to his wife's labor market activity until a change in the family law in 1976 (Rheinstein & Glendon, 1978). In the Netherlands, the husband's opinion on important family decisions, such as children's education, legally outweighed the mother's stance until a family law change in 1984 (Holtrust & de Hondt, 1997, p. 286). One generation later, women in the Netherlands escaped the legal power exerted by their husbands, and, according to our research, to a large degree also their male partners' direct normative influence on their labor market involvement.
However, our analysis could still detect that women are more likely to exit the labor market if their partner holds relatively traditional attitudes net of women's own gender attitudes. This suggests that the deactivating force of the partner's gender attitudes might have a higher effectiveness than the activating force of the partner's gender attitudes. One interpretation of this is that the activating force of men's gender attitudes is not required to encourage women to participate in the labor market as this is what structural constraints and societal norms encourage. Women's "normal" life course in the Netherlands is to start participating in the labor market after completing education, which is why most women are probably in (part-time) employment when they start cohabiting with a partner unless the women are highly traditional at a young age. Hence, women who do not have highly traditional gender attitudes are therefore more likely to end up in employment than outside the labor force, but in the long term, particularly the group of women with less than highly egalitarian gender attitudes may also be more easily encouraged to exit the labor market if they end up in a partnership with a man who holds traditional gender attitudes.
Earlier longitudinal studies found an effect of women's gender ideology on their labor force participation (Corrigall & Konrad, 2007;Cunningham, 2008b). We could show that this effect holds true for predicting labor market entries as well as labor market exits while taking into account the male partners' attitudes. However, we do not find evidence for a direct association between women's gender ideology and changes in their number of hours worked. A similar pattern was found by Verbakel (2010) with religion as proxy for gender ideology. Overall, it seems like women in the Netherlands change the extent of their employment based on other factors than their partner's and their own gender attitudes. Structural constraints set by motherhood seem more relevant for changes in women's working hours.
Finding no relation between (women's or their partner's) gender ideology at a given wave and subsequent changes in women's hours worked does not mean that gender attitudes do not influence the total number of hours women spend in the labor market. In contrast to remaining economically active (as a reference category to labor market exits) or inactive (as a reference category to labor market entries), no changes in the number of hours worked does not necessarily reflect a specific level of support for traditional gender ideology. Women can either not change their hours worked having a full-time job or having a small part-time job with the former rather being an indicator of less and the latter of more traditional gender attitudes. Hence, the lack of a significant association between traditional gender ideology and changes in hours worked only suggests that once women in partnerships settled on a certain extent of involvement in the labor market, their own or their partners' gender ideology are not a strong driver for change in the quantity of their labor market involvement. The extent of employment at the initial step into the labor market (or shortly after having entered a partnership), however, may still be shaped by women's or their partners' gender ideology.
We find relatively little evidence for a moderating role of gender ideology with regard to the effects of the male partners' labor market resources. Hence, it does not seem like the male partners' labor market resources are used instrumentally by either women or men to realize their attitudes about the role of women in paid and domestic work. In fact, our only substantial finding is that regardless of gender ideology, the chances of women to exit the labor market somewhat increase with a higher income of the partner, which is in line with household specialization and exchange theory. The mixed findings of earlier studies regarding the effect of the partner's labor market resources on women's labor market participation remain therefore puzzling. We encourage future research to focus on the role of other beliefs in influencing women's labor market participation. Items about gender attitudes are often framed so that they measure how people think gender roles should be defined. However, it may be that men's perception of what kind of gender roles are actually endorsed by other men is more predictive of their behavior (Thébaud & Pedulla, 2016). Others have suggested that personal attitudes toward a partner focusing on domestic work are more relevant for predicting the share of housework than abstract gender beliefs (Poortman & Van Der Lippe, 2009). This might also apply to paid work but has rarely been tested.
An alternative explanation might lie in the larger narrative that the couple increasingly loses its function as economic unit as partners tend to focus on their labor market career more independently from each other (Blossfeld & Drobnič, 2001). This development would imply a general decline of male partners' influence on women's paid work.
As plenty of earlier research has shown, we found strong evidence for the influence of children on women's labor market transitions. Consistent with our hypotheses, we found that giving birth increases women's probability to exit the labor market and reduces their number of hours worked per week. Similarly, one or more children reaching school age increases the number of hours in paid employment (if there has been no new childbirth in the meantime). However, there was no strong occurrence of (re-)entry when children reached school age. It seems that relatively recent Dutch family policies designed to increase women's labor market attachment are more successful with keeping women who give birth active in the labor market (despite a substantial reduction of their working hours) than (re-)activating those women who are not participating (cf. Begall & Grunow, 2015).
Furthermore, we find that a reduction in paid working hours after giving birth and an increase in hours after the child reaches school age is dependent on the gender ideology of women but not their partner's, which is similar to findings in the United Kingdom (Schober, 2013). This suggests that the effect of children on women's hours in paid employment does not only reflect an adaptation to institutional structures set by family and employment policies but also is shaped by women's gender ideology. In contrast, male partners' gender ideology matters less for how women divide their time between child care and employment.
A limitation of our study might be its focus on women who were in the same partnership during at least two waves. Given that chances to split up are higher among couples with opposing values and beliefs, our sample might be biased toward women who are above average willing to suit their partner's expectations. However, this conclusion is not as clear-cut as it appears. For instance, men in long-term relationships may also be a selected group that is more adaptive in their gender attitudes than other men. Furthermore, couples split up for numerous reasons while having very similar gender attitudes and, finally, we also disregard those who transitioned from single to cohabiting; a group who might have particularly convergent attitudes. Hence, although it seems plausible that our sample includes relatively more couples with similar than widely divergent views, this does not necessarily imply that we overestimate the effects of attitudes on behavior. Nonetheless, it may be a fruitful avenue for future research to take into account divorce and remarriage in the analysis of gender ideology and women's labor market behavior if more than three waves are available.
In this study, we examined various ways in which gender ideology may affect the labor force participation of women in partnerships. We found evidence showing that changes in women's involvement in the labor market are not only reactions to economic pressure and institutional constraints but also to women's and marginally their partners' gender attitudes.