(Grand)childlessness and depression across men and women’s stages of later life

ABSTRACT

The literature on how family status and health in later life relate is extensive. Although research has focused on the health effects of grandparenthood and grandparenting, explorations of whether ageing without children can lead to mental health impairments have achieved mixed results. We bridge empirical traditions to investigate the relationship between family status and mental health in Europe by using data from the Survey of Health, Ageing and Retirement in Europe (SHARE) and sampling roughly 160,000 men and women aged 50–89 from 17 European countries. Mental health is evaluated through depressive symptoms on the EURO-D scale and compared between childless or grandchildless people and grandparents at different ages. To identify the association between (grand)parenthood and mental health status net of confounders, we perform inverse probability-weighted regression adjustment (IPWRA). The rich information SHARE provides facilitates considering the common factors that may influence (grand)parenthood and depressive symptoms. Results show that the three groups hardly differ in mental health: grandchildless men and grandfathers reported fewer depressive symptoms, if any, than childless men aged 70–79. Overall, while mental health does not seem to relate to family status per se, it could be crucial when accounting for the entire individual life course.

KEYWORDS:

• Grandparents
• grandchildless
• childless
• EURO-D
• inverse probability weighted regression adjustment
• SHARE
• SHARELIFE

Introduction

This article considers whether family status in mid and later life determines individuals’ mental health, focusing on depressive symptoms among people with no children (‘childless’), people having children but no grandchildren (‘grandchildless’) and people with grandchildren (‘grandparents’). Mental health includes aspects of psychological, emotional, and social well-being and enables resilience against psychological and physical illnesses, with positive externalities on longevity. Thus, identifying social groups at risk of poor mental health is a valuable approach to health promotion and disease prevention (Soldevila-Domenech et al., 2021) to ensure the sustainability of long-term care systems.

Family ties and their absence are usually considered protective or risk factors for mental well-being. However, whether grandparents’ mental health differs from that of grandchildless individuals or lifetime childless people remains an open question. While one line of research has explored the association between mental health and grandparenthood or grandparenting – with inconsistent results (for a review, see Danielsbacka et al., 2022) – a parallel research thread has examined the association between childlessness and mental health, achieving similarly mixed findings (e.g. Dykstra & Wagner, 2007; Quashie et al., 2021).

Existing research on the association between grandparenthood and mental health often compares grandparents with a vague group of ‘others’ (Yang, 2021), often excluding childless people from the analysis (e.g. Bordone & Arpino, 2019; Sheppard & Monden, 2019; Tanskanen et al., 2019). Previous studies have thus excluded a critical and increasing section of the population: the lifetime childless. Research on childlessness and mental health in mid- or late-life also tends to compare the childless with all societal ‘others’ (e.g. Quashie et al., 2021), overlooking potential differences between grandparents and grandchildless older adults. Thus, the literature has so far considered childlessness and grandparenthood as two contrasting states in the reproductive history of a person, which has resulted in less attention being paid to the intermediate stage when parents have not (yet) transitioned to grandparenthood (for counterexamples, see Danielsbacka & Tanskanen, 2016; Powdthavee, 2011).

Grandchildless people are a group of interest. With the increase in the average age of first childbirth and a growing rate of childless individuals in many developed countries (Beaujouan & Sobotka, 2019; Sobotka, 2017), an increasing share of older individuals do not experience grandparenthood at all (Skopek, 2021), whether because their children become parents later or because their children remain childless.

The present study provides a systematic account of the empirical evidence on grandparenthood, (grand)childlessness and mental health in later life. It also extends the current literature by considering whether there exists an association between mental health and family status, broadly conceived by distinguishing between whether a person is (a) childless, (b) grandchildless or (c) a grandparent. In so doing, we attempt to help reconcile two isolated research strands.

Only two previous studies have explicitly accounted for differences in the three family statuses considered here: first, regarding self-reported life satisfaction of people over 40 in the UK (Powdthavee, 2011), and second, on self-evaluated happiness among the baby-boomer generation in Finland (Danielsbacka & Tanskanen, 2016). Aside from expanding empirical evidence to include a broader range of countries and a different indicator of mental health – depression – we consider each family status in an age-specific gendered structure to account for the fact that not having (grand)children at younger ages could have substantially different implications from not having them at older ages and that this pattern might be gender specific.

A challenge for research on the relationship between family states and mental health is that both the experience of family transitions and depressive symptoms are not randomly distributed. For example, health during childhood affects both the transition to (grand)parenthood and health in later life. Our analytical strategy thus addresses the concern of non-random selection into family states and depression by exploiting the rich array of information available from the Survey of Health, Ageing and Retirement in Europe (SHARE, Börsch-Supan et al., 2013) and applying inverse probability weighted regression adjustment (IPWRA) estimators (Petersen et al., 2006) to consider eventual common factors that may bias our estimates.

Theoretical background

The relation between family status and mental health in later life can be understood through two main pathways: first, as a selection effect and second, by focusing on the importance of family relationships for individuals’ well-being. The life course framework and its paradigmatic principles can help reconcile these approaches.

The selection effect suggests that confounding factors, such as early life-course events and circumstances, may influence the relationship between family status and mental health. From a life course perspective, the principle of life-span development (Elder et al., 2003) and cumulative inequality theory (Ferraro et al., 2009) suggest that experiences (e.g. advantages and disadvantages) and resources acquired earlier in life shape later life outcomes; early-life events can account for both psychological well-being and (grand)parental status, leading to observed differences in mental health across family-status groups. For example, financial hardship and health problems may hinder partnership formation and transitions to parenthood in early adulthood while also shaping health, including mental health, in later life (Pakpahan et al., 2017; Zimmer et al., 2016).

Much of the research has explored the role of family relationships in individuals’ well-being. The life course perspective’s principle of linked lives (Elder et al., 2003) highlights that family members are linked in various ways throughout their life stages. For example, relationships with children remain salient in later life: emotionally close and supportive ties can enhance well-being, while conflictual and strained relationships can undermine it (see Thomas et al., 2017; Umberson et al., 2010; Umberson & Thomeer, 2020). Family members can also mutually exchange time and resources: While adult children usually provide older parents with social support and care, older parents support adult children financially (Albertini et al., 2007) and provide grandchildren with care (Hank & Buber, 2009). Finally, grandparenthood, as a valued and desired role, can confer a sense of meaning and purpose (Mahne & Motel-Klingebiel, 2012; Thiele & Whelan, 2008; Villar et al., 2021; Werner et al., 1998).

Central to the life course perspective is the timing principle, which emphasizes that life transitions, behaviours, and events have varying consequences according to the timing of an individual’s life (Elder et al., 2003). Therefore, age is a crucial factor to consider when studying the implication of family status for well-being. In this context, the concept of normative timing (Furstenberg, 2005; see also ‘prescriptive timetables’ by Neugarten et al., 1965; and ‘cultural age deadlines’ by Settersten & Hägestad, 1996) refers to societal expectations regarding the appropriate age for specific life transitions. Individuals are aware of these expectations and consider their timing and whether they are ‘early’, ‘on time’, or ‘late’. These transitions also serve as ‘age reminders’ and influence age identity (Kaufman & Elder, 2003). Accordingly, early grandparenthood can make individuals feel older at a younger age (Bordone & Arpino, 2016) and may lead to lower life satisfaction or psychological well-being.

In addition, the timing of transitions – the age when they occur – can create role overlap and competing obligations (Leopold & Skopek, 2015). For example, young grandparents (<65, the approximate retirement age in many European countries) may experience role overlap between employment, grandparental childcare (Leopold & Skopek, 2015), and competing caregiving duties toward adult children or grandchildren and older parents (who may be alive if individuals become grandparents relatively early in life; see Železná, 2018). As off-time grandparenthood and role accumulation (Goode, 1960) could trump the positive externalities (e.g. a sense of purpose) of grandparenthood and lead to physical and psychological burden (Kim et al., 2019), we hypothesize that at younger ages, grandparents report more depressive symptoms than grandchildless and childless people (H1).

The normative timing and role overlap concepts suggest a different relationship between family status and mental health in later life. Remaining grandchildless at the normalized and expected age could be a source of psychological distress for those anticipating it (older parents vs childless individuals); conversely, being a grandparent later in life and caring for grandchildren can make grandparents feel more youthful (Bordone & Arpino, 2016; 2022). This perception of youth is a protective factor against depression (Keyes & Westerhof, 2012; Westerhof & Barrett, 2005). As people retire and free up time from paid employment, the risk of role overlap decreases, and they may no longer need to provide simultaneous support to older relatives and their children or grandchildren. Older grandparents could benefit from the physical activity and cognitive stimulation that comes with providing childcare (Arpino & Bordone, 2014) if they are not too old to engage with their grandchildren. Nevertheless, even if they face challenges in daily activities, older individuals may gain a sense of purpose and benefit from an extended family network to counteract loneliness and social exclusion. We thus expect that at older ages, grandparents report fewer depressive symptoms than grandchildless and lifetime childless people (H2).

Gender is an essential factor to consider when examining the role of family status on mental health throughout different life stages, including later ones. For example, in previous and older cohorts, women have traditionally filled the roles of full-time housemakers and family caregivers (Leopold et al., 2018). Women are thus likelier to be ‘sandwiched’ between two generations – older parents and (grand)children – in need of care and shoulder a high accompanying psychological burden (Grundy & Henretta, 2006; Železná, 2018). In addition, as traditional carers for family members, women may feel deeper empathy for family members than men (Rotkirch & Janhunen, 2010), so they may emotionally suffer more than men from the lack of (grand)children. Although role overlap decreases with age and grandmothers’ greater responsibility for childcare (compared to grandfathers) could benefit their mental health, recent evidence suggests that grandparenthood is highly valued by men as an opportunity to make up for lost time with their children (Airey et al., 2021; Mann, 2007). Finally, grandfathers also actively engage in grandparental childcare (Di Gessa et al., 2020), so older men could also experience the age-related burden and the benefits of family status for mental health, similar to women.

The possibility of no interaction between gender and age is also worth considering. As depressive symptoms tend to increase with age, and women tend to suffer from depression more than men (Buber & Engelhardt, 2011), there could be a ‘ceiling effect’ for older women whose mental health is neither alleviated nor exacerbated by family status. However, given the potentially competing expectations of gender-based differences regarding family status and mental health, we do not formulate specific hypotheses on this relationship.

Literature review

The following sections review two distinct streams of research on the relationship between family status and health in mid and later life: one focuses on grandparental childcare or being a grandparent, and the other on childlessness (see also Appendix Tables A1 and A2). From a life course perspective, however, this division fails to capture the dynamic and evolving nature of family status and its relation with mental health.

Grandparenthood and mental health

Ample literature discusses grandparental well-being concerning non-custodial grandparental childcare (for reviews, see Danielsbacka et al., 2022; Kim et al., 2017). With few exceptions (for Germany, Ates, 2017; and across countries, Brunello & Rocco, 2019), empirical evidence across Europe supports a positive association between non-cohabitant grandparents’ provision of childcare and their well-being. Childcare stimulates grandparents’ cognitive functioning (Arpino & Bordone, 2014); has a positive effect on their happiness (Danielsbacka & Tanskanen, 2016), subjective well-being (Arpino et al., 2018; Mahne & Huxhold, 2015), and subjective health (Di Gessa et al., 2016a, 2016b); and protects them from depression (Arpino & Gómez-León, 2020). Typically, these studies focus only on grandparents (i.e. providing vs not providing childcare for grandchildren) and exclude middle- and older-aged grandchildless and childless individuals from their samples.

Among studies that integrate older parents with no grandchildren into their analyses, the idea that grandparents are better off than grandchildless individuals is well-supported (see Appendix, Table A1). Cross-sectional evidence from SHARE shows that grandparents tend to have higher life satisfaction across gender and levels of education, although the association varies by country, and its magnitude is generally limited (Arpino et al., 2018). However, investigations that exploit longitudinal variation in the same data reveal more heterogeneous results: becoming a grandmother is related to fewer depressive symptoms, but becoming a grandfather has no effect (Bordone & Arpino, 2019; Di Gessa et al., 2020; Sheppard & Monden, 2019). By expanding the analysis to more countries, Yang (2021) finds significant contextual differences: becoming a grandparent reduces depressive symptoms in lower-income countries, while the opposite is true in higher-income countries. Nevertheless, Tanskanen et al. (2019) conclude that no relation exists between the transition to grandparenthood and a change in depressive symptoms. Most of these studies also exclude childless people from their analyses. In doing so, they miss out on comparing individuals who may expect to become grandparents (grandchildless) and childless individuals who are aware they will never have grandchildren.

Only two studies include older childless individuals in their analyses. Based on data from Understanding Society, Powdthavee (2011) shows that grandparents display the highest self-reported life satisfaction, while no differences emerge between childless and grandchildless individuals. Danielsbacka and Tanskanen (2016) support this finding in a Finnish study which concludes that grandparents report a significantly higher level of happiness. However, when adjusted for several covariates, the differences between the three groups are no longer statistically significant.

From the literature reviewed, despite extensive research, it remains unclear whether grandparents’ mental health differs from that of grandchildless or childless people.

Childlessness, parenthood, and later life mental health

A separate field of research has investigated the relationship between parenthood and mental health in mid and later life. In this case, childless individuals are often compared with all others regardless of the presence of grandchildren. Although this comparison is meaningful when analyzing the association between parenthood and mental health during early adulthood (e.g. Aassve et al., 2012; Myrskylä & Margolis, 2014), considering grandparents and grandchildless parents as homogenous groups could obscure crucial differences in mental health in late life.

The evidence on the association between childlessness and mental health in later life is mixed (see Appendix, Table A2). A long-standing debate thus continues regarding whether childless individuals face social isolation, loneliness, and a lack of support as they age (Keizer et al., 2010; O’Bryant, 1985; Pinquart, 2003; Wenger et al., 2007; Zoutewelle-Terovan & Liefbroer, 2018).

Extensive evidence from cross-sectional studies at the European level shows childlessness to be negatively associated with mental health, even when accounting for variations in individual characteristics related to marital and socioeconomic conditions (Buber & Engelhardt, 2008; Dykstra & Keizer, 2009; Dykstra & Wagner, 2007; Grundy et al., 2019; Hansen et al., 2009; Huijts et al., 2013; Kendig et al., 2007; Quashie et al., 2021 for Hungary and Czechia). Some studies also report that men suffer more from a lack of children than women, both in terms of depression (Buber & Engelhardt, 2008; Kendig et al., 2007) and life satisfaction (Bauer et al., 2022; Dykstra & Wagner, 2007), while others find that childlessness is associated with increased depressive symptoms only for women (Grundy et al., 2019).

Other studies find opposing results, showing that parents in general or parents of many children tend to be more depressed than childless people (Hank & Wagner, 2013; Moor & Komter, 2012). This pattern has been observed in research conducted in Southern Europe (Gibney et al., 2017) and Hungary and Belgium (Quashie et al., 2021). Some analyses report no differences between parents and childless people (Hank, 2010 for Germany; Hank & Wagner, 2013; Huijts et al., 2013; Keenan & Grundy, 2019; Quashie et al., 2021). Likewise, some studies confirm that childless people do not experience support deficits in later life and instead possess more extensive support networks than parents, relying more on voluntary associations, public services, and emotional support from non-relatives (Albertini & Mencarini, 2014; Schnettler & Wöhler, 2016). Finally, mixed results have emerged from longitudinal studies, which can account for pre-existing mechanisms related to both childlessness and mental health. Some report no effect of childlessness on mental health (Keenan & Grundy, 2019), while others indicate that being childless is linked to the worsening of mental health only for some groups: women living in Western countries (Grundy et al., 2019); and older age groups, for example married, widowos, and divorced women (Mikucka, 2020 for Switzerland). In contrast, some evidence reports age trajectories in life satisfaction to be lower for mothers than childless women in their 70s (Bauer et al., 2022).

Previous research thus provides inconsistent evidence on the association between childlessness and mental health. Nevertheless, several life-course factors (previous social, economic and health conditions) are crucial in predicting both childlessness and late-life mental well-being.

Data and methods

Data and sample

To answer our research question, we employ data from eight waves of the SHARE carried out between 2004/2005 and 2019/2020 (Börsch-Supan, 2022a, 2022b, 2022c, 2022d, 2022e, 2022f, 2022g, 2022h; Börsch-Supan et al., 2013). SHARE is a biennial, cross-national panel survey providing longitudinal micro-data on the health, social, and economic conditions of people aged 50 and older living in private households across 28 European countries and Israel. Further information on, e.g. sampling design, response rate, and refreshment, can be found elsewhere (Bergmann et al., 2019). Waves 3 and 7 are the SHARELIFE modules, which gather retrospective information on several life domains of respondents throughout their life course (Börsch-Supan, 2022c, 2022g).

Our sample for analysis includes respondents who participated in either of the SHARELIFE modules, are between 50 and 89 years old and belong to one of the following European countries: Austria, Belgium, Croatia, Czech Republic, Denmark, Estonia, France, Germany, Greece, Italy, Luxemburg, Netherlands, Poland, Slovenia, Spain, Sweden, and Switzerland. We exclude from the sample foreign-born individuals. After sample selection and the deletion of missing information on key variables (see section ‘Variables’),¹ our sample comprises roughly 160,000 observations, as shown in Table 1.

Table 1. Sample distribution by family status (treatment), age group, and gender.

	Men					Women
	50–59	60–69	70–79	80–89	Total	50–59	60–69	70–79	80–89	Total
Childless (C): N	2,077	2,543	1,557	592	6,769	1,877	2,258	1,765	1,139	7,039
%	12.75	9.6	8.03	8.08	9.74	8.17	6.97	7.36	11.11	7.85
Grandchildless (T1): N	8,222	5,718	1,753	427	16,120	9,910	5,240	1,731	533	17,414
%	50.48	21.59	9.04	5.83	23.19	43.11	16.17	7.22	5.2	19.43
Grandparents (T2): N	5,989	18,228	16,088	6,308	46,613	11,201	24,899	20,482	8,578	65,160
%	36.77	68.81	82.94	86.09	67.07	48.73	76.86	85.42	83.69	72.71
Total	16,288	26,489	19,398	7,327	69,502	22,988	32,397	23,978	10,250	89,613

Source: SHARE waves 1–8 (N = 159,115).

Variables

We measure mental health using the EURO-D depression scale, an additive scale ranging from 0 to 12 on self-reported depressive symptoms: depressive mood, pessimism, suicidality, guilt, sleep, interest, irritability, appetite, fatigue, concentration, enjoyment, and tearfulness (Prince et al., 1999). Therefore, 0 indicates the absence of depressive symptoms, while 12 indicates severe depression. Depression caseness is usually associated with a threshold of four symptoms (Courtin et al., 2015). The scale’s validity is evident in its use in many other health measures; it is internally consistent and validated across European countries (Costa et al., 2008; Prince et al., 1999).

We distinguish whether an individual is (a) childless, (b) grandchildless or (c) a grandparent depending on whether they (a) never had any biological or adopted children, (b) had at least one biological or adopted child during his life but no grandchildren, (c) had at least a grandchild, during various stages of life: 50–59, 60–69, 70–79, and 80–89 years old.

As we aim to estimate the effect of family states on mental health, we also consider several individual life-course characteristics that may drive the selection of individuals into (grand)parenthood: these will be employed in the treatment model (see the ‘Analytical Strategy’ section). To maximize the possibility that these characteristics affect parenthood and are not their consequence, we measured them at ages when the respondent was unlikely to be a parent. These characteristics were retrieved from the SHARELIFE modules (Börsch-Supan, 2022c, 2022g).

The first set of variables measures individual life conditions during childhood (i.e. before age 16). These include self-reported health (excellent and good vs fair, poor, and varied a great deal); summary count of stressful events (having missed a month or more of school, having had parents drinking heavily or with mental health problems, having experienced financial hardship or difficult living arrangements, top coded to 2); the number of books in the household (none or very few; one bookshelf; one bookcase or more). The family’s socioeconomic background is proxied by the father’s level of education (primary vs secondary and more).

The second set of variables refers to the respondent’s early adulthood (between 16 and 25 years old) and whether they have experienced severe stress, financial hardship, or poor health (measured with verbatim questions). Also, we measure whether the respondent was still living in the parental home by age 20 as a proxy for partnership formation.²

Finally, we include the respondent’s educational level – primary, secondary, and tertiary education – and country of residence, corresponding to the birth country.

Our analytical strategy allows for double robustness as the outcome (depressive symptoms) can be corrected for non-random treatment assignments. First, we employed a set of variables similar to the treatment model in the outcome model (see the below section on the analytical strategy), measured at different points in the life course. Second, we added a set of covariates to the respondent’s childhood (as above) and a set of covariates to the respondent’s early adulthood and adulthood (16–48 years old, differently than before), namely, whether they had ever experienced severe stress, financial hardship, and poor health. Third, we measured the respondent’s situation in mid and later life one year before the interview: whether the respondent was living with a partner, was still in employment, or suffered from poor health. Finally, we added controls for country of residence and educational level.

Analytical strategy

To identify the association between grandparenthood and mental health, net of confounders, we combine inverse probability weighted (IPW) and regression adjustment (IPWRA)³ (Petersen et al., 2006). We aimed to compare the mental health scores of three groups of individuals:

1. Childless – control group (C)

2. Grandchildless – treated group 1 (T1)

3. Grandparents – treated group 2 (T2)

As mentioned above, comparing the mean score between the groups could be biased by confounding. IPW allows us to condition our estimates on a vast array of information provided in the SHARE; in this way, these groups can be considered identical except for the transition to (grand)parenthood (the treatments). Any difference in depressive scores between C and T1 and C and T2 is the average impact of the treatments.

Let a treatment t be dichotomous (e.g. being childless vs grandparent). IPW aims at balancing treatment and comparison groups on a given set of covariates. The potential outcome Y_t is the outcome that an individual would have under a given condition, t. Each person has two potential outcomes, Y₁ if they received the treatment (t = 1, is grandparent), and Y₀ if they did not (t = 0, is childless). If we could observe (under identical conditions) the mental health score when a person is childless (Y₀) and, simultaneously, the mental health score when the same person is a grandparent (Y₁), the difference between the two mental health scores Y₁–Y₀ would be the measure of the impact of t. However, this situation does not exist in the real world, where we can only observe one of two potential outcomes: individuals can be either a grandparent or childless.

Given the nature of observational data, t is not assigned randomly to the population under study, but there might be confounding due to common causes affecting the treatment and the outcome. Being childless or not a grandparent and experiencing depression might not be directly associated: they might be mediated by a third common factor, e.g. poor health during young adulthood, which influenced partnership formation (leading to no children) and mental health in later life. In this example, individuals with poor health during adulthood could be oversampled in the childless group.

The IPW estimator weighs the data to balance the composition of treated and comparison groups: those having an unlikely or rare treatment status given their covariates are assigned higher weights than those who, given their covariates, are likelier to have observed treatment status. Following our example, individuals who were in poor health during young adulthood and are now childless would receive smaller weights than those who had poor health and are now grandparents. As such, previous poor health is evenly distributed across childless and grandparents in the reweighted dataset. In the Appendix (see Tables A3 and A4), we report covariate balancing before and after weighting to show that after weighting, their distribution is almost identical between treatment and control groups: the standardized differences are close to zero, and the variance ratios close to 1. Once weighted by the inverse probability of treatment, we can invoke a consistency assumption: If a subject (S1) has not received the treatment, their observed outcome equals (is ‘consistent with’, in the jargon) the potential outcome under the control condition, Y₀; for a subject (S2) who has received the treatment, their observed outcome is consistent with the potential outcomes under treatment, Y₁. The mean of the difference between Y₁ and Y₀ is the average treatment effect (ATE): $\mathrm{ATE}=E\left(Y_1-Y_0\right)$Operationally, weights are estimated via the treatment model. Using multinomial logistic regression, we estimate a model for the probability of being childless, grandchildless, or a grandparent as a function of a series of regressors. The inverse of this predicted probability is then used as a weight for comparing the average mental health score of the treated and the control group. Grandparents are assigned weights inversely proportional to the probability of being a grandparent, conditional on the covariates observed for that person, and so on for (grand)childless individuals. The variables we include in the treatment model refer to the respondent’s childhood (health, stressful events, books at home and family socioeconomic background) and early adulthood (stress, financial hardship, and poor health), as well as including partnership formation, and the respondent’s educational level and country of residence (for details, see the paragraph discussing the variables).

After computing the treatment model, as previously illustrated, IPWRA uses the estimated weights to compute weighted linear regression models for the outcome at each treatment level (the outcome model). IPWRA is a doubly robust estimator because it remains consistent should either the outcome or the treatment model be misspecified (Słoczyński et al., 2022). In the outcome model, we include all the variables referring to conditions during childhood (health, stressful events, books at home and family socioeconomic background), early adulthood and adulthood (severe stress, financial hardship, and poor health), as well as in mid and later life (cohabitation with a partner; employment status; poor health) (see ‘Variables’). At this point, the mean depression score for the three groups can be compared (ATE).

As previously highlighted, family transitions can assume different meanings at different life stages: not being a grandparent by age 50 (when few people already are) is crucially different from not having grandchildren when 75, when most people have performed the transition. Similarly, the family network becomes more crucial for support later in life when health inevitably deteriorates. Finally, grandparenthood may not be equally associated with mental well-being in women and men. For these reasons, our analyses are divided by age groups and performed separately by gender.

Additionally, we add cluster standard errors at the individual level: since SHARE is biennial, individuals could have repeated observations within each age group; keeping only one observation per individual in each age group thus does not affect our results.

Full models are available in the online supplementary material (Tables S1–S12).

Our approach relies upon several fundamental assumptions. First is the conditional independence assumption (CIA), meaning that once we control all observable variables, the potential outcome is independent of the treatment. In the present study, we believe the theory-oriented choice of covariates supported by the richness of information contained in SHARE should minimize the chance of omitted-variable bias. Second is the overlap assumption, meaning that all individuals have a positive probability of receiving the treatment; thus, a sufficient overlap in the joint distribution of the covariates between treated and control groups makes the groups comparable. The overlap is fully met in each age group of our male and female samples, as shown in the Appendix (Figure A1–A2).

Descriptive results

Table 1 shows the distribution of our sample by treatment status, age group and gender.

Most individuals in our sample are grandparents: 67% of men and almost 73% of women have at least one grandchild. Predictably, grandparenthood is strongly related to age: while less than 37% of men aged 50–59 are grandparents, the percentage reaches 86% in the group 80–89. This is reflected in the percentage of grandchildless individuals decreasing with age (e.g. from 51% at 50–59 to 6% at 80–89, respectively). The pattern is similar for women: for older age groups, the percentage of grandmothers increases, and that of grandchildless mothers decreases. However, women tend to become grandmothers earlier in life compared to men: for example, at 50–59, almost 49% of women are already grandmothers, whereas the share jumps up to 83% for women aged 80–89 years old.

The share of childless people is more stable during mid to late life than in earlier stages. However, a slight decrease emerges, among men, with age (from 13% at the age of 50–59 to 8% at the age of 80–89), whereas an increase is found among women (8% vs 11%). There could be multiple reasons for differences in the share of childlessness across age groups, including men’s longer reproductive window and the sample’s changing composition and attrition rate.

Ageing itself appears to be associated with mental health outcomes. Figure 1 shows the relation between mental health scores, family status, and age group by gender. Women in all age groups tend to suffer more depressive symptoms than men. For example, by age 50–59, men report suffering from two symptoms on average, with women reporting 2.5 symptoms. For both men and women, the experience of depressive symptoms strongly increases with ageing. However, while men generally never reach depression caseness (4 symptoms or more), most women approach the threshold when aged 70–79.

Figure 1. Average number of depressive symptoms among men (left panel) and women (right panel) aged 50 and older who are childless, grandchildless, and grandparents.

Source: SHARE waves 1–8 (N = 159,115).

Men who are childless appear to be more depressed than grandchildless and grandfathers across all age groups, while only minor differences emerge between the latter two groups. For women, differences are less evident across the three groups, and childless women resemble grandmothers across all age groups. Grandchildless women are those who display fewer depressive symptoms among younger respondents. However, they seem to lose an initial advantage with age, as they overtake childless women and grandmothers by reporting 3.5 depressive symptoms by 80–89.

Is there a relationship between family status and mental health?

Next, we investigate the extent to which family status impacts mental health scores across classes of age, net of confounding factors.

The complete IPWRA estimates for men and women are reported in Supplementary Tables S1–S6 and S7–S12, respectively. Those tables include estimates of the determinants of treatment (i.e. being childless, grandchildless, or grandparent) and the outcome equations (i.e. depressive symptoms). Treatment models for men (Tables S1–S2) and for women (Tables S7–S8) confirm the importance of early-life characteristics for family status: for example, poor self-reported health in early life is associated with a higher likelihood of being childless at age 50–69 rather than parent or grandparent. Outcome models also show an association between early-life characteristics and poor mental health later in life for both men (Tables S3–S5) and women (and S9–S11): again, people who have suffered health problems, stress and financial hardship in childhood and early adulthood report higher depressive symptoms in later life. This evidence confirms the importance of considering confounding factors when studying depressive scores in mid- and later life between childless, grandparents and grandchildless individuals, as a specific family status could follow a pre-existing mental health condition rather than being a determinant.

Figures 2 and 3 show results from the IPWRA estimation for men and women, respectively. The black symbols (dot and diamond) report the average treatment effects (ATEs), which indicate the differences in symptom number in the mental health score between the treatment (T) and the control (C) group on the left y-axis. The dots signify the difference between grandchildless (T1) and childless (C) individuals, while the diamonds represent the difference between grandparents (T2) and childless people (C). The grey bars provide the potential outcome means (POMs) of age-specific depressive symptoms for the childless (C), the baseline category, referring to the right-hand side y-axis.

Figure 2. ATEs on EURO-D mental-health score (black symbols, left y-axis), and POMs of EURO-D mental-health score (grey bars, right y-axis), by age groups. Men.

Note: 95% confidence intervals are displayed. GrChildless = Grandchildless. Results from IPWRA and complete estimates are available in online supplementary material, Tables S1–S12.: Source: SHARE waves 1–8 (N = 69,502).

Figure 3. ATE on EURO-D mental-health score (black symbols, left y-axis), and POM of EURO-D mental-health score (grey bars, right y-axis), by age groups. Women.

Note: 95% confidence intervals are displayed. GrChildless = Grandchildless. Results from IPWRA and full estimates are available in online supplementary material, Tables S1–S12. Source: SHARE waves 1–8 (N = 89,613).

Figure 2 displays the results for men. As the descriptive evidence (Figure 1) shows, childless men’s mental health worsens with age (see the grey bars on the right y-axis). Specifically, men’s depressive symptoms increase from almost 2 in their 50s to almost 2.5 depressive symptoms in their 80s, without ever constituting clinical depression. ATEs (black dots and diamonds, left-hand y-axis) are very close to zero, meaning there are no substantive differences in mental health between childless people and our treatment groups (i.e. grandchildless and grandparents) across age groups, except for the group 70–79. At that age, grandchildless men report 0.22 fewer symptoms, and grandfathers 0.14 fewer symptoms than childless men, suggesting that mental health at this age is somewhat less favourable for childless men than parents, regardless of the presence of grandchildren.

We performed a t-test to test for statistically significant differences between the two treated groups (i.e. grandchildless vs grandparents), again not detecting any relevant difference. In substantive terms, we reject our hypotheses: results indicate that family status does not influence depression scores in men from mid to late life, once accounting for confounding, except for the age group 70–79.

Figure 3 displays the results for women; like for men, the number of depressive symptoms for childless women increases with age. At all stages of their life course, women report more depressive symptoms than men and pass the threshold for severe depression by the age of 80–89. Thus, our hypotheses are rejected in this case: the results do not reveal any statistically significant (dis)advantage in the mental health of childless women compared with grandchildless or grandmothers, as ATEs are negligible in size and not statistically significant. In addition, for women, a t-test provides no evidence of significant differences in depression scores between the grandchildless and grandmothers.

Conclusion and discussion

The present research attempts to contribute to the vast literature on family status and mental health. First, we organized the empirical literature on the topic by outlining two streams of research which have not yet been in dialogue. Studies that analyze the consequences of childlessness for mental health in later life traditionally compare childless older people with parents, forgetting that some of them are not only parents but also grandparents. Conversely, studies analyzing the consequences of grandparental childcare focus only on grandparents; and studies on being a grandparent compare older people who have grandchildren with older adults who have no grandchildren, often excluding childless people. By doing so, previous research has overlooked one piece of the puzzle that constitutes the picture of family statuses in older age.

Thus, the second contribution of the study was to compare the mental health outcome of older people who never had children (i.e. childless) with that of older people who have at least one child but no grandchildren (i.e. grandchildless) and people who are both parents and have at least one grandchild (i.e. grandparents). Before this study, only two articles distinguished these three family statuses, examining different health outcomes in two countries. Powdthavee (2011) considered (grand)parenthood in later life in the UK, finding evidence of a protective effect of having grandchildren on life satisfaction. On the other hand, in Finland, Danielsbacka and Tanskanen (2016) studied the consequences of grandparenthood on happiness and found no indication that grandparenthood is associated with better happiness than parents or childless people.

Building on previous evidence, we provided three additional contributions to understanding the complex relationship between family status and mental health outcomes in later life. First, unlike earlier studies, we focused on an indicator more directly connected with mental health status and manifested in depressive symptoms. Second, we distinguished age groups when examining the association between family status and mental health. Third, from a life course perspective, we recognized that family status might entail benefits or burdens for mental health according to the stage of life in which it is experienced. Additionally, we analyzed the association between family states and mental health on a larger scale, focusing on the European level.

To do this, we used individual data from 17 European countries from the eight waves of the SHARE carried out between 2004–2005 and 2019–2020. A common challenge for research on (grand)parenthood and mental well-being in later life is that the transition to (grand)parenthood is not randomly distributed among people; instead, it may be associated with various confounding demographic and socioeconomic factors affecting mental health. Therefore, to yield unbiased estimates of the (grand)parenthood effect, we employed a doubly robust treatment effect estimator, which employs the inverse of the predicted probabilities of being childless, grandchildless or grandparents and combines them with regression adjustment (IPWRA).

We hypothesized that, at younger ages, grandparents reported higher depressive symptoms compared to grandchildless and childless individuals due to the accumulation of roles and making the transition ‘off-time’ according to societal age norms. At later ages, our hypothesis was of a protective effect of grandparenthood when role overlap, and competing caregiving obligations are likely to decrease. Social support to (grand)children is supposed to confer life meaning and purpose, with a positive spilover on cognition and physical activity.

The descriptive results partially support our hypotheses, with younger grandparents reporting slightly more depressive symptoms than grandchildless individuals. However, contrary to our expectations, there is no evidence of an advantage to grandparenthood at older ages, as the three groups converge in the number of depressive symptoms. Regardless of family status, depressive symptoms become more severe with age. Childless men have the highest depressive symptoms, while childless women report similar scores to grandmothers.

Once common factors that could influence (grand)parenthood and depressive symptoms are considered, the evidence does not support the idea that family status relates to mental health in later life, at least regarding depressive symptoms. Contrary to our hypotheses, at younger ages, we detect no differences in mental health scores among the three groups under study: for women, differences do not emerge in later life; for men, the grandchildless individuals and grandfathers are slightly better off than childless individuals in terms of depressive symptoms, but only when they are 70–79. Moreover, grandchildless men and grandfathers have similar scores in this age group, so we cannot attribute this effect solely to the ‘rejuvenating’ role of having grandchildren ‘in time’. Perhaps as individuals age, they are less likely to be depressed if able to rely on an extended family network composed of only adult children or grandchildren.

Nevertheless, these differences did not persist for the oldest group (80–89 years old), supporting the notion of a ‘ceiling’ effect: as depressive symptoms strongly increase with age, family status does not alleviate or exacerbate mental imbalance. Overall, our results could be considered aligned with those of Danielsbacka and Tanskanen (2016), who examined grandparenthood in Finland, and other studies in non-European contexts (e.g. Quashie et al., 2021).

This study is not without limitations. First, we used a more robust method than previous studies, but unobservable characteristics may still bias our estimates. Given the need to include controls applicable to all three groups (childless, grandchildless, and grandparents), we could not include further information on partners or children, whose characteristics (e.g. partner’s ill health, see Walker & Luszcz, 2009) and life events (e.g. children’s unemployment, see Albertini & Piccitto, 2022) could also impact the mental health of the respondents. The grandchildren’s age, for example, could be highly relevant, as it can strongly influence grandparental childcare obligations. In addition, as SHARE data only collect the youngest grandchild’s birth date, it was impossible to measure when someone became a grandparent if it did not happen during the observation window.

Second, we did not further explore potential sources of heterogeneity beyond age and gender that could mask the absence of differences in mental health between the three groups. Future research could investigate how factors such as marital or socioeconomic status may moderate the association between family status and depression in later life by affecting, for example, social support networks. A natural progression of this work would involve exploring variations across different geographical contexts; recent research indicates that becoming a grandparent reduces depression in low-income countries but increases it in high-income countries (Yang, 2021). Similarly, exploring the moderating role of contextual factors, such as family and active ageing policies, that relieve individuals’ dependence on the family (both in terms of support expected by adult children and given to grandchildren), enhancing longer and healthy lives (Boudiny, 2013), is an intriguing topic for future research.

Finally, our groups are broadly defined by family states themselves, not by the nature of family relationships. We do not distinguish between (grand)parents with or without contacts or support obligations with their descendants. We recognize that family status’s detrimental or protective effect on mental health depends, among other things, on contacts, the amount of support exchanged, and geographical proximity (Albertini & Arpino, 2018). Although the intersection of many dimensions would have made several treatment groups hard to handle due to sample size and covariates balancing, our approach provides further evidence in this direction, leading to our study’s central message: merely having (grand)children is not the point with respect to mental health. Rather, our study confirms that an individual’s past life experiences and circumstances play a more significant role in determining the risk for depression in later life (Pakpahan et al., 2017).

The present findings align with the current EU-level policy discourse on preventing, limiting, and postponing the multifarious challenges associated with ageing (European Commission, 2021). Adopting a life-cycle approach to ageing involves promoting healthy and active ageing and lifelong learning from early childhood. The former aims to instil healthy lifestyles that positively impact health status throughout the life course, until older ages, with positive externalities on the labour market and social protection systems. In this regard, public policies such as the EU4Health programme can be crucial in disease prevention and ensuring accessibility, affordability, and quality of healthcare services. The latter emphasizes the investment in knowledge, skills, and competencies, from early childhood education and care to higher education and beyond. Education is associated with numerous outcomes, including better health, cognitive functioning, and employability – even in later life. This framework thus implies that healthier individuals could experience strong intergenerational relationships as meaningful and purposeful without overburdening family members with caregiving responsibilities.

Declarations

All authors whose names appear on the submission contributed equally to the manuscript.

Acknowledgement

This paper uses data from SHARE Waves 1, 2, 3, 4, 5, 6, 7, 8 and 9 (DOIs: 10.6103/SHARE.w1.800, 10.6103/SHARE.w2.800, 10.6103/SHARE.w3.800, 10.6103/SHARE.w4.800, 10.6103/SHARE.w5.800, 10.6103/SHARE.w6.800, 10.6103/SHARE.w7.800, 10.6103/SHARE.w8.800, 10.6103/SHARE.w8ca.800, 10.6103/SHARE.w9ca800), see Börsch-Supan et al. (2013) for methodological details.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

The SHARE data collection has been funded by the European Commission, DG RTD through FP5 (QLK6-CT-2001-00360), FP6 (SHARE-I3: RII-CT-2006-062193, COMPARE: CIT5-CT-2005-028857, SHARELIFE: CIT4-CT-2006-028812), FP7 (SHARE-PREP: GA N°211909, SHARE-LEAP: GA N°227822, SHARE M4: GA N°261982, DASISH: GA N°283646) and Horizon 2020 (SHARE-DEV3: GA N°676536, SHARE-COHESION: GA N°870628, SERISS: GA N°654221, SSHOC: GA N°823782, SHARE-COVID19: GA N°101015924) and by DG Employment, Social Affairs & Inclusion through VS 2015/0195, VS 2016/0135, VS 2018/0285, VS 2019/0332, and VS 2020/0313. Additional funding from the German Ministry of Education and Research, the Max Planck Society for the Advancement of Science, the U.S. National Institute on Aging (U01_AG09740-13S2, P01_AG005842, P01_AG08291, P30_AG12815, R21_AG025169, Y1-AG-4553-01, IAG_BSR06-11, OGHA_04-064, HHSN271201300071C, RAG052527A) and from various national funding sources is gratefully acknowledged (see www.share-project.org).

Notes

1 Our final sample includes N = 159,115 observations (n, individuals = 46,325), and results from the following procedure. Overall, SHARE includes n = 139,620 and N = 383,647. First, we excluded from the analytical sample individuals who answered the SHARELIFE modules in neither wave 3 nor wave 7 (N = 75,586, n = 48,421). Second, we excluded individuals who did not provide information on their childhood situation (N = 67,267, n = 30,396); over half of missing records relate to respondents’ father’s education, which was asked only in waves 5, 6, or 8. We also deleted individuals who were not born in the country of interview (or had missing information on their birthplace, N = 24,265, n = 6,137). In addition, we deleted individuals who did not provide information on Euro-D (N = 41,357). The high number of missing observations result from wave 7, as only respondents surveyed in wave 3 (SHARELIFE Interview) were asked this question in wave 7.

Further, we deleted people who were either younger than 50 or older than 89 (N = 4,537) and a listwise deletion was performed for people lacking information on the remaining covariates (N = 2,153). As the final step, we examined the sample size for each country and retained only those countries where there were at least 100 observations for each family status. Therefore, we excluded Bulgaria, Cyprus, Finland, Hungary, Israel, Latvia, Lithuania, Malta, Portugal, Romania and Slovakia (N = 9,367).

2 Including in the model a variable measuring partnership formation in more detail (e.g. ever had a partner: yes/no) would violate the overlap assumption (see ‘Analytical Strategy’). Childless individuals are by far the least likely to have formed a partnership during their life course. Conditioning on an explicit partnership variable would make the probability of receiving the treatment significantly different for the three groups we formed. For the cohort under study, leaving the parental home often corresponded with starting a family (Angelini & Laferrère, 2013), making living status at the age of 20 a reasonable proxy for partnership formation.

3 We implemented the IPWRA approach with the Stata© command teffects ipwra.

Appendix

Table A1. Synthetic review of peer-reviewed articles investigating the association between grandparenthood and mental health in mid- and later-life.

Author	Sample	MH indicator	Data	Eu country(es)	Main result about grandparenthood and MH
Arpino et al. (2018)	GC vs GP, age 50–84	Life satisfaction	SHARE 1–2	14	GPs have better MH across gender and educational level (with country differences).
Bordone and Arpino (2019)	GC vs GP, age 50–84	Depression (EURO-D scale)	SHARE 1–2, 4–6	17	Becoming first-time GP increases MH for women, not men (with country differences).
Danielsbacka and Tanskanen (2016)	GP, GC, CH, age 62–67	Happiness	Survey data 2012	Finland	GPs are happier than GC and CH. No significant differences after controls added (gender, marital status, health, socio-economic status, religion, friends, and relative contact)
Di Gessa et al. (2020)	GC vs GP, age 50+	Life satisfaction, Depression (EURO-D)	SHARE 4–6	13	Becoming a first-time GP via daughters increases MH for women, no change for additional grandchildren.
Powdthavee (2011)*	GP, GP no alive children, GC, vs CH, age 40+	Life satisfaction	Understanding Society 1	UK	Being a GP is positively associated with MH, while GC is not.
Sheppard and Monden (2019)	GC vs GP, age 50+	Depression (EURO-D scale), Life satisfaction, Subjective life expectancy	SHARE 4–6	15	Becoming a GP increases MH for women, not men.
Tanskanen et al. (2019)	GC vs GP, age 50–90	Life satisfaction, Quality of life (CASP scale), Depression (EURO-D scale)	SHARE 2, 4–6	13	Becoming a GP is not associated with changes in MH for both men and women.
Yang (2021)	All vs GP, age 50–90	Depression (CES-D scale)	ELSA 5–7, SHARE 4–6	14	Becoming a GP is associated with increased MH for both men and women.

Note: The table includes published papers on peer-reviewed journals, with European data, focusing on mental health. *Exception: Powdthavee (2011) since extensively cited. MH = Mental health; GP = Grandparent; GC = Grandchildless; CH = Childless.

Table A2. Synthetic review of peer-reviewed articles investigating the association between childlessness and mental health in mid- and later-life.

Author	Sample	MH indicator	Data	Eu country(es)	Main result about childlessness and mental health
Dykstra and Wagner (2007)	CH vs PA, age 70–89.	Life satisfaction	Berlin Aging Study; Dutch survey on the Living Arrangements and Social Networks of Older Adults	Amsterdam, Berlin	Netherlands: CH men have lower MH than PA; worst MH for women who outlived their children. Germany: PA unrelated to MH.
Kendig et al. (2007)	CH vs PA, age 65–85	Depression (feeling low spirit or depressed; CES-D scale), Difficulty of falling asleep	Survey of Living Conditions in Finland; Longitudinal Aging Study Amsterdam:	Finland, the Netherlands	Finland: formerly married CH men have more depression and more difficulties falling asleep than formerly married fathers. CH women have less difficulty in falling asleep and similar depression than PA. Netherlands: No differences CH-PA. Never married CH women least likely to report sleeping problems.
Buber and Engelhardt (2008)	CH vs PA, age 60+	Depression (EURO-D scale)	SHARE 1	11	CH have lower MH than PA among men, not women. PA with more children have better MH than CH among men, not women.
Dykstra and Keizer (2009)	CH men vs Fathers, age 40–59	Life satisfaction scale, Mental health index, Problem behaviour	The Netherlands Kinship Panel Study	the Netherlands	CH men and non-coresident fathers have lower MH than men cohabiting with children. No differences CH-PA after control for partnership history.
Hansen et al. (2009)	CH vs PA, age 40–80	Life satisfaction, Depression (CES-D Scale)	The Norwegian Life Course, Ageing and Generation Study	Norway	Women: PA have better life satisfaction than CH. Men: living with a child relates to better life satisfaction than being CH. No differences CH-PA in depression, across gender and marital status.
Hank (2010)	CH vs PA, age 50–75	Mental health score (from exploratory factor analysis)	The German Socio-Economic Panel	Germany	No differences CH-PA, across West and East Germany.
Moor and Komter (2012)	CH vs PA, age 20–85	Depression (CES-D scale)	Generation & Gender Survey 1	5	PA have lower MH than CH. Only women with children aged > 12 have better MH than CH women after controls (health, economic and financial situation)
Hank and Wagner (2013)	CH vs PA, age 65+	Depression (EURO-D scale)	SHARE 1–2	15	No differences CH-PA.
Huijts et al. (2013)	CH vs PA, age 40+	Psychological wellbeing (CES-D scale)	European Social Survey 3	24	No differences CH-PA
Gibney et al. (2017)	CH vs PA, age 55–75	Depression (EURO-D scale)	SHARE 2, SHARELIFE	18	CH have better MH in Southern Europe; No differences CH-PA in North, West and East Europe.
Grundy et al. (2019)	CH vs PA, age 65–80	Depression (CES-D score)	Generation & Gender Survey 1–2	9	Being CH reduces MH in Eastern countries, only for women in Western counties.
Keenan and Grundy (2019)	CH vs PA, age 50–79	Depression (EURO-D scale)	SHARE 1–2	10	Being CH not associated with changes in MH.
Mikucka (2020)	CH vs PA, age 60–89	Life satisfaction	Swiss Household Panel	Switzerland	The effect of being CH on MH depends on marital status and age: being CH decreases MH at older age (i.e. 75+) for married and widows. Among the divorced, being CH increases/decreases MH for men/women.
Quashie et al. (2021)	CH vs PA, age 50+	Depression (felt depressed in the week prior the interview)	SHARE 4–5	17	CH have lower MH in Czechia, the Netherlands, Spain and Sweden, better MH in Hungary. After controls (place of living, employment status, education, income) association remains only for Czechia and Hungary, and a negative association appears for Belgium.
Bauer et al. (2022)	CH vs PA, age 50–85	Life satisfaction	SHARE 2, 4–6	18	CH men have lower life satisfaction in all age groups, except for the oldest, and for women in the youngest or oldest age groups. After controls (cohort, country, marital status, education, and period) differences are reduced. No differences in life satisfaction trajectories emerge, except for mothers having lower life satisfaction than CH women in their 70s.

Note: The table includes published papers on peer-reviewed journals, with European data, focusing on mental health. MH = Mental health outcome; GP = Grandparent; GC = Grandchildless; CH = Childless; PA = Parents (no distinction GC/GP).

Table A3. Diagnostic statistics for covariate balance over treatment groups: standardized differences and variance ratios for the raw data and the weighted sample. Men.

Grandchildless (T1)	50–59				60–69				70–79				80–89
	Standardized differences		Variance ratio		Standardized differences		Variance ratio		Standardized differences		Variance ratio		Standardized differences		Variance ratio
	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted
Childhood (<age 16)
Poor self-reported health	−0.24	−0.01	0.54	0.99	−0.14	−0.01	0.70	0.98	−0.07	−0.01	0.84	0.97	−0.09	−0.05	0.77	0.88
N stressful events	−0.08	0.01	0.81	0.98	−0.13	0.02	0.80	1.09	−0.11	0.00	0.80	0.97	0.01	−0.01	0.91	0.83
Books in household (ref. none/few)
One bookshelf	0.02	−0.02	1.02	0.98	−0.03	−0.01	0.96	0.99	0.09	0.04	1.14	1.05	0.07	−0.01	1.09	0.99
One bookcase or more	0.12	0.02	1.01	1.00	0.13	0.03	1.03	1.01	0.00	−0.02	1.00	0.99	−0.01	−0.06	0.98	0.94
Father education (ref. primary)	0.06	0.00	0.99	1.00	0.04	0.02	1.00	1.00	0.16	0.02	1.02	1.00	0.31	0.02	1.23	1.01
Early adulthood (age 16–25)
Bad health	−0.25	0.02	0.34	1.08	−0.18	0.02	0.39	1.13	−0.11	0.03	0.55	1.16	−0.02	0.00	0.89	1.02
Ever stress	−0.06	0.02	0.82	1.07	−0.03	0.01	0.91	1.03	−0.02	0.01	0.90	1.06	−0.02	−0.01	0.92	0.97
Financial hardship	−0.07	0.00	0.77	1.01	−0.08	0.02	0.71	1.08	−0.07	0.00	0.77	0.99	−0.01	0.01	0.98	1.02
In parental home (age 20)	0.03	−0.01	0.96	1.01	−0.02	−0.02	1.03	1.03	0.07	0.05	0.88	0.92	0.12	0.06	0.82	0.90
Country (ref. Austria)
Germany	−0.10	0.00	0.77	0.99	0.00	0.00	1.00	1.01	0.14	0.01	1.45	1.02	0.16	−0.02	1.74	0.94
Sweden	0.10	−0.02	1.63	0.94	0.03	0.01	1.11	1.03	−0.1	0.02	0.73	1.05	−0.06	−0.02	0.83	0.94
Netherlands	−0.04	0.00	0.82	1.01	0.02	0.00	1.09	1.00	0.07	0.02	1.57	1.10	0.10	−0.01	2.05	0.92
Spain	−0.01	0.00	0.98	0.99	−0.03	−0.01	0.93	0.96	−0.11	−0.02	0.76	0.95	−0.12	0.01	0.78	1.01
Italy	0.14	−0.02	1.43	0.94	0.22	−0.03	1.67	0.92	0.25	−0.04	1.75	0.91	0.19	−0.01	1.53	0.98
France	0.03	0.01	1.11	1.02	−0.02	0.02	0.93	1.05	−0.13	0.03	0.64	1.11	−0.19	0.04	0.60	1.13
Denmark	0.09	0.02	1.29	1.04	−0.07	0.00	0.79	1.01	−0.15	0.01	0.56	1.03	−0.09	0.01	0.73	1.04
Switzerland	−0.06	−0.01	0.81	0.98	−0.02	0.00	0.95	1.00	0.02	0.01	1.05	1.03	0.06	0.01	1.25	1.04
Belgium	−0.09	0.00	0.82	1.01	−0.15	0.02	0.67	1.05	−0.19	0.01	0.58	1.03	−0.24	−0.02	0.55	0.96
Czech Republic	0.11	0.03	1.80	1.12	0.06	0.00	1.32	1.01	0.03	0.00	1.16	0.99	0.18	0.01	2.47	1.04
Luxembourg	−0.03	0.00	0.81	1.02	0.02	0.00	1.13	0.97	0.05	−0.01	1.48	0.93	0.09	0.01	2.06	1.05
Slovenia	−0.04	−0.01	0.83	0.96	0.01	−0.01	1.03	0.97	0.05	0.02	1.40	1.11	0.14	0.02	2.94	1.12
Estonia	−0.05	−0.01	0.79	0.97	−0.02	0.01	0.88	1.05	−0.01	0.00	0.96	1.01	−0.02	−0.01	0.94	0.97
Greece	−0.01	−0.01	0.95	0.91	0.03	0.00	1.35	0.97	0.06	0.00	1.76	0.95	0.11	0.01	2.74	1.16
Poland	−0.08	−0.01	0.51	0.95	−0.05	0.00	0.55	0.95	0.07	−0.01	2.36	0.90	−0.02	−0.01	0.69	0.93
Croatia	0.01	0.01	1.06	1.09	−0.01	0.00	0.91	0.97	−0.04	−0.01	0.69	0.90	0.07	0.00	1.83	1.03
Educational level (ref. primary)
Secondary	−0.06	−0.02	1.02	1.01	0.02	−0.03	1.00	1.01	0.14	0.01	1.00	1.00	0.08	−0.01	1.02	1.00
Tertiary	0.20	0.01	1.21	1.01	0.14	0.01	1.15	1.01	0.12	−0.03	1.16	0.96	0.10	0.06	1.19	1.09

Table

Grandparents (T2)	50–59				60–69				70–79				80–89
	Standardized differences		Variance ratio		Standardized differences		Variance ratio		Standardized differences		Variance ratio		Standardized differences		Variance ratio
	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted
Childhood (<age 16)
Poor self-reported health	−0.17	0.00	0.67	0.99	−0.12	0.00	0.74	0.99	−0.03	0.02	0.93	1.04	−0.03	−0.04	0.93	0.91
N stressful events	−0.05	0.01	0.82	0.92	−0.11	0.00	0.83	1.02	−0.11	0.00	0.82	1.01	0.01	−0.01	0.96	0.91
Books in household (ref. none/few)
One bookshelf	0.05	−0.02	1.06	0.98	0.02	0.01	1.03	1.01	0.10	0.01	1.15	1.02	0.01	−0.05	1.02	0.94
One bookcase or more	0.02	0.02	1.00	1.00	0.03	0.02	1.01	1.01	0.02	−0.01	1.01	0.99	0.12	−0.02	1.15	0.98
Father education (ref. primary)	0.06	0.00	0.99	1.00	0.02	0.02	1.00	1.00	0.07	0.01	1.02	1.00	0.25	0.00	1.20	1.00
Early adulthood (age 16–25)
Bad health	−0.24	0.01	0.35	1.04	−0.14	0.01	0.52	1.06	−0.07	0.02	0.72	1.14	−0.03	0.01	0.87	1.07
Ever stress	−0.05	0.01	0.83	1.05	−0.08	0.00	0.74	1.01	−0.05	0.01	0.83	1.04	−0.08	0.00	0.73	0.99
Financial hardship	0.05	0.00	1.18	1.00	0.03	0.02	1.13	1.07	0.04	0.01	1.14	1.03	0.09	−0.01	1.36	0.97
In parental home (age 20)	−0.19	−0.01	1.23	1.01	−0.1	−0.01	1.16	1.02	−0.01	0.05	1.01	0.93	0.09	0.07	0.86	0.89
Country (ref. Austria)
Germany	−0.14	0.00	0.67	1.01	−0.13	0.00	0.67	1.00	−0.02	0.01	0.94	1.02	0.05	−0.01	1.23	0.95
Sweden	0.13	−0.02	1.88	0.93	0.09	0.01	1.37	1.02	0.03	0.02	1.10	1.05	0.03	0.00	1.09	1.00
Netherlands	−0.11	0.01	0.56	1.04	0.02	0.00	1.10	0.99	0.10	0.02	1.81	1.14	0.10	0.00	2.08	1.02
Spain	−0.21	−0.02	0.51	0.95	−0.13	−0.01	0.70	0.98	−0.08	−0.02	0.82	0.96	−0.03	−0.01	0.95	0.98
Italy	−0.09	−0.03	0.73	0.92	−0.04	−0.03	0.89	0.93	0.05	−0.04	1.13	0.91	0.02	−0.01	1.06	0.97
France	0.08	0.01	1.28	1.03	0.05	0.01	1.17	1.03	−0.09	0.02	0.76	1.06	−0.21	0.01	0.57	1.05
Denmark	0.06	0.01	1.18	1.04	0.05	0.01	1.15	1.03	−0.02	0.02	0.95	1.06	0.00	0.01	0.99	1.02
Switzerland	−0.21	0.00	0.42	1.00	−0.2	0.00	0.48	0.99	−0.13	0.00	0.64	1.01	−0.04	0.00	0.86	1.00
Belgium	−0.04	0.00	0.92	1.00	−0.05	0.01	0.89	1.03	−0.1	0.01	0.77	1.02	−0.16	0.00	0.70	0.99
Czech Republic	0.34	0.03	4.14	1.14	0.31	0.01	3.24	1.03	0.26	−0.01	2.83	0.98	0.26	0.02	3.31	1.07
Luxembourg	−0.16	0.00	0.18	1.00	−0.07	0.00	0.53	0.99	−0.04	−0.01	0.67	0.95	0.00	0.00	1.00	0.98
Slovenia	0.04	−0.01	1.18	0.98	0.09	−0.01	1.56	0.97	0.13	0.01	2.07	1.06	0.21	0.03	4.09	1.16
Estonia	0.25	−0.01	2.30	0.97	0.12	0.00	1.71	1.01	0.05	0.00	1.21	0.99	−0.01	0.00	0.95	1.00
Greece	−0.12	−0.02	0.26	0.83	−0.05	0.00	0.56	0.96	−0.01	−0.01	0.89	0.92	0.03	0.01	1.45	1.09
Poland	0.01	0.00	1.04	1.00	0.02	0.00	1.16	0.99	0.04	0.00	1.67	0.94	0.06	−0.02	2.42	0.85
Croatia	0.01	0.01	1.06	1.08	0.04	0.00	1.31	1.01	0.01	0.00	1.07	0.97	−0.01	0.01	0.94	1.05
Educational level (ref. primary)
Secondary	0.16	−0.02	0.90	1.01	0.12	−0.01	0.96	1.00	0.10	0.03	1.01	1.00	0.05	0.01	1.02	1.00
Tertiary	−0.07	0.01	0.92	1.01	−0.02	0.01	0.98	1.01	0.07	−0.03	1.10	0.97	0.11	0.04	1.18	1.07

Source: SHARE waves 1–8 (N = 69,502).

Table A4. Diagnostic statistics for covariate balance over treatment groups: standardized differences and variance ratios for the raw data and the weighted sample. Women.

Grandchildless (T1)	50–59				60–69				70–79				80–89
	Standardized differences		Variance ratio		Standardized differences		Variance ratio		Standardized differences		Variance ratio		Standardized differences		Variance ratio
	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted
Childhood (<age 16)
Poor self-reported health	−0.17	0.02	0.66	1.05	−0.23	0.01	0.60	1.02	−0.1	0.04	0.82	1.10	−0.03	0.01	0.94	1.04
N stressful events	−0.12	0.02	0.71	0.94	−0.14	0.02	0.72	0.99	−0.02	0.02	0.92	1.00	0.03	−0.01	1.19	1.10
Books in household (ref. none/few)
One bookshelf	−0.01	0.00	0.98	1.01	0.03	−0.03	1.04	0.96	0.03	−0.05	1.04	0.94	−0.01	−0.02	0.99	0.97
One bookcase or more	0.03	−0.01	1.00	1.00	−0.07	0.03	1.00	1.01	−0.1	0.02	0.96	1.01	−0.2	0.03	0.85	1.03
Father education (ref. primary)	−0.05	−0.04	1.02	1.01	−0.01	−0.03	1.00	1.00	0.02	0.00	1.00	1.00	−0.1	−0.01	0.96	1.00
Early adulthood (age 16–25)
Bad health	−0.16	−0.01	0.48	0.96	−0.15	−0.02	0.51	0.90	−0.05	−0.01	0.81	0.97	0.01	0.01	1.03	1.03
Ever stress	−0.13	−0.03	0.70	0.92	−0.09	0.03	0.78	1.09	−0.03	0.02	0.89	1.07	−0.12	0.00	0.60	1.02
Financial hardship	−0.09	0.01	0.76	1.04	−0.07	0.03	0.80	1.08	−0.04	0.01	0.90	1.02	−0.02	0.06	0.96	1.16
In parental home (age 20)	−0.05	−0.02	1.03	1.00	−0.01	0.02	1.01	0.99	−0.08	0.02	1.09	0.99	−0.13	−0.03	1.17	1.03
Country (ref. Austria)
Germany	−0.07	−0.03	0.82	0.92	0.05	0.00	1.14	1.01	0.17	0.01	1.75	1.03	0.11	−0.01	1.61	0.96
Sweden	0.09	0.01	1.52	1.05	0.05	0.01	1.21	1.04	0.03	−0.04	1.12	0.90	0.06	−0.01	1.24	0.97
Netherlands	−0.03	0.01	0.88	1.04	−0.03	0.01	0.87	1.05	−0.05	−0.01	0.77	0.94	0.05	−0.01	1.35	0.93
Spain	0.05	0.00	1.16	0.99	0.03	−0.01	1.09	0.98	0.06	0.01	1.17	1.02	0.10	0.01	1.22	1.03
Italy	0.05	0.01	1.12	1.02	0.08	0.02	1.20	1.05	0.02	0.00	1.05	1.01	0.09	0.03	1.34	1.12
France	0.04	−0.01	1.13	0.98	−0.01	0.01	0.95	1.04	−0.06	0.00	0.84	1.01	−0.02	0.00	0.95	0.99
Denmark	0.04	0.00	1.14	1.00	−0.04	−0.01	0.88	0.96	−0.05	−0.04	0.80	0.88	−0.1	0.01	0.67	1.03
Switzerland	−0.12	0.00	0.68	0.99	−0.11	0.01	0.75	1.02	−0.03	0.02	0.90	1.09	0.05	0.03	1.19	1.12
Belgium	−0.06	0.02	0.87	1.04	−0.14	0.05	0.68	1.15	−0.17	0.03	0.65	1.09	−0.16	−0.01	0.68	0.98
Czech Republic	0.08	0.01	1.50	1.02	0.08	−0.05	1.43	0.88	0.09	0.00	1.43	1.01	−0.04	0.02	0.76	1.08
Luxembourg	0.00	0.00	0.99	0.98	0.05	0.00	1.43	1.02	0.04	−0.02	1.42	0.85	−0.03	0.00	0.77	1.03
Slovenia	0.08	−0.01	1.64	0.96	0.01	−0.01	1.06	0.96	0.06	−0.01	1.35	0.94	0.08	−0.03	1.50	0.90
Estonia	0.00	0.01	1.02	1.04	−0.03	−0.01	0.88	0.98	−0.03	0.00	0.92	1.00	−0.1	−0.01	0.75	0.97
Greece	−0.08	0.00	0.59	1.02	0.00	0.00	0.97	0.98	0.01	−0.01	1.17	0.89	−0.01	−0.02	0.86	0.78
Poland	−0.02	−0.02	0.82	0.89	−0.02	0.00	0.77	1.01	0.02	0.01	1.70	1.11	0.05	0.01	2.13	1.09
Croatia	0.01	0.01	1.13	1.05	−0.01	−0.03	0.94	0.77	−0.05	0.01	0.58	1.12	0.00	−0.01	0.95	0.92
Educational level (ref. primary)
Secondary	−0.02	0.00	1.01	1.00	−0.03	0.02	1.01	0.99	0.09	0.04	0.99	1.00	−0.09	0.07	0.98	1.02
Tertiary	0.01	0.00	1.01	1.00	0.01	−0.01	1.01	0.99	−0.15	−0.01	0.80	0.99	−0.13	−0.06	0.75	0.87

Table

Grandparents (T2)	50–59				60–69				70–79				80–89
	Standardized differences		Variance ratio		Standardized differences		Variance ratio		Standardized differences		Variance ratio		Standardized differences		Variance ratio
	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted	Raw	Weighted
Childhood (<age 16)
Poor self-reported health	−0.09	0.02	0.81	1.05	−0.17	0.01	0.70	1.02	−0.14	0.03	0.75	1.07	−0.12	0.03	0.78	1.08
N stressful events	−0.05	0.02	0.86	0.98	−0.12	0.01	0.75	0.98	−0.06	0.00	0.88	0.99	−0.01	−0.01	1.11	1.13
Books in household (ref. none/few)
One bookshelf	0.02	0.00	1.03	1.00	0.04	−0.02	1.05	0.97	0.02	−0.03	1.03	0.96	0.04	0.00	1.06	1.00
One bookcase or more	−0.11	0.00	1.00	1.00	−0.16	0.02	0.98	1.01	−0.16	0.02	0.92	1.01	−0.2	−0.01	0.84	0.99
Father education (ref. primary)	−0.07	−0.03	1.02	1.01	−0.08	−0.02	1.01	1.00	−0.05	0.03	0.99	1.01	−0.06	−0.01	0.98	0.99
Early adulthood (age 16–25)
Bad health	−0.1	0.01	0.67	1.03	−0.13	−0.01	0.57	0.94	−0.07	−0.01	0.73	0.95	−0.04	0.00	0.86	1.00
Ever stress	−0.07	−0.03	0.84	0.93	−0.09	0.01	0.77	1.04	−0.01	0.02	0.98	1.07	0.01	0.00	1.04	0.99
Financial hardship	0.06	0.01	1.17	1.02	0.05	0.02	1.16	1.05	0.04	0.01	1.11	1.03	0.02	0.03	1.06	1.08
In parental home (age 20)	−0.46	−0.02	1.08	1.00	−0.29	0.01	1.15	1.00	−0.28	0.01	1.29	0.99	−0.16	−0.02	1.21	1.02
Country (ref. Austria)
Germany	−0.14	−0.02	0.67	0.95	−0.06	0.00	0.83	1.01	0.02	0.00	1.07	0.99	0.04	−0.01	1.21	0.96
Sweden	0.08	0.01	1.47	1.05	0.13	0.00	1.58	1.01	0.09	−0.05	1.32	0.88	0.05	−0.01	1.18	0.96
Netherlands	−0.06	0.01	0.73	1.06	0.00	0.00	1.00	1.02	−0.03	−0.02	0.86	0.92	0.05	−0.01	1.36	0.95
Spain	−0.1	−0.01	0.74	0.98	−0.03	0.01	0.91	1.02	0.04	0.02	1.10	1.04	0.03	0.01	1.06	1.02
Italy	−0.15	−0.02	0.65	0.96	−0.1	0.03	0.76	1.09	−0.04	0.02	0.90	1.05	0.07	0.04	1.27	1.12
France	0.10	−0.01	1.38	0.99	0.02	−0.01	1.06	0.97	−0.08	−0.02	0.78	0.94	−0.14	0.00	0.69	1.00
Denmark	0.04	0.00	1.11	0.99	0.06	−0.01	1.23	0.96	0.08	−0.03	1.37	0.89	0.01	0.01	1.05	1.02
Switzerland	−0.27	0.01	0.35	1.03	−0.29	0.00	0.37	1.02	−0.16	0.02	0.55	1.08	−0.08	0.02	0.74	1.09
Belgium	−0.06	0.02	0.86	1.04	−0.07	0.02	0.83	1.07	−0.13	0.02	0.72	1.07	−0.12	0.00	0.77	0.99
Czech Republic	0.32	0.01	3.43	1.05	0.31	−0.03	2.84	0.93	0.27	0.03	2.53	1.07	0.25	0.01	2.87	1.04
Luxembourg	−0.12	0.01	0.33	1.09	−0.04	0.00	0.67	1.01	−0.01	−0.02	0.94	0.84	−0.05	0.00	0.58	0.95
Slovenia	0.21	0.00	2.83	0.99	0.11	0.00	1.65	1.00	0.12	−0.02	1.84	0.92	0.14	−0.03	1.86	0.89
Estonia	0.23	0.01	2.19	1.02	0.06	−0.01	1.24	0.96	−0.07	−0.01	0.81	0.98	−0.09	−0.01	0.76	0.98
Greece	−0.18	−0.01	0.21	0.93	−0.05	0.00	0.59	1.00	0.00	−0.02	1.01	0.84	0.01	−0.01	1.09	0.92
Poland	0.07	−0.01	1.69	0.92	0.02	0.01	1.17	1.08	0.07	0.02	3.71	1.26	0.05	0.01	2.38	1.12
Croatia	0.06	0.01	1.55	1.05	0.02	−0.03	1.14	0.83	0.02	0.01	1.18	1.06	0.03	−0.01	1.40	0.95
Educational level (ref. primary)
Secondary	0.13	−0.01	0.94	1.00	0.02	0.04	0.99	0.99	0.03	0.04	1.00	1.00	−0.05	0.01	0.99	1.00
Tertiary	−0.26	0.01	0.77	1.01	−0.18	−0.01	0.81	0.99	−0.19	−0.02	0.74	0.97	−0.13	−0.04	0.74	0.92

Source: SHARE waves 1–8 (N = 89,613).

Figure A1. Estimated densities of the probability of getting each treatment level, by age group. Men. Source: SHARE waves 1–8 (N = 69,502).

Figure A2. Estimated densities of the probability of getting each treatment level, by age group. Women. Source: SHARE waves 1–8 (N = 89,613).