Grandparenthood, grandchild care and employment

A growing number of studies find a negative relationship between grandparenthood and employment, both at the extensive and at the intensive margin. In this literature, the extensive margin is operationalised as the probability of being employed or self-employed as opposed to inactive, and the intensive margin is indicated by the number of working hours. Performing survival analysis on data from the Health and Retirement Study in the United States of America (USA), Lumsdaine and Vermeer (Reference Lumsdaine and Vermeer2015) find that, among older women, the birth of the first or of an additional grandchild leads to an increase in the probability of retirement by 8.5 and 1.4 per cent, respectively. Rupert and Zanella (Reference Rupert and Zanella2018) apply an instrumental variable approach to data from the US Panel Study of Income Dynamics, exploiting variation in the timing of becoming a grandparent by gender of the first-born adult child, as daughters tend to have children earlier than sons. They find that becoming a grandmother reduces women's working time by 30 per cent, but that grandparenthood has no effect on men's employment. Asquith (Reference Asquith2018) exploits variation in US state-by-year access to reproductive technologies to instrument grandparent status, and finds that grandfathers are 5.7 per cent more likely to be retired for each additional grandchild, while grandmothers work 120 hours less per year and are 8.4 per cent less likely to work in response to each additional grandchild.

Among the European studies, Backhaus and Barslund (Reference Backhaus and Barslund2019) apply the same identification strategy as Rupert and Zanella (Reference Rupert and Zanella2018) to pooled SHARE data from nine European countries. They find that being a grandmother reduces women's probability of working by 20 per cent, with no effect on men. Analyses of women in Austria (Frimmel et al., Reference Frimmel, Halla, Schmidpeter and Winter-Ebmer2017) and England (Zanasi et al., Reference Zanasi, Sieben and Uunk2020) find that the birth of the first grandchild is linked with an increase in the probability of leaving the labour market. Applying survival analysis to Swedish register data, Kridahl (Reference Kridahl2017) shows that grandparents of both sexes retire earlier than non-grandparents. Studies analysing cross-national comparative data also suggest that, across Europe, being a grandparent is associated with stronger preferences towards early retirement (Hochman and Lewin-Epstein, Reference Hochman and Lewin-Epstein2013) and with women's early retirement behaviour (Van Bavel and De Winter, Reference Van Bavel and De Winter2013). Collectively, these findings suggest that in Europe becoming a grandparent induces individuals to leave the labour force and/or to work fewer hours. Moreover, the negative effects of grandparenthood appear to be greater for women, with lower or no effects for men (Backhaus and Barslund, Reference Backhaus and Barslund2019). However, it remains unclear whether, among grandparents, the provision of grandchild care is linked with employment. This is important to understand in order to identify whether public child-care provision can support delayed retirement by relieving grandparents of care responsibilities.

A few studies investigate the association between grandchild care provision and employment. Using SHARE, Hank and Buber (Reference Hank and Buber2009) analyse the correlates of grandchild care and find that, in Europe, employed grandparents are less likely to provide it than non-employed grandparents. Using the same data-set to examine the predictors of retirement, De Preter et al. (Reference De Preter, Van Looy and Mortelmans2013) find that older workers who look after their grandchildren at least weekly are over twice as likely to retire at follow-up as those who do not. This evidence, however, is difficult to interpret, since the negative association may be explained by two non-mutually exclusive mechanisms: selection and role conflict.

Grandparents may select into paid work and family care based on both observable and unobservable characteristics. Among the observable characteristics, grandparents’ health, socio-economic status and family structure simultaneously determine their ability and necessity to work and provide child care (De Preter et al., Reference De Preter, Van Looy and Mortelmans2013; Arpino and Bordone, Reference Arpino and Bordone2017). Unobservable characteristics relate to grandparents’ personality traits and preferences towards work and family care. These include family needs and attitudes towards gender roles and family obligations. In particular, the fact that some individuals are more family-oriented while others are more career-oriented may result in grandparents who provide grandchild care having lower employment rates or working hours, even in the absence of any direct association between grandchild care and employment (Lakomý and Kreidl, Reference Lakomý and Kreidl2015). Lakomý and Kreidl (Reference Lakomý and Kreidl2015) analyse SHARE data to study how different employment statuses relate to the intensity of grandchild care. They find a positive association between being in part-time – as opposed to full-time – work and higher frequencies of grandchild care for paternal grandmothers. However, the association does not hold when controlling for time-invariant heterogeneity across grandparents using individual fixed effects, suggesting that it is mostly attributable to selection. The authors conclude that the choice between full-time and part-time work may be influenced by the same unobserved factors that also impact the frequency of grandchild care.

Accounting for selection, any direct negative association between grandchild care and employment may be explained by role conflict theory (Goode, Reference Goode1960). Both working and looking after grandchildren require time and energy (Lakomý and Kreidl, Reference Lakomý and Kreidl2015). Grandparents may be unable to combine the two roles and thus not engage in (or give up) either one or the other. Role conflict is most likely to arise when grandparents are faced with daily or almost-daily child-care responsibilities. As it has been shown in relation to informal care-givers of older adults (Heitmueller, Reference Heitmueller2007), net of selection, care provision is only negatively associated with labour market participation at high frequencies (i.e. over 20 hours a week). Intuitively, looking after a grandchild once a week or less often is unlikely to pose work–family reconciliation problems, as grandparents may easily combine full-time employment with grandchild care performed on weekends or days off from work. In this article, I consider grandchild care performed daily or almost daily, and test for whether it is in conflict with employment among grandparents aged 50–69. I use a simultaneous equations approach to account for the fact that grandparents may select into employment and daily grandchild care based on observed and unobserved characteristics.

Familism in child-care policies and daily grandchild care

The comparative literature on welfare regimes and intergenerational support defines familism (or familialism) as the extent to which families, rather than the state or the market, are considered responsible for the care and financial support of their dependent members, particularly young children and non-self-sufficient older adults (Leitner, Reference Leitner2003; Saraceno, Reference Saraceno2016). With respect to informal care-giving, familistic policies generally strengthen family care responsibilities (Leitner, Reference Leitner2003). They may do so through the absence of state- or market-provided alternatives to family care, in what has been named ‘familism by default’ (Saraceno and Keck, Reference Saraceno and Keck2010); or by actively supporting the family caring function through financial assistance (e.g. cash-for-care transfers) and employment protection to family care-givers (e.g. leave schemes), resulting in a model of ‘supported familism’ (Saraceno and Keck, Reference Saraceno and Keck2010; Saraceno, Reference Saraceno2016). By contrast, de-familising policies are those aimed at unburdening families from their support functions (Leitner, Reference Leitner2003). ‘De-familisation’ predominantly works through the provision of services, either directly by the state, or subsidised and provided via the market (Saraceno, Reference Saraceno2016). Since the provision of services and leave schemes are not mutually exclusive, some policy patterns may also result in ‘optional de-familisation’, by which family care-givers can choose between outsourcing the care or providing it themselves (Saraceno and Keck, Reference Saraceno and Keck2010; Herlofson and Hagestad, Reference Herlofson, Hagestad, Arber and Timonen2012).

Policies that partition responsibilities and obligations between the family, the state and the market inevitably reflect and regulate the division of responsibilities between genders (Lewis, Reference Lewis1992). Since care is predominantly women's work, familism is strongly related to the gender division of roles, and to the extent to which working-age women participate in the labour market (Saraceno and Keck, Reference Saraceno and Keck2011; Thévenon, Reference Thévenon2011). In general, de-familisation through service provision promotes women's employment by relieving them of care responsibilities (Saraceno and Keck, Reference Saraceno and Keck2011). Empirical evidence shows that, in countries where child-care services are widespread and parental leave schemes are generous (usually found in Northern Europe), the prevalence of grandchild care is high, but its intensity is relatively low, with most grandparents providing care once a week or less often (Igel and Szydlik, Reference Igel and Szydlik2011; Bordone et al., Reference Bordone, Arpino and Aassve2017). This reflects the fact that, since many mothers work, grandparents are often needed to complement formal child care (Herlofson and Hagestad, Reference Herlofson, Hagestad, Arber and Timonen2012).

At the other end of the policy spectrum, familism by default is not conducive to mothers’ employment, because it does not offer alternatives to family care (Saraceno and Keck, Reference Saraceno and Keck2011). In countries characterised by familistic child-care policies (usually in Southern Europe), relatively low female employment means that the demand for grandparental care is lower than in the North, and fewer grandparents provide any care (Igel and Szydlik, Reference Igel and Szydlik2011; Di Gessa et al., Reference Di Gessa, Glaser, Price, Ribe Montserrat and Tinker2016). However, the scarce provision of services (often combined with low availability of part-time employment) mean that working mothers may require full-time care from the grandparents (Arpino et al., Reference Arpino, Pronzato and Tavares2014). Thus, a large proportion of grandparents who provide care tend to do so daily (Bordone et al., Reference Bordone, Arpino and Aassve2017). The implications of long leave schemes for women's participation in the labour market are ambiguous, depending on the extent to which employment protection to family care-givers is combined with the possibility of outsourcing care (Lewis, Reference Lewis2006; Thévenon, Reference Thévenon2011), as well as on the availability of part-time employment for mothers (Bordone et al., Reference Bordone, Arpino and Aassve2017). Thus, mothers’ need for daily grandchild care may depend on whether services are provided in addition to generous parental leave.

Attempting to position European countries on to the familism/de-familisation spectrum is an arduous task, because every country presents a combination of different policy approaches with respect to different dimensions of intergenerational support (Saraceno and Keck, Reference Saraceno and Keck2010). However, countries can be classified according to specific policy areas of interest. In the following analysis, I adopt a classification of European countries that is based solely on policies related to the care of young children, which are relevant to the study of grandchild care (Bordone et al., Reference Bordone, Arpino and Aassve2017), and its association with grandparents’ employment. Saraceno and Keck (Reference Saraceno and Keck2010) propose two indicators for classifying countries according to their child-care policy mix: the percentage of children under the age of three enrolled in formal child care, an indicator of de-familisation; and effective parental leave, defined as the duration of paid parental leave multiplied by the replacement rate of the leave benefit, an indicator of supported familism. Depending on the level of each indicator, countries can be grouped into four policy regimes, which differ in the share of child-care responsibilities borne by families, and in the extent to which working-age mothers participate in the labour market:

1. (1) Optional de-familisation: both child-care service utilisation and weeks of effective parental leave are high (i.e. above the across-country average). Families can easily outsource child care to public services, but they may also choose to provide care themselves, as mothers’ employment is protected through generous leave schemes.

2. (2) Service de-familisation: child-care service utilisation is high, but the length of effective parental leave is low (i.e. below the across-country average). De-familisation occurs predominantly through service provision or subsidisation, and it promotes mothers’ employment by relieving them of child-care responsibilities.

3. (3) Supported familism: formal child-care utilisation is low, while effective parental leave is high. The welfare state encourages family care by supporting parents in taking care of children. Long periods of leave may encourage mothers to remain in the labour market by granting job protection but, in combination with the low provision of services, they may also make it harder for them to return to work.

4. (4) Familism by default: both child-care service utilisation and effective leave are low. Families are implicitly expected to take care of children, but not supported in this role, making it difficult for mothers to reconcile work and child-care responsibilities.

While following the policy regimes proposed by Sarceno and Keck (Reference Saraceno and Keck2010) and used by Bordone et al. (Reference Bordone, Arpino and Aassve2017) in their study of grandchild care provision, this classification differs from those used by the authors in two ways. First, Saraceno and Keck identify three types of regimes (familism by default, supported familism and de-familisation), but add that ‘there may also be a fourth variant that offers an option between supported familialism and de-familialisation, but this is a rare case’ (2010: 676). I explicitly consider this fourth variant as optional de-familisation. Given my focus on potential role conflict between grandchild care and employment, I find it particularly important to distinguish between ‘optional’ and ‘service’ de-familisation. As I argue below, the two regimes may lead to different levels of need for daily grandparental care, resulting in it having different associations with employment. Leitner (Reference Leitner2003) and Herlofson and Hagestad (Reference Herlofson, Hagestad, Arber and Timonen2012) use the term ‘optional familialism’ to refer to a child-care policy context where family care is supported through extensive leave, but can easily be outsourced to public services. I rename this cluster ‘optional de-familisation’ in order to draw attention to the availability of child-care services.

Second, unlike Bordone et al. (Reference Bordone, Arpino and Aassve2017), I use child-care enrolment (as opposed to ‘child-care coverage’) to indicate de-familisation. Following previous studies (Di Gessa et al., Reference Di Gessa, Glaser and Tinker2016), I select enrolment because it combines a supply-side element (the availability of services) with actual take-up rates for formal child care, which are more strongly related to unobserved country-level factors linked to grandchild care, such as norms about family responsibilities and cultural preferences about the outsourcing of child care (Arpino et al., Reference Arpino, Pronzato and Tavares2014; Di Gessa et al., Reference Di Gessa, Glaser, Price, Ribe Montserrat and Tinker2016). As such, I consider it a more ‘complete’ indicator of the level of de-familisation. For comparability with Bordone et al. (Reference Bordone, Arpino and Aassve2017), in Appendix S1 (in the online supplementary material) I present an alternative classification of countries based on child-care coverage, defined as the number of weeks of coverage available if all children were enrolled (Saraceno and Keck, Reference Saraceno and Keck2008).

While empirical research consistently finds a North–South gradient in the prevalence and frequency of grandchild care provision in Europe (Igel and Szydlik, Reference Igel and Szydlik2011; Bordone et al., Reference Bordone, Arpino and Aassve2017), nothing is known about whether some child-care policy regimes are more conducive than others to work–family reconciliation or role conflict among grandparents. In this study, I test for heterogeneity in the relationship between daily grandchild care and grandparents’ employment across the four child-care regimes described above.

In the optional de-familisation country cluster, grandparents are not needed nor normatively expected to act as full-time care-givers. They may select into employment or daily child care, or combine the two roles out of personal preference. Thus, I do not expect to find evidence of role conflict in these countries. In the service de-familisation group, the absence of generous leave schemes (combined with high service utilisation) may impose some requests on grandparents’ time, as grandparents may be needed to complement formal child care, for instance by picking up children from nurseries. While these commitments are unlikely to result in a lower probability of working (extensive margin), employed grandparents may work fewer hours if performing such daily tasks, resulting in a negative association at the intensive margin. In the supported familism cluster, grandparents may act as full-time care-givers for mothers who remain in employment, and thus be less likely to work, resulting in role conflict at the extensive margin of employment. In countries characterised by familism by default, traditional gender roles are persistent (Leitner, Reference Leitner2003), with older women much less likely to work than men. Grandmothers may self-select into employment or care for grandchildren full-time. Thus, for grandmothers, I do not expect to find role conflict at the extensive margin. However, I expect employed grandmothers who provide daily care to work substantially fewer hours than those who do not. By contrast, grandfathers are unlikely to provide grandchild care or, if they do so, to combine it with employment. Thus, I expect to observe a large negative association between daily child care and the probability of working for grandfathers, which may partly or fully be explained by selection.

Data and sample selection

SHARE is a multi-disciplinary longitudinal survey representative of the population aged 50 and over in various European countries and Israel, excluding individuals living in institutions (Börsch-Supan et al., Reference Börsch-Supan, Brandt, Hunkler, Kneip, Korbmacher, Malter, Schaan, Stuck and Zuber2013). I analyse pooled data from the first (2004–2005) to the sixth (2015) wave of SHARE. For each respondent, only information collected during the first wave at which they were interviewed is used. Thus, for respondents interviewed at all waves, only the Wave 1 observation is considered; for respondents present from Wave i onwards, only the ith wave observation is considered. The third (2008–2009) and seventh (2017) waves are excluded, as the former only contains retrospective information, while the latter only contains regular modules for respondents who were previously interviewed (and retrospective modules for new respondents). Pooling first-observation data allows as much information as possible to be retained while treating the sample as cross-sectional. The main focus of the study lies in differences in employment outcomes between grandparents who provide daily grandchild care and grandparents who do not, rather than in changes in employment over time within grandparents. I analyse data from the 20 European countries present in SHARE Waves 1–6. These are Austria, Belgium, Croatia, Czech Republic, Denmark, Estonia, France, Germany, Greece, Hungary, Ireland, Italy, Luxembourg, The Netherlands, Poland, Portugal, Slovenia, Spain, Sweden and Switzerland.

I restrict the analytic samples to grandmothers and grandfathers aged 50–69 who have at least one grandchild aged 0–14. The age range for grandparents covers the transition to retirement (Riedel and Hofer, Reference Riedel and Hofer2013), and it is commonly used in studies of later-life employment (Pollak and Sirven, Reference Pollak and Sirven2016). The age range for grandchildren isolates grandchild care performed when grandchildren are young, and it is similar to those used in previous research (Hank and Buber, Reference Hank and Buber2009; Bordone et al., Reference Bordone, Arpino and Aassve2017). However, the two child-care policy indicators (child-care services and effective parental leave) predominantly relate to care provision to much younger children (i.e. under three). While it is not possible to restrict the samples to grandparents of very young children due to the small number of observations, I assess the sensitivity of the results to lowering the bandwidth to ages 0–10 (Bordone et al., Reference Bordone, Arpino and Aassve2017) and 0–5, which gives the smallest possible sample allowing meaningful estimation.

When studying the determinants and consequences of grandchild care provision it is essential to adopt a multigenerational perspective (Igel and Szydlik, Reference Igel and Szydlik2011). Therefore, grandparent–parent dyads are used as the primary units of analysis. In the data-set, each grandparent has a number of dyads corresponding to the number of adult children (parents) who have children of their own (grandchildren) aged 0–14. In SHARE, the questionnaire section on grandchild care provision is only asked to the ‘family respondent’, the first person in a couple to start the main interview after completing the cover screen. The analyses are therefore restricted to these respondents. Since grandparents with co-resident grandchildren share household consumption with them, co-residential grandchild care may motivate paid work as a source of income, and thus have a different association with grandparents’ employment than non-co-residential care (Ho, Reference Ho2015). Since this mechanism is beyond the scope of this study, observations for which the grandparent lives in the same household as any grandchild (3.9% of the sample) are excluded from the analytic sample.

The analyses of the probability of employment (extensive margin) are based on the samples of grandmothers and grandfathers aged 50–69 who report being either economically active (i.e. employed or self-employed) or economically inactive (i.e. retired or home-makers). Grandparents who are unemployed at the time of the interview (5.3% of the sample) are excluded, since unemployment has been found to have a different association from other statuses with grandparental child care (Lakomý and Kreidl, Reference Lakomý and Kreidl2015). Grandparents who report being ‘permanently sick or disabled’ (5.5% of the sample) are also excluded, as they are unlikely to provide daily grandchild care and retaining them may confound the association.

After excluding grandparents who do not meet the inclusion criteria or have missing values on any variable of interest, the samples for the analyses of the probability of employment consist of 16,976 grandmother–parent dyads corresponding to 11,164 grandmothers; and 11,092 grandfather–parent dyads corresponding to 7,393 grandfathers. For the analyses of average weekly working hours (intensive margin), the samples are further restricted to grandmothers and grandfathers who report being employed or self-employed at the time of the interview. This results in sample sizes of 5,975 grandmother–parent dyads corresponding to 4,161 grandmothers; and 4,735 grandfather–parent dyads corresponding to 3,298 grandfathers.

Measures

The primary outcome of analysis is grandparents’ employment. To study differences in employment at the extensive margin, a binary variable indicating whether a grandparent reports being employed or self-employed (as opposed to retired or a home-maker) at the time of the interview is coded. For the intensive margin, a continuous variable is used for the self-reported number of weekly working hours among employed or self-employed grandparents.

The main explanatory variable is a binary indicator of daily grandchild care provision. Grandparents are classified as providing grandchild care for the parent in each dyad if the grandparent reports looking after any young children (0–14) of that parent ‘almost daily’.

A set of controls for grandparents’ characteristics are included, which have been found to correlate with both employment and grandchild care provision (Komp et al., Reference Komp, Van Tilburg and Broese Van Groenou2010; Igel and Szydlik, Reference Igel and Szydlik2011; Lakomý and Kreidl, Reference Lakomý and Kreidl2015; Arpino and Bordone, Reference Arpino and Bordone2017), all measured at the time of the interview. Grandparents’ age is categorised into five-year groups to control for non-linear decreases in labour force participation by age, and for the fact that older grandparents may be less likely to provide grandchild care (Hank and Buber, Reference Hank and Buber2009). To account for joint decision-making among couples about retirement timing (Riedel et al., Reference Riedel, Hofer and Wogerbauer2015) and for the fact that non-married grandparents are usually less likely to provide care (Hank and Buber, Reference Hank and Buber2009), a variable combining marital status and partner's work status is included, categorised into whether respondents are not married, married to a partner who works for pay or married to a partner who does not work. In addition, total household size is controlled for.

Higher-educated and wealthier individuals tend to work until later in life (Komp et al., Reference Komp, Van Tilburg and Broese Van Groenou2010) and are generally more likely to provide grandchild care, but less likely to provide it daily (Igel and Szydlik, Reference Igel and Szydlik2011; Arpino and Bordone, Reference Arpino and Bordone2017). Using the International Standard Classification of Education (ISCED) 1997 codes, grandparents’ educational attainment is classified into three levels: low (up to lower-secondary education, ISCED 0–2), intermediate (upper secondary education and vocational training, ISCED 3–4) and high (tertiary education, ISCED 5–6). Controls are added for household net worth, which is the sum of all household assets minus liabilities, adjusted by household size and split into five quantile groups calculated separately by country; and rural (as opposed to urban) dwelling, which has been found to be positively associated with grandchild care and negatively with employment (Van der Meer, Reference Van der Meer2006; Arpino and Bordone, Reference Arpino and Bordone2017).

Since healthier individuals are more likely to be engaged in both grandchild care and work (De Preter et al., Reference De Preter, Van Looy and Mortelmans2013; Arpino and Bordone, Reference Arpino and Bordone2017), controls are added for grandparents’ health status as indicated by the number of limitations with activities of daily living (ADL). I control for depressive status, coded as binary using the Euro-D scale, where those reporting four or more depressive symptoms are considered ‘depressed’. Because depression is potentially endogenous to both employment and grandchild care provision (Kim et al., Reference Kim, Kang and Johnson-Motoyama2017), its coefficient should be interpreted with caution. However, excluding it from the model does not change the results. Since a partner's health status may also influence the decision to work (De Preter et al., Reference De Preter, Van Looy and Mortelmans2013), an indicator of whether the grandparent has a partner who suffers from any ADL limitations is included.

The characteristics of the parents are important determinants of grandchild care provision (Hank and Buber, Reference Hank and Buber2009; Igel and Szydlik, Reference Igel and Szydlik2011), and they may confound the association between grandchild care and employment. Given that grandparents with more adult children tend to be less likely to provide frequent grandchild care for each of them (Di Gessa et al., Reference Di Gessa, Glaser, Price, Ribe Montserrat and Tinker2016), the total number of adult children (parents) with children of their own (grandchildren) aged 0–14 is controlled for. Among the characteristics of the parents in each dyad, controls are included for gender, as mothers are more likely to receive grandchild care; and work status, as working parents have greater need for grandparental care (Hank and Buber, Reference Hank and Buber2009; Di Gessa et al., Reference Di Gessa, Glaser, Price, Ribe Montserrat and Tinker2016). The age of the parent's youngest child (grandchild) is split into four categories (0–2, 3–5, 6–10 or 11–14 years old) to account for varying child-care needs at different ages of the grandchild (Hank and Buber, Reference Hank and Buber2009). Finally, a control variable for geographical proximity of the parent in the dyad is added. Proximity is measured in kilometres and split into four categories according to whether the parent lives in the same building as the grandparent (after excluding those living in the same household), within 5 kilometres (km), between 5 and 25 km away or farther than 25 km away. While proximity is strongly correlated with grandchild care provision (Hank and Buber, Reference Hank and Buber2009), the association should be interpreted with caution, given that grandparents and their adult children may move closer to each other for the purpose of providing or receiving grandchild care.

To control for variation across SHARE countries in the average rates of grandchild care provision (Bordone et al., Reference Bordone, Arpino and Aassve2017) as well as in older adults’ labour market participation (Riedel et al., Reference Riedel, Hofer and Wogerbauer2015), country fixed effects (dummy variables) are included. Wave fixed effects account for demographic and socio-economic changes over time (Arpino and Bordone, Reference Arpino and Bordone2017), and are included as dummy variables for each SHARE wave (2004–2005, 2006–2007, 2011, 2013 and 2015).

In order to classify SHARE countries into the four typologies described above (optional de-familisation, service de-familisation, supported familism and familism by default), I draw two indicators from the Multilinks (2011) database: the percentage of children aged 0–2 enrolled in formal child care and effective parental leave, as defined above (Saraceno and Keck, Reference Saraceno and Keck2010). Multilinks (2011) indicators have been specifically constructed for comparative analysis of welfare regimes, and come from a theoretically driven database (Saraceno and Keck, Reference Saraceno and Keck2008). Unlike other policy databases, Multilinks (2011) provides single measures of composite concepts (e.g. effective leave), reducing complexity and facilitating comparisons. The data are available for 2004 and 2009. For countries first observed in SHARE Waves 1 and 2 (Austria, Belgium, Czech Republic, Denmark, France, Germany, Greece, Ireland, Italy, The Netherlands, Poland, Spain and Sweden), I take the average value of each indicator between the two years. For countries observed from SHARE Wave 4 onwards (Estonia, Hungary, Luxembourg, Portugal and Slovenia), only the 2009 values are used. Multilinks data are not available for Switzerland and Croatia, which are dropped from the analyses by country groups.

Figure 1 plots the two indicators against each other to show how countries fare with respect to their child-care policies. The horizontal line indicates the mean percentage of children in formal child care across the 18 countries (27.9), while the vertical line marks the cross-country average length of effective parental leave in weeks (9.3). The quadrant plot delineates four country groups that, despite some heterogeneity, share similar characteristics:

1. (1) Optional de-familisation: Denmark, France, Luxembourg, Slovenia and Sweden.

2. (2) Service de-familisation: Belgium, The Netherlands, Portugal and Spain.

3. (3) Supported familism: Austria, Czech Republic, Estonia, Germany and Hungary.

4. (4) Familism by default: Greece, Ireland, Italy and Poland.

Source: Multilinks (2011).

Figure 1. Classification of Survey of Health, Ageing and Retirement in Europe countries into four groups based on child-care service utilisation and effective parental leave in 2004–2009. The horizontal and vertical lines indicate cross-country average values for child-care utilisation and parental leave, respectively.

Statistical analysis

The analysis is conducted in two stages in order to handle the potential selection of grandparents into daily grandchild care and work. First, single-equation multivariate regression models of each employment outcome ($y_1^\ast$) are fitted on a binary indicator for grandchild care provision (y ₂) and a set of covariates (X):

$$y_1^{\rm \ast } = {\rm \;}\beta _0{\rm X} + \delta _0y_2 + {\rm \;}\varepsilon _0.$$

Probit regressions are fitted for the probability of being employed/self-employed and, for the sub-samples of grandmothers and grandfathers in paid work, linear regressions are fitted for self-reported weekly working hours. As argued above, the association between grandchild care and employment may be affected by selection even after controlling for the covariates, because of unobserved characteristics associated with both employment and daily grandchild care provision. These include grandparents’ preferences, family characteristics and values, as well as any changes in preferences that occurred in anticipation of having to provide grandchild care. In the presence of unobserved grandparental characteristics that are correlated with participation in both roles, ${\hat{\delta}}_0$ is biased since it does not isolate the difference in employment or working hours that is attributable to daily grandchild care provision from that attributable to such unobserved factors.

Second, to address the fact that employment and grandchild care may be jointly determined, the coefficients on grandchild care from single-equation models are compared to those obtained using a recursive bivariate approach (Heckman, Reference Heckman1978; Maddala, Reference Maddala1983). This is a system of equations in which employment (y ₁) and grandchild care (y ₂) represent two distinct outcomes regressed on a common set of covariates X. Continuous latent variables $y_1^\ast$ and $y_2^\ast$ are assumed to be underlying the observed binary outcomes y ₁ and y ₂. The error terms of the two equations are correlated with each other to account for the potential presence of unobservable characteristics associated with participation in both activities. Within the system, grandchild care is included as an endogenous binary regressor in the equation for employment:

$$y_1^{\rm \ast } = {\rm \;}\beta _1{\rm X} + {\rm \;}\delta _1y_2 + \varepsilon _1$$

$$y_2^{\rm \ast } = {\rm \;}\beta _2{\rm X} + {\rm \;}\varepsilon _2$$

where the error terms ɛ₁ and ɛ₂ follow a bivariate normal distribution with mean 0 and variance 1, and ρ = cov(ɛ₁, ɛ₂).

For the analytic samples of grandmothers and grandfathers, a recursive bivariate probit model is fitted for the probabilities of being (self-)employed and providing daily grandchild care, controlling for the common set of covariates and including grandchild care as a binary regressor in the equation for employment. For the sub-samples of grandmothers and grandfathers in paid work, hybrid (linear-probit) models are fitted, with weekly working hours treated as continuous, again including daily grandchild care as a dummy variable in the employment equation alongside the full set of covariates.

The estimated coefficient on grandchild care ${\hat{\delta}}_1\;$represents the difference in the probability of being employed or in working hours associated with daily grandchild care provision, controlling for the full set of covariates and allowing for covariance in the latent errors. In the absence of an instrument for the endogenous variable, the system of equations does not allow the direction of causality to be discerned. However, it isolates the ‘structural’ (or direct) association between grandchild care and employment from that attributable to selection (Heckman, Reference Heckman1978). Comparing the coefficient on grandchild care from the single equation model (${\hat{\delta}}_0$) to the one from the recursive bivariate model (${\hat{\delta}}_1$) allows assessment of whether unobserved confounders correlated with participation in both activities alter the association between them.

After conducting analyses on the pooled data across countries, I test for heterogeneity in the association across the four country groups described above. This is done by including an interaction term between daily grandchild care and the country group indicator in the equations for the probability of employment (bivariate probit recursive model) and weekly working hours (linear-probit recursive model).

In all sets of models, standard errors are clustered to handle the correlation of observations referring to different dyads within the same grandparent. The calibrated cross-sectional weights provided in SHARE (Börsch-Supan and Jurges, Reference Börsch-Supan and Jurges2005) are used to address differential inclusion probabilities and non-response, dividing the weights by the number of dyads to restore representativeness with respect to individual grandparents. The recursive bivariate models are fitted using the ‘cmp’ package in Stata 15 (Roodman, Reference Roodman2011; StataCorp, 2017).