Introduction
Across western societies, grandparents play an important role in caring for their grandchildren. As with the unpaid work of women in the male breadwinner model (Folbre, Reference Folbre2001), the care grandparents provide is largely unrecognised and gendered in that it is mainly provided by grandmothers (Sear and Coall, Reference Sear and Coall2011). Making visible this caring role is especially important in the UK where grandparents support high levels of mothers’ participation in paid work but simultaneously face an extension of their own paid working lives due to the raising of the state pension age (Pensions Act, 2011; 2014). Gray (Reference Gray2005) highlighted this tension a decade ago in this journal but the academic literature has paid surprisingly little subsequent attention.
From the angle of maternal employment, the contemporary UK context has increased in interest for two reasons. Firstly, mothers’ participation has grown rapidly to reach 69.6 per cent by 2014 (ONS, 2014). The participation of first-time mothers of babies under the age of one had even exceeded this figure, reaching 74 per cent by the millennium (Kanji, Reference Kanji2010). Although part-time work for mothers has become normative in the UK (Himmelweit and Sigala, Reference Himmelweit and Sigala2004), mothers’ full-time employment has also increased (Soobedar, Reference Soobedar2011) resulting in considerable variation in the duration of mothers’ working hours. Grandparents’ childcare is particularly relevant in this respect because previous studies associated it with a greater intensity of mothers’ work (Dimova and Wolff, Reference Dimova and Wolff2011).
Secondly, decisions about work and care in the UK have been made against the backdrop of some of the highest childcare costs in Europe (Kenjoh, Reference Kenjoh2005) and in sharp contrast to the recent situation in Germany. Childcare costs have significantly impacted on women's participation (Del Boca and Vuri, Reference Del Boca and Vuri2007). The experience elsewhere in Europe is that, where there is a shortage of childcare or its cost is high, the intensity of grandparents’ childcare increases to fill the void (Attias-Donfut and Wolff, Reference Attias-Donfut, Wolff, Arber and Attias-Donfut2000).
Recent studies in European and North American contexts have started to make visible grandparents’ specific role in raising mothers’ labour force participation (Dimova and Wolff, Reference Dimova and Wolff2008; Reference Dimova and Wolff2011; Aassve et al., Reference Aassve, Arpino and Goisis2012; Compton and Pollak, Reference Compton and Pollak2014). These studies quantified an often substantial role played by grandparents while highlighting considerable differences in the size of effects between countries. Further evidence of cross-country variation has been provided in comparative studies of the intensity of grandparents’ childcare (Igel and Szydlik, Reference Igel and Szydlik2011; Di Gessa et al., Reference Di Gessa, Glaser, Price, Ribe and Tinker2016).
This article investigates the effect of grandparents’ care on mothers’ paid work in terms of both participation (the extensive margin) and hours of work (the intensive margin) in the UK. Attention is on mothers because caring has largely been constructed as a maternal responsibility. The study focuses on mothers with at least one child of primary school entry age as childcare demands vary considerably with children's ages (see for example McKay's Reference McKay2004 study of low-income families in the UK). The analysis examines potential differences in the grandparents’ role in lone as opposed to dual-parent households, building on the potentially differential importance of grandparents in lone-mother and step-parent households (Attar-Schwartz et al., Reference Attar-Schwartz, Tan, Buchanan, Flouri and Griggs2009; Ruiz and Silverstein, Reference Ruiz and Silverstein2007; Henderson et al., 2009). The final investigation is into the impact of mothers’ educational attainment on labour force participation.
Literature review and theoretical background
Grandparents’ care for grandchildren is much in evidence in European countries other than the UK. In their study of ten European countries using the Survey of Health, Ageing and Retirement in Europe (SHARE) data (which excludes the UK), Hank and Buber (Reference Hank and Buber2009) found a high prevalence of grandparents’ childcare even where they least expected it in Sweden, Denmark and France. In line with expectations because of limited publicly available childcare, the intensity of care was greatest in Italy, Spain and Greece.
The cross-country differences in grandparents’ childcare illustrate one aspect of how the welfare state context and features of the labour market influence the expression of intergenerational relationships (Szydlik, Reference Szydlik2012). These institutional factors play an important role in the widely diverging patterns of grandparent participation, which do not readily conform to Esping-Andersen's European welfare-state typology (Aassve et al., Reference Aassve, Arpino and Goisis2012). According to Saraceno and Keck's (Reference Saraceno and Keck2010) categorisation of states based on the degree of familialisation, the UK fell into a medium category among the 26 EU member states with regard to the provision of childcare for children aged three to six years and with regard to the degree of public provision of childcare. The USA provides a still different view into how institutional and family cohesion condition outcomes across generations: grandparents were the main source of childcare for up to 23 per cent of pre-schoolers (Laughlin, Reference Laughlin2013).
Research quantifying the effect of grandparents’ childcare on mothers’ labour force participation has varied in its findings. Dimova and Wolff's (Reference Dimova and Wolff2011) study, which pooled the SHARE data for ten countries, highlighted how grandparents’ childcare raised maternal participation in paid work. Aassve et al.’s (Reference Aassve, Arpino and Goisis2012) analysis of seven countries from the Gender and Generations Survey (again excluding the UK) found considerable variation across countries with regard to whether grandparents’ childcare raised mothers’ labour force participation: in France, Germany, Bulgaria and Hungary grandparents’ childcare had a positive and statistically significant impact on mothers’ participation; while in Russia, Georgia and the Netherlands it did not. In the Italian context of limited formal childcare, Arpino et al. (Reference Arpino, Pronzato and Tavares2014) found, using an instrumental variables approach, an extremely large effect: Italian mothers helped by grandparents were 32 percentage points more likely to work. A much smaller, but still important, effect of 5.1–6.2 per cent was found in the USA for married women in receipt of grandparents’ childcare (Compton and Pollak, Reference Compton and Pollak2014).
It is well recognised that demographic changes have lengthened the opportunities for intergenerational relationships and exchange (Bengtson, Reference Bengtson2001; Uhlenberg, 2005). Grundy et al. (Reference Grundy, Murphy and Shelton1999) found that, at age 20, over 80 per cent of the population in the UK had a living grandparent. Tan et al.’s (2010) study reported that 94 per cent of maternal grandmothers met their grandchildren at least twice a year, while 57 per cent met on a weekly or daily basis. Cohort changes in age at first birth mean that, on average, grandparents in the UK had their children at younger ages than subsequent cohorts so that, in terms of age, today's grandparents are able to look after their grandchildren; an opportunity which may not endure (Gray, Reference Gray2005; see also, for Russia, Utrata, Reference Utrata2011).
The grandparent literature categorises forms of intergenerational resource flow in terms of six distinct forms of solidarity comprised of different types of shared activities. Policy and institutional environments shape the type of assistance provided; so that, where formal care is more available, grandparents have channelled their support into financial gifts and financial support (Dimova and Wolff, Reference Dimova and Wolff2011). In the UK, grandparents seem to provide care specifically to assist parents to work (La Valle et al., Reference La Valle, Finch, Nove and Lewin2000 and McKay, Reference McKay2004). Examining European data from SHARE, Dimova and Wolff (Reference Dimova and Wolff2011) showed that the effect of care was to raise mothers’ participation, in contrast to financial transfers from grandparents which did not raise participation.
We might therefore expect that grandparents’ care is provided specifically to help mothers into work and that it raises mothers’ participation, either through filling a care void or because parents prefer its relational qualities (Wheelock and Jones, Reference Wheelock and Jones2002). This leads to the following hypotheses:
H1: Grandparents’ childcare raises mothers’ participation.
H2: Grandparents’ childcare raises the extent of mothers’ participation.
At least two competing explanations have been put forward for why grandparents provide care, based on exchange and altruism (Coall and Hertwig, 2010). Ample research in anthropology and sociology describes the subtlety and complexity of gift exchange. Bourdieu (Reference Bourdieu1977) sets out how gift exchange, like social exchange, is enacted over time. This temporality would be manifest in the exchange of childcare which grandparents provide in return for elder care by parents and grandchildren.
Reciprocity forms part of the definition of functional solidarity, one of Bengtson and Robert's (Reference Bengtson and Roberts1991) six components of intergenerational solidarity between adult children and their parents. Economists view this reciprocity as an intertemporal exchange in which parents invest in the expectation of future returns (Coall and Hertwig, 2010). Both economists and sociologists envisage alternatives to this kind of exchange thesis: for economists, grandparents could invest altruistically because their grandchildren's utility enhances their own utility (Becker and Tomes, Reference Becker and Tomes1986). For sociologists, this is a form of unconditional solidarity.
The gendered nature of ‘gift’ exchanges that grandmothers provide to grandchildren had received little attention until Ashwin et al. (Reference Ashwin, Tartakovskaya, Ilyina and Lytkina2013) argued that care contributions provided by women were so taken for granted that they did not demand reciprocation. The provision of care was not necessarily of unconditional solidarity or even altriusm. Grandmothers’ complaints about non-reciprocation from their daughters cast doubt on both altruism and reciprocation theories. Utrata (Reference Utrata2011) argued that sometimes grandmothers feel compelled to provide informal childcare as part of performing the necessary roles of women in their age group, which require prioritising the interests of the younger generation. If this pressure does come to bear on older women we would expect a high proportion of grandmothers to provide care because it is normative.
The intergenerational decision-making process would be related to the opportunities and claims to resources of three generations: children, parents and grandparents (Hagestad, Reference Hagestad2006). Testing this theory in practice requires empirical analyses which include characteristics of the three generations of grandparents, parents, and grandchildren, although the requisite data is rarely available in quantitative studies. However, pursuing this multi-generational approach, grandparents may be more likely to look after their grandchildren if the mother has a higher level of education. The intergenerational improvement in women's educational attainment means that in many cases mothers are likely to have higher attainment than grandparents. This leads us to the third hypothesis:
H3: Mothers’ work opportunities, as demonstrated by educational attainment, condition whether grandparents provide care.
Grandparents’ involvement also depends on family structure (Steinbach, Reference Steinbach2012), with grandparents more involved when the mother does not have a partner. In the USA the prevalence of grandparents’ childcare for never married, divorced and separated mothers was found to be around seven percentage points higher than in married mother households, based on analysis of 2008 data from the Survey of Income and Program Participation (Laughlin, Reference Laughlin2013). Poorer families are less likely to be able to provide financial transfers and thus providing care may be a more feasible method of providing intergenerational support, as Gray (Reference Gray2005) argued in relation to the UK.
In the UK, as elsewhere, household structure is related to poverty rates (Brewer and Gregg, Reference Brewer and Gregg2002). The risk of poverty in lone-parent households is particularly high, at 35 per cent, and even higher, at 55 per cent (Bradshaw, Reference Bradshaw2010) if the lone parent is not working. The relatively lower level of attainment (Kiernan and Smith, Reference Kiernan and Smith2003; Kanji, Reference Kanji, Mayes and Thompson2012) and lesser labour force experience of women in the UK who start out as lone mothers presents formidable challenges to participating in paid work. Ho (Reference Ho2015) found that in the USA increases in lone mothers’ income led to a decrease in grandparent-provided care and an increased uptake of formal care. A similar effect would result from lowering the cost of formal childcare. Indeed previous policies to reduce childcare costs for lower income families and lone mothers were successful in the UK: Francesconi and Van der Klaauw (Reference Francesconi and Van der Klaauw2007) found that the childcare element in the Working Families’ Tax Credit played a very important role in raising lone mothers’ labour force participation. Similarly, Duncan et al. (Reference Duncan, Paull and Taylor2001) found the price of childcare had a strongly negative effect on the probability of choosing formal care for both single parents and married women. In contrast, Viitanen (Reference Viitanen2005) found that childcare subsidies had only a modest effect in the UK, suggesting that the relatively low childcare price elasticity of employment may be a result of a preference for care by relatives, as Wheelock and Jones (Reference Wheelock and Jones2002) also argued. An alternative explanation could relate to the shortage of childcare places. This leads to the fourth hypothesis:
H4: The impact of grandparents’ childcare on the probability of lone mothers entering paid work is larger than that for partnered mothers.
Method
Data
I used data from the Millennium Cohort Study (MCS), a nationally representative sample of babies born in the UK over a twelve-month period which started in 2000. The MCS is a stratified one-stage cluster sample with finite population sampling and varying probabilities of selection. The analysis used data from the third round of the survey (MCS3) which took place when the survey children entered primary school at around age five. The survey comprised 12,013 partnered mothers and 2,938 lone mothers at MCS3. Details on the sample and attrition are provided in Plewis (Reference Plewis2007). After listwise deletion of observations with any missing values on the variables of interest, the analysis used data from 14,429 mothers at MCS3. The use of mothers’ attitudes from MCS1 was responsible for most of the reduction in the sample size.
Variables
Dependent variables
The first dependent variable was the mother's paid work status, which was set at zero if she was not working and at one if she was in paid work.
The second dependent variable was the mother's usual weekly hours in paid work.
Key independent variables
Grandparent childcare
This variable for grandparents’ childcare was constructed from two MCS3 variables. The first relates to the situation where any of the grandparents look after the cohort member before or after school during a typical week as the first named source of care. The alternative sources of care were non-resident parents, the other parent, friends, other relatives, nursery or an after-school club. The second variable indicates if grandparents looked after the cohort child at the weekend. In both cases respondents were asked the question irrespective of whether the mother worked. The responses to these two variables were combined to make a dichotomous variable equal to one if the grandparents provided care. As a robustness check, the variable was also coded as equal to one in cases where grandparents looked after grandchildren on a weekday after school only. The results are comparable. The combined variable results are reported below.
Grandmother's age
In recognition of Hagestad's (Reference Hagestad2006) call to include characteristics of all three generations, I ran a separate regression (not reported because this variable was insignificant and required restricting the analysis to those with a living grandmother) with the cohort member's maternal grandmother's age in years in the grandparents’ childcare equation, as previous studies found that maternal grandmothers provided the bulk of grandparents’ childcare (Aassve et al., Reference Aassve, Arpino and Goisis2012; Kanji, Reference Kanji, Mayes and Thompson2012). Younger grandmothers may be more able to provide childcare than older ones.
Distance from grandparents
I constructed a continuous variable to measure distance from maternal grandparents as measured by travel time to grandparents. In their study of proximity of middle-aged couples to their parents, based on the British Household Panel Survey, Chan and Ermisch (Reference Chan and Ermisch2015) found that having a child reduced the probability of living a long travel time from the woman's parents, but not from the man's parents. This finding suggests that the connection with the woman's parents is more critical in terms of having a grandparent who provides care. Compton and Pollak (Reference Compton and Pollak2014) argued that geographical distance to mother or distance to the mother-in-law had an effect on mothers’ labour force participation only through its effect on care. The variable is constructed as follows: zero equals grandparents are unavailable because they are no longer alive, have no relationship or are outside the UK; one equals grandparents living one hour or more away from parents but in the UK; two equals 30 to 60 minutes away; three equals 15 to 30 minutes away and four equals less than 15 minutes away. Chan and Ermisch (Reference Chan and Ermisch2015) dichotomised distance categories by whether the parents lived near (less than 15 minutes away) or further away, which seems unnecessarily restrictive in relation to whether grandparents can provide care.
Education
For educational attainment I used a five-category variable employing the UK's National Vocational Qualification (NVQ) scale. NVQ level zero covered those with no qualifications or qualifications obtained overseas (previous analysis suggested that the effects in this dataset are similar). NVQ1 encompassed qualifications of two GCSE passes at grades D-G, equating to twelve years of compulsory schooling. NVQ2 included those with two GCSE passes at grades A-C. NVQ3 equates to A-levels or equivalent vocational qualification, equating to 14 years of schooling. NVQ4 categorised those with tertiary education – that is, a degree, further degree, vocational degree, equivalent qualification in nursing or diploma in higher education.
Control variables
Child suffers if the mother works
Respondents were asked the question ‘A child is likely to suffer if his or her mother works before he/she starts school’ and asked to indicate (1) strongly agree, (2) agree, (3) neither agree nor disagree, (4) disagree, (5) strongly disagree and (6) can't say. The answers agree and strongly agree were coded one, the other answers were coded zero. This measure is not ideal because the question was asked at MCS1 but is included because mothers’ attitudes towards working have been found to be influential (Kanji, Reference Kanji2010). This question accounts for a part of what would otherwise be unobserved heterogeneity.
Urban-rural
Living in the major population centre of Greater London is an important predictor of employment (McKay, Reference McKay2004) but is unavailable so I constructed a UK-wide variable for living in a rural or urban area (the reference category).
Number of children
Studies have shown that the number of children impacts on women's participation. It may also impact on the extent to which grandparents provide care (Gray, Reference Gray2005).
Number of younger children
An additional variable is included for the number of younger children: it is coded zero if there are no younger children, one if there is one younger child and two if there are two or more younger children.
Analytic approach
The analysis comprised three parts. First, I used descriptive statistics to explore the normativity of grandparents’ childcare, indicated by its prevalence. Significance of differences between lone and partnered mothers was tested.
In the second step, I modelled participation using bivariate probit (with participation and grandparents’ childcare equations) and instrumental variables approaches. If unobserved characteristics determined both grandparents’ provision of care and mothers’ labour force participation, then univariate probit or logit estimation would result in significant bias. A positive correlation between the provision of grandparents’ childcare and mothers’ participation would result if a latent factor such as parental support facilitated a mother's participation and also the likelihood of grandparents’ childcare. A negative correlation could result if grandparents provided care to mothers with lesser educational attainment, and thus less access to formal care; who were less likely to be in work, as previous studies of the UK have shown (Kanji, Reference Kanji2010; Reference Kanji, Mayes and Thompson2012). A further explanation could be a mismatch in values regarding childcare between mothers who wanted to work and their parents who were less likely to provide childcare. The test for this endogeneity was if the error terms in equations specifying participation and grandparents’ childcare were correlated (Wooldridge, Reference Wooldridge2002).
The advantage of the bivariate probit model was that the coefficient of grandparents’ childcare in the participation equation was net of grandparents’ childcare selectivity effects. The causal effect of grandparents’ childcare on mothers’ participation is the difference between the probability of each mother working with access to grandparents’ care minus the probability of each mother working without access to grandparents’ care averaged across all mothers. By contrast, the IV estimator measures the effect of grandparents’ childcare on all those who received it (‘compliers’), the local average treatment effect (LATE) and the average treatment on the treated (ATT) (Angrist et al., Reference Angrist, Imbens and Rubin1996) which Arpino et al. (Reference Arpino, Pronzato and Tavares2014) show coincide in the case of grandparents’ childcare. The results of Angrist et al. (Reference Angrist, Imbens and Rubin1996) depend on the monotonicity requirement that there should be no defiers: that is, those who take the treatment of grandparents’ childcare when they are not assigned to the treatment group. In our case it is unlikely that, when grandparents live far away, mothers would still be able to take the grandparents’ childcare treatment; thus the monotonicity requirement is satisfied.
A potential source of bias is that heteroskedasticity of the jointly distributed error terms in the bivariate probit model could result in biased estimates of the ATE (average treatment effect) (Chiburis et al., Reference Chiburis, Das and Lokshin2012). The alternative is IV estimation, which is consistent for the local average treatment effect. As heteroskedasticity was found, both biprobit and IV estimation were reported and the results used to gauge the level of bias.
I employed distance from maternal parents (and in robustness checks, paternal parents) as an instrument because it had little bearing on mothers’ labour force participation but directly influenced grandparents’ provision of care, as found by Compton and Pollak (Reference Compton and Pollak2014). Indeed, the correlation between travel time from maternal grandparents and mothers’ labour force participation was only 0.07, while the correlation between giving care and travel time from maternal grandparents was 0.27. The data provided the unusual opportunity to verify whether parents had intentionally moved close to grandparents, which would risk invalidating the instrument. This was not the case as only 3 per cent of parents cited the desire to move close to relatives as one of their reasons for moving even though 49 per cent of children experienced at least one house move by their fifth birthday (Ketende et al., Reference Ketende, Macdonald and Joshi2010). Similarly Del Boca et al. (Reference Del Boca, Piazzalunga and Pronzato2017) reported that only 3.2 per cent of mothers had moved to be nearer their family or friends in the two years surrounding the birth of a child and 1.2 per cent had moved in with family, based on analysis of the British Household Panel Survey. Further confidence in the validity of the instrument comes from controls of educational attainment and mother's age which could be related to mothers living close to the maternal grandmother and also directly affect the outcome of mothers’ employment decisions (see Del Boca et al., Reference Del Boca, Piazzalunga and Pronzato2017).
The third stage in the analysis was to model the impact of grandparents’ childcare on mothers’ hours of work (the intrinsic margin). I used an instrumental variables approach to model the outcome accounting for the potential endogeneity of grandparents’ childcare, again using distance from grandchild, although in this case a binary variable (in effect a treatment variable) was constructed which distinguished those living close enough to use grandparent care. Although grandparents’ childcare is binary, Heckman and Robb (Reference Heckman and Robb1985) show that Ordinary List Squares (OLS) estimates are consistent in such cases.
The model is executed using Stata's ivregress command, which calculates mothers’ working hours and grandparents’ childcare equations simultaneously and provides corrected standard errors. We confined the analysis of working hours to working mothers because our interest was in whether mothers were enabled to work longer hours, conditional on their working. A Tobit model would take into account censoring on the dependent variable and in effect average across working and non-working mothers.
In order to account for differences between mothers in different educational groups and to distinguish between lone and partnered mothers, the average marginal effects of grandparents’ childcare (the average of the marginal effects calculated for each mother) on the participation of women with selected characteristics were reported.
Results
Descriptive analyses of mothers’ participation and grandparents’ childcare
The summary statistics in Table 1 showed that, by MCS3, grandparents were providing 36.1 per cent of the before- or after-school care during the school week or weekend care for lone mothers and 32.4 per cent for partnered mothers, regardless of whether the mother was working. An additional 2.9 per cent of children were looked after by a childminder who was named as the first source of before- or after-school care and the grandparent as the second source of out-of-school care. The proportions of grandparents’ childcare were higher for working partnered mothers at 38.8 per cent and 45.1 per cent for working lone mothers (not shown in the table). The very high levels of grandparents’ childcare for working and non-working mothers lend support to the idea put forward in H1 that grandparents’ childcare is normative in the UK.
Summary statistics for lone and partnered mothers
(weighted results; standard deviation in parentheses)
It was striking that a very high proportion of both lone and partnered mothers lived close to the maternal grandparents: 43.4 per cent of lone and 39.8 per cent of partnered mothers lived less than 15 minutes away from the maternal grandparents. These findings were consistent with Murphy et al.’s estimate (Reference Murphy, Grundy and Shelton1999) of 60 per cent of 30–50 year olds living within 30 minutes’ travel time of their mother. The data from the MCS showed that a lower proportion of paternal grandparents lived near to the cohort child than maternal grandparents, consistent with Chan and Ermisch's (Reference Chan and Ermisch2015) finding that there is greater proximity to maternal grandparents. Nonetheless paternal grandparents also lived in proximity, with over 50 per cent of couples who are parents living within an hour from the male partner's parents. In Table 1 the category outside the UK or not applicable applies to about 20 per cent of children: included in this category are 6.8 per cent of children who had a maternal grandmother living outside the UK; the rest of the category applies to those who report no relationship with their mothers or whose mother has died.
The survey data did not provide full details of which grandparent provided after-school care at MCS3. At MCS1 maternal grandfathers provided 8.2 per cent of the care in working lone-mother households, compared to 3.1 per cent for working partnered mothers; for lone mothers, other relatives and friends filled the part that male partners played in their children's care. The proportion of working lone mothers who used formal care when the child was around nine months old was actually slightly higher (34 per cent) than that of partnered mothers (32 per cent), bearing out Duncan et al.’s (Reference Duncan, Paull and Taylor2001) finding that similar proportions of lone and partnered mothers used formal care. Furthermore Ho's (Reference Ho2015) research on the USA showed that lone mothers did not choose grandparents’ childcare because they were averse to formal childcare.
Multivariate results
The value of rho showed a significant and high degree of negative correlation between the residuals of the participation and grandparents’ childcare equations, which implied that a variable that makes mothers more likely to participate in the labour market also made them less likely to receive grandparents’ childcare. Because we are using data with clustering – which means the observations are not independent – we cannot compute a true likelihood ratio, but instead compute the pseudo-likelihood (see Sribney, 2013). The modified Hosmer Lemeshow test of goodness of fit implied significant correlation in the error terms.
Notwithstanding its advantages, the bivariate probit model can be at significant risk of providing biased estimates if heteroskedasticity is present. I rejected the null hypothesis of no significant deviation from normality of the jointly distributed error term based on the Murphy score test (Murphy, Reference Murphy2007), chi2(9) = 65.98, Prob > chi2 = 0.0000; the result implied excess kurtosis and skewness, which could introduce significant bias into the parameter estimates (Chiburis et al., Reference Chiburis, Das and Lokshin2012). In order to correct for heteroskedasticity in the error terms, and to gain some insight into the degree of bias in the bivariate probit model estimates, I also employed an instrumental variables regression with robust standard errors. The endogeneity of grandparents’ childcare is, as in the biprobit model, established. The robust score test (see Wooldridge, Reference Wooldridge2002) result again led to strong rejection of the null hypothesis that grandparents’ childcare was exogeneous F (1, 14388) = 28.32 (p=0.000). The Stock Yogo (Staiger and Stock, Reference Staiger and Stock1997) test of weak instruments was strongly rejected, the F statistic (minimum eigenvalue) of 736 is substantially over the critical value of 10 suggested by Staiger and Stock (Reference Staiger and Stock1997) when there is one endogenous regressor.
Comparison of the marginal success probability in the participation equation of the bivariate probit model with the estimates of the IV model in Table 2 presents only minor variation in parameter estimates; although, as would be expected, there were substantial differences in standard errors. The limited difference in the parameter estimates provided reassurance that the bivariate probit estimates were not substantially biased and that the causal effect of grandparents’ childcare on mothers’ labour force participation, estimated at 11.9 percentage points through the bivariate probit model, was not substantially biased. A further robustness check involved running the grandparents’ childcare variable only, including before- and after-school care during the school week (excluding weekend care); the level of the effect was calculated at 10.66 per cent reinforcing that weekend grandparents’ childcare also impacts on mothers’ participation.
The effect of grandparents’ childcare on mothers’ participation: biprobit and instrumental variables estimates
*p < .05. **p < .01. ***p < .001. The column headed ME provides the marginal probabilities of mothers working. The average marginal effect of grandparent care on the probability that the mother is working – the conditional probability of working given grandparent care minus the conditional probability of working given no grandparent care, letting grandparent care = 0 and 1 for each observation, is calculated at 125 per cent. Analyses performed using svy command with Stata 12. Robust standard errors.
In the models reported in Table 2, as we would expect based on previous analyses, mothers of young children in the UK with higher levels of education were much more likely to be in work (Kanji, Reference Kanji2010). The participation of mothers is substantially higher in all groups relative to the reference category of no education and is particularly elevated for those who have had some kind of tertiary education. In comparative analysis, Kenjoh (Reference Kenjoh2005) found a particularly strong effect of education on the odds of working full-time in the UK, which is interpreted to be a result of the especially low level of state intervention to support lower income women into employment (see also Joshi et al., Reference Joshi, Macran and Dex1996). Mothers’ participation increases with age up to a certain point where it starts to diminish. Lone mothers are about twelve percentage points less likely to be in work controlling for age and educational attainment, narrowing the 20 percentage points raw difference between lone and partnered mothers, as the summary statistics in Table 1 indicated.
Perhaps counter-intuitively, but supporting Aassve et al.’s (Reference Aassve, Arpino and Goisis2012) findings for France, the results of the grandparents’ childcare equation in the biprobit specification in Table 2 indicated that mothers were more likely to be in receipt of grandparents’ childcare if they had attained NVQ3 over NVQ2 and less likely to receive grandparents’ childcare if they had attained degree level over NVQ2. Dimova and Wolff (Reference Dimova and Wolff2008) found that more highly educated mothers were more likely to be provided with grandmother care, although Arpino et al. (Reference Arpino, Pronzato and Tavares2014) found that less-educated mothers were more likely to receive grandparents’ childcare in Italy. Although we cannot directly compare the educational attainment of grandparent-mother ‘diads’ as in Dimova and Wolff (Reference Dimova and Wolff2008), it seems likely that mothers have better work prospects than the grandparents (largely grandmothers) because of the strong cohort effect on women's educational attainment as well as their age or youth advantage in the labour market (Utrata, Reference Utrata2011). In terms of H3 it is not a straightforward story of higher education leading to more grandparents’ childcare. Consistent with the findings of Dimova and Wolff (Reference Dimova and Wolff2011), I interpreted the educational gradient as meaning that intergenerational care was on average motivated by the desire to help mothers into work, which supports H1. There was no evidence that the age of the grandmother impacted on the provision of grandparents’ childcare (results not reported). In line with other studies, the presence of more children was associated with lower participation and also with a lower likelihood of grandparents providing care (tests for the endogeneity of additional children were rejected). Our results further echoed Aassve et al.’s (Reference Aassve, Arpino and Goisis2012) finding that the provision of grandparents’ childcare is not strongly associated with mothers’ characteristics, based on the seven countries of the Gender and Generations Survey.
An interesting side issue is the significance of grandparents living together. Grandfathers were the main care provider for over 8 per cent of lone mothers at MCS1 while the presence of both grandparents may facilitate one of them providing care. The significance of grandfather's presence is consistent with Glaser et al. (Reference Glaser, Ribé Montserrat, Waginger, Price, Stuchbury and Tinker2010) who show that older people with more resources, for example those with a partner or with higher levels of wealth or educational attainment, are more likely to provide help; those in poor health or single are less likely to provide support. Hank and Buber's (2009) study found that the presence of a partner for grandmothers was not significant in whether or not they provided care, so that if grandfathers were providing care then it was a direct provision of care rather than supporting grandmothers.
Most importantly these results show the substantial impact of grandparent-provided care in the UK. The instrumental variables regression estimates the LATE at 33.3 percentage points; in the bivariate model the effect of grandparents’ childcare on the marginal probability of working, given that grandparents’ childcare was available, was estimated at 31.5 percentage points. The unconditional average causal effect of grandparents’ childcare on raising mothers’ participation estimated by the biprobit model was 11.86 percentage points. Table 3 differentiates the strength of the average marginal effect at different values of the explanatory variables: lone and partnered mothers and different levels of educational attainment. The results show that education generally outweighs partnership status. Interestingly grandparents’ childcare raised the probability of mothers’ working most at NVQ3, equivalent to A-level qualifications. A high proportion, 22.8 per cent of lone mothers have no qualifications which implies a much lower probability of working. Grandparents’ care raised the participation of mothers without qualifications by 9.3 percentage points for partnered mothers and 8.0 percentage points for lone mothers. This effect should be interpreted against their much lower probabilities of working.
Biprobit average marginal effects for lone and partnered mothers as at different levels of educational attainment.
Instrumental variables regression of the working hours of mothers conditional on their working shows no impact of grandparents’ childcare on the working hours of those already in work (results available in Appendix 1 – online supplementary material). Tests of the instrument confirmed that the equation was not weakly identified.
Conclusion
The results of this analysis clearly show that grandparents in the UK have played an important role in mothers’ labour force participation. Grandparents’ childcare raised the participation of mothers, in the group that had access to grandparents’ childcare, by 33 percentage points (LATE), which is very much in line with the effect of 32 percentage points that Arpino et al. (Reference Arpino, Pronzato and Tavares2014) found for Italy and much higher than the 6 per cent effect that Compton and Pollak (Reference Compton and Pollak2014) found for the USA. Participation of mothers is much higher in the UK than in Italy underlining that participation is about both finding a job and having access to care. In the UK, the causal effect of grandparents’ childcare on the labour force participation of mothers with at least one child in the school entry age group (age four to five in the UK) was 12 percentage points (AME), which is substantial given the overall participation rate of 56 per cent at MCS3. The lower participation rate of mothers with one child of school entry age (versus mothers of children of all ages) illustrates how the ages of children affect mothers’ labour market outcomes.
Until now the grandparents’ role has been unquantified and largely peripheral to discussions about mothers’ participation in the UK. It is increasingly untenable to ignore the extent of the effect of grandparents’ childcare when the circumstances affecting the supply of that care are changing. Gray (Reference Gray2005) argued that the implications of providing grandparents’ childcare have not figured in plans for achieving increased participation in paid employment for older people, an issue that is pressing as deferment of the state pension is enforced under the Pensions Acts of 2011 and 2015. Grandmothers who provide care may already be in a precarious financial position due to an intermittent paid work history caused by prior caring.
Not enough is known about the opportunities for caregiving grandparents to continue in paid work, and the potential conflicts between providing unpaid care and trying to participate in paid work.
The analysis underscored that grandparents’ childcare plays an important role for mothers at all levels of qualification. While it raised the probability of participating by around 14 percentage points for mothers with NVQ4 (equivalent to degree level qualifications), it also made a substantial impact on mothers without qualifications raising their probability of participating by eight percentage points for lone mothers and nine percentage points for partnered mothers against much lower starting probabilities of participating. The larger size of the percentage point impact on participation at higher education groups is consistent with previous research such as Dimova and Wolff's (Reference Dimova and Wolff2011) study of SHARE. The importance of grandparents’ childcare was underscored for lone mothers for whom participation is more challenging, not least because of substantial differences in observable educational and age characteristics. The significant impact of grandparents’ childcare at all education levels is likely to mask differences in the circumstances in which that care is given; the provision of grandparents’ childcare for mothers without qualifications may present a more difficult challenge than that provided for mothers with higher level qualifications. We simply do not know enough about the type and extent of grandparents’ childcare and its implications for those in different socio-economic circumstances. Further research is clearly required.
The role that grandparents’ childcare plays in extending mothers’ hours at work has received little attention as Aassve et al. (Reference Aassve, Arpino and Goisis2012) and Dimova and Wolff (Reference Dimova and Wolff2011) have previously noted. Using the MCS data I found that grandparents’ care did not extend mothers’ hours for those already in work, which is consistent with Compton and Pollak's (Reference Compton and Pollak2014) analysis of the USA. Dimova and Wolff's (2012) analysis of pooled European countries found that grandparents’ childcare was associated with a higher degree of participation, but their ordered probit analysis also partly measured participation.
Policies which rely on transferring responsibilities for childcare to grandparents are likely to have had profound effects on grandparents, especially grandmothers, and on intergenerational relationships. This is a prime example of how the political structures the personal. There is conflicting evidence on whether parents use grandparents’ childcare because they prefer it, as Wheelock and Jones (Reference Wheelock and Jones2002) argue, or because they lack childcare alternatives. Further research is required to settle this important question. However, the evidence is clear that working parents in the UK rely heavily on grandparents and the degree of intergenerational solidarity is surprisingly high in what many regard as a liberal market economy.
Acknowledgments
The author is grateful to the anonymous reviewers for their insightful and constructive comments on an earlier version of this article.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S004727941700071X
