Class Inequality in Parental Childcare Time: Evidence from Synthetic Couples in the ATUS

Abstract

The time that parents spend teaching and playing with their young children has important consequences for later life achievement and attainment. Previous research suggests that there are significant class inequalities in how much time parents devote to this kind of developmental childcare in the United States. Yet, due in part to data limitations, prior research has not accounted for how class inequalities in family structure, assortative mating, and specialization between partners may exacerbate or ameliorate these gaps. We match parental respondents within the American Time Use Survey (ATUS) to generate synthetic parental dyads, which we use to estimate, in turn, the contributions of family structure, assortative mating, and specialization to class gaps in parental time spent in developmental care of children aged 0–6. We find some evidence that accounting for class differences in family structure widens income gaps in total parental time in developmental childcare of young children. Further, we show that assortative mating of parents widens educational gaps in developmental childcare, whereas specialization between partners marginally widens these class divides. Although the net effect of these three processes on income-based gaps in childcare time is modest, accounting for these three processes more than doubles education-based gaps in total parental developmental childcare as compared to maternal time alone. Our findings from this novel empirical approach provide a more holistic view of the extent and sources of inequality in parental time investments in young children’s cognitive and social development.

Early childhood is likely the most consequential period for the development of socio-emotional and cognitive skills, in large part because early skill development lays a necessary foundation for future skill acquisition (Heckman 2006). Because children are most receptive to environmental influences during early childhood, investments in children during this period play a crucial role in child development. Although it has proven very difficult to establish causal effects of parental time investment in childcare time on later life outcomes (e.g., Milkie et al. 2015), there is a large body of research that shows consistent positive associations. In particular, for young children, parental childcare time in developmentally enriching activities such as playing or talking with children, reading with children, or helping children with homework (hereafter “developmental childcare time”) is positively associated with cognitive outcomes (Fiorini and Keane 2014; Hsin and Felfe 2014; Del Bono et al. 2016; Cano et al. 2019; Li and Hamlin 2019), child behavior (Hsin and Felfe 2014; Gialamas et al. 2019), and academic performance (e.g., Price and Kalil 2018). However, this developmental childcare time is sharply stratified by parental socioeconomic class in the United States (e.g., Kalil et al. 2012; Hsin and Felfe 2014; Altintas 2016). Socioeconomic divides in developmental childcare time for young children are thus one mechanism among several (including, e.g., socioeconomic divides in exposure to environmental stressors; see McEwen and McEwen 2017) that may lead to inequality in child well-being and later life attainment.

Scholarship on parental developmental childcare time generally examines gaps in maternal or paternal childcare time, not in combined in maternal or paternal childcare time, not in combined parental childcare time (though see Sandberg and Hofferth, 2001; Folbre et al. 2005; Phillips 2011 for important exceptions using the 1997 Panel Study of Income Dynamics Child Development Supplement (PSID-CDS)). Yet, family structure is sharply stratified by income and education in the United States—children born to college-educated and higher-earning parents are much more likely to grow up in two-parent households than less advantaged children (McLanahan 2004). Class disparities in family structure are likely to lead to wider class gaps in total parental developmental childcare time than in maternal or paternal care alone, simply through the effect of the presence of an additional household parent. Further, the United States has high levels of marital homogamy with the respect to education (Mare 2016) and earnings (Schwartz 2010), both of which could have further effects on inequality in the family contexts of children (Schwartz 2013). The result then is that existing research that examines only one parent’s developmental childcare time may seriously underestimate differences in total developmental childcare time by education and income. It is also possible that the effects of family structure and assortative mating on class gaps in total developmental childcare received by household children are partially reduced to the extent that parents specialize in either paid work or childcare (Becker 1981).

Previous research has not directly estimated the separate effects of the above three mechanisms—class gaps in family structure, class-based assortative mating, and specialization between partners—on class gaps in parental developmental childcare time. The investigation of these effects on total parental developmental childcare time has been hampered by the fact that the highest-quality data on time use in the United States, the ATUS, only contains a time diary for one adult respondent per household. Our research fills this gap in the literature by successively estimating the contributions of family structure, assortative mating, and specialization between partners to education and income-based gaps in total parental developmental childcare time received by young children (defined here as children aged 0–6). Our general method is to create synthetic parent dyads by matching ATUS respondents to alter in the ATUS data, in order to estimate the developmental childcare time provided by ATUS respondents’ coresident partners (or non-coresident co-parent).

We find that accounting for family structure marginally widens gaps in parental developmental childcare time by household income, but not by maternal education. Educational gaps in developmental childcare are, however, significantly widened after further accounting for assortative mating and specialization between partners, though these two processes do not drive wider income-based gaps. In sum, accounting for all three processes—family structure, assortative mating, and specialization—serves to widen maternal education-based gaps in this parental childcare time by over twenty-seven minutes per day, or over 150% the size of education-based gaps in maternal childcare time alone. We also estimate that these processes widen income-based gaps in developmental childcare time by over 40% relative to income-based gaps in maternal childcare time. Given the importance of parental developmental childcare time with young children for the future well-being of children, our results suggest that total parental developmental childcare time is a mechanism through which class gaps in family structure, assortative mating, and specialization between partners translate into inequality in child outcomes.

Background

Class Gaps in Parental Childcare Time

Parental socioeconomic class strongly influences parental childcare time, with parents in higher-income households (Ramey and Ramey 2010; Schneider et al. 2018) and with higher levels of education (Ramey and Ramey 2010; Guryan et al. 2008; Kalil et al. 2012) performing more childcare. Although income and education are strongly correlated, scholars of the family have placed importance on both dimensions of socioeconomic class as distinct drivers of unequal parental investment in children. Research on parental behaviors associated with the concerted cultivation of children has shown that parental education is the greatest predictor of these behaviors (Cheadle and Amato 2011:697–8), and income, net of education, also appears to allow parents to increase their investments in their children (Kornrich and Furstenberg 2013; Schneider et al. 2018).

Further, these class gaps in maternal and paternal investments of time by class appear to have widened in recent years in the United States. Charting childcare time in the United States, Ramey and Ramey (2010) find that the gap between college-educated and less-educated mothers widened substantially after the mid-1990s. Schneider et al. (2018) also find evidence in the United States of widening divides between households in the top income decile and lower-income households in terms of childcare time, with fathers of high-income households in particular appearing to increase their childcare time relative to other fathers since 2000.¹ Though much research on parental childcare time focuses on a broader conceptualization of childcare than play and/or educational activities, and on childcare performed for all and not just young children, similar socioeconomic gaps in developmental childcare time can be seen among households with young children in the United States (Kalil et al. 2012; Hsin and Felfe 2014; Altintas 2016).

However, this existing research has been constrained both empirically and theoretically. In terms of empirics, the extant literature generally focuses on maternal or paternal childcare time separately and thus does not account for the total childcare time that children receive from both mothers and fathers. In terms of theory, although demographic perspectives have emphasized the role of class divergence in the familial settings in which childcare takes place, existing work has not attempted to account for how these transformations may have affected class gaps in parental childcare. Below, we discuss two crucial theoretical perspectives that will extend our understanding of the differentiation of childcare time by class: family structure and assortative mating.

Class Gaps in Family Structure

Theories of the second demographic transition (Lesthaeghe 2010) and of diverging destines (McLanahan 2004) would suggest that these well-documented class gaps in maternal childcare time might be exacerbated by class gaps in family structure. Children whose parents are highly educated (McLanahan 2004) or higher income (e.g., Martin 2006) are also more likely to grow up in a two-parent household, due to educational and income divides in non-marital fertility, marriage, and divorce (e.g., Martin 2006; Gibson-Davis and Rackin 2014).

Class gaps in family structure could affect class gaps in total parental childcare time through an intensive margin (i.e., married parents performing more childcare than unmarried parents) or through an extensive margin (i.e., the number of parents who are present and performing care). Although there is some evidence that individual mothers who are married spend more time in childcare than those who are unmarried (Bianchi et al. 2006; Kendig and Bianchi 2008), the differences are not large (Pepin et al. 2018; Bianchi 2011). However, it is likely that the effect on the extensive margin could have large effects on the total amount of parental childcare time received by the children of higher- and lower-SES parents. For example, non-coresidential fathers spend very little time performing childcare (Carlson and Berger 2013), whereas married fathers spent more than seven hours per week caring for children (Parker and Wang 2013). Indeed, Kalil et al. (2013) find large and significant differences in total caregiving time received by children in married parent households as compared to those growing up with a single mother.

Yet, research on parental childcare time has not directly investigated how class gaps in family structure are implicated in class gaps in developmental childcare time. An important exception is the work of Sandberg and Hofferth (2001:429), who examine the 1997 PSID-CDS and find that children whose mothers had a college degree spent three more hours each week in the care of either parent than children whose mothers did not have a college degree. They further report that this education gap was one hour wider among all family structures than among two-parent families, which suggests that a meaningful portion of the education gaps in parental time are due to education gaps in family structure. However, this research relies on data that is currently over two decades old, measures a different construct of childcare time (total time children were co-present with parents), and only reports a difference in means without controlling for other factors known to affect childcare time (e.g., family income, number of household children, and age of the youngest household child).

Assortative Mating

Beyond family structure, demographic theories of mate selection and homogamy raise the question of if class gaps in parental developmental childcare time might be driven in part by parents matching based on educational attainment or income, both of which are predictors of parental childcare time. Research finds that assortative mating has increased over the past several decades along lines of education (Mare 201). Similarly, economic homogamy has increased in the United States (Schwartz 2010), driven primarily by a rise in dual-earner households rather than increased matching based on earnings potential (Gonalons-Pons and Schwartz 2017). This increased class divide in the educational composition and earnings of married couples are likely to lead to increased class gaps in total parental developmental childcare time simply through additive effects, because highly educated and high-earning parents perform more developmental childcare. Further, the joint determination of childcare time between partners may also be affected by the degree of assortative mating between partners if homogamous couples have more similar time use preferences, and if these preferences are reinforcing such that there are multiplicative effects on parental childcare time.

Previous research has not isolated the contribution of the effects of assortative mating toward class gaps in parental childcare time, though some scholars have made indirect estimates. Altintas (2016) creates estimates of gaps in total parental childcare time by parental education, using the individual time diary data in the 1966–2012 American Heritage Time Use Survey. When taking into account education-based gaps in father absence and assuming full educational homogamy among married partners, Altintas (2016) finds much wider education-based gaps in total parental childcare time than in either maternal or paternal childcare time. Though useful, this analysis does not disentangle the effects of assortative mating on class gaps in total parental childcare time from the effects of class gaps in family structure. Further, the assumption of full educational homogamy is probably too strong, given that just over half of marriages are between educationally homogamous individuals (Schwartz and Mare 2005).

Specialization Between Partners

Finally, economic theories of the household (Becker 1981) suggest that total parental childcare time could be affected by within-household specialization between partners, whereby one partner spends relatively more time in paid work, whereas the other spend relatively more time in unpaid developmental childcare. Such specialization could come about by each partner spending more time in activities in which they have a comparative advantage (Becker 1981), or by higher-earning partners using their bargaining power to spend less time in household production (Gupta and Stratton 2008). This perspective proceeds from the general assumption in theory and research on specialization that childcare is fundamentally similar to other forms of unpaid labor, such as routine housework. But, there is also good reason to suspect that the household dynamics shaping time spent in these two forms of unpaid labor may in fact operate very differently, with parents of all social classes viewing intensive parenting as an investment good rather than a routine and unpleasant task (Ishizuka 2019).

Studies find mixed evidence of specialization between partners in childcare resulting from differences in earnings or time spent in paid work. Examining married mothers and fathers in the 2003–2007 ATUS, Raley et al. (2012) show that mothers’ employment and time spent in paid work are associated with fathers performing more childcare, both as measured in absolute terms as well as a ratio to the time mothers spend performing childcare. However, Connelly and Kimmel (2009) find no effect of the relative wage between spouses on either spouse’s allocation of time toward childcare, though in later work (Connelly and Kimmel 2010), they find that husbands’ weekly employment time is positively related to wives’ time in caregiving.

Any specialization between partners could reduce income-based gaps in total parental childcare time if, for example, a sufficiently large portion of higher-income households are characterized by high-earning men who provide relatively little childcare. On the other hand, if higher-income households are more likely to be characterized by dual earners with similar marginal productivities in work than are lower-income households, accounting for specialization between partners could actually reveal wider income-based gaps in parental childcare time. As with family structure and assortative mating, research has yet to examine the effect of specialization between partners in paid work and childcare on class gaps in childcare time.

Our Contribution

Previous literature that has examined class gaps in parental childcare time has generally relied on the large-scale and high-quality time diary data available in the ATUS (e.g., Bianchi et al. 2004; Guryan et al. 2008; Ramey and Ramey 2010; Kalil et al. 2012; Altintas 2016). However, estimating total parental childcare time using the ATUS—and how this time is shaped by family structure, assortative mating, and specialization between partners—is difficult because the ATUS only collects a time diary from one respondent per household. Though some estimates of class gaps in total parental childcare time have relied on the 1997 PSID-CDS (Folbre et al. 2005; Phillips 2011), these studies rely on data that is now over twenty years old and do not investigate how class gaps in family structure, class-based assortative mating, and specialization between partners may affect class gaps in total parental childcare time.

We advance the literature on inequalities in total parental childcare time by making novel use of the ATUS. We focus on parental developmental childcare time—including time spent in play and in educational activities such as reading and homework help—for children aged 0–6. By focusing on parental developmental childcare for younger children, we focus on the type of care that studies taking both panels (Fiorini and Keane 2014; Del Bono et al. 2016; Cano et al. 2019; Gialamas et al. 2019; Li and Hamlin 2019) and quasi-experimental approaches (Hsin and Felfe 2014; Price and Kalil 2018) have shown is most beneficial for child outcomes, performed at an age, where these investments are most consequential (Heckman 2006).

Our innovation is to exploit the detailed data in the ATUS on both the characteristics of diarists and the characteristics of their spouses or coresident partners (hereafter “partners”) to simulate couples and so calculate total parental developmental childcare time. In brief, we match diarists (described by a vector of characteristics R_i) who have partners (described by vector P_i) and who jointly share a vector of household-level characteristics (H_i) with alters R_j who have similar partner characteristics as P_i, who are partnered with individuals P_j with the similar characteristics as R_i, and also match on household-level characteristics. We then impute parental developmental childcare time by matched alters R_j as estimates of the time in childcare by P_i, which allows us to estimate total parental developmental childcare time by R_i and P_i. We perform a similar matching algorithm to match time diaries of single parents with time diaries of non-coresident parents of the opposite sex.

By using this matching technique, we create three sets of synthetic parental dyads, sequentially accounting for characteristics related to family structure, assortative mating, and specialization between partners on class gaps in total parental developmental childcare time. By comparing class gaps in maternal developmental childcare time to the estimated class gaps in total parental developmental childcare time in the first set of synthetic parental dyads, we are able to estimate the contribution of class gaps in family structure to class gaps in total parental developmental childcare time. Similarly, we estimate the effect of class-based assortative mating on class gaps in total parental developmental childcare time by comparing the class gaps in total parental developmental childcare time we observe in the second set of dyads to those in the first set of dyads. Finally, by comparing class gaps in total parental developmental childcare time between the second and third set of dyads, we estimate the effect of specialization between partners on class gaps in total parental developmental childcare time.

Data and Methods

American Time Use Survey

The ATUS is a nationally representative time-diary survey administered by the Bureau of Labor Statistics that is designed to provide estimates of time use in the United States. Because the ATUS is administered to a sample of respondents from the Current Population Survey (CPS), it contains detailed demographic and labor force information for both respondents and other members of the respondent’s household. We focus on ATUS respondents aged 18–65 who are living with one or more of their own children (defined by the ATUS as biological, step-, or adopted children) aged 0–6, or who have an own child aged 0–6 with whom they do not live. We further limit our sample to those who were not flagged as filling out a low-quality time diary, did not fill out their diary on a holiday, and did not have missing values for household income. Here, we drop approximately 12% of observations. As described in Appendix A, we re-weight our sample at this step to ensure that the resulting sample is demographically similar to the original sample.

We focus on ATUS respondents who filled out their final CPS survey after October 2003, when a more granular coding scheme for household income was introduced. The resulting sample consists of 16,769 mothers and 11,854 fathers surveyed between the years of 2003 and 2018. We access ATUS data through the American Time Use Survey Extract Builder (ATUS-X; Hofferth et al. 2020).

Parental Childcare Time

Although the ATUS includes detailed information on the characteristics of all household members, the ATUS only includes time diaries for the respondent. We measure parental developmental childcare time as the number of minutes in a diary day that a respondent reports playing with, reading to or with, talking with or listening to, and helping or teaching their own household children aged 0–6 (ATUS activity codes 030102–030107, 030201, and 030203). This definition of developmental childcare time incorporates the two types of direct engagement with children—play, and engagement in educational activities—which extant research highlights as most beneficial for the cognitive and non-cognitive outcomes for young children. Conversely, our measure does not capture time spent performing basic physical care of children, or childcare time where the parent is not directly engaging with children (including time spent managing children’s activities or in travel related to caring for children), both of which are less strongly associated with child outcomes.

For all parents in our sample living with their own children, we calculate this measure of childcare as that done in the presence of their own household child aged 0–6. In order to estimate total parental time received by children in single-parent households, we calculate a similar measure of developmental childcare time (ATUS activity codes 040102–040107, 040201, and 040203) performed by parents who have children with whom they do not live. We use the hierarchical time-diary extracts available on ATUS-X to identify developmental childcare time that is performed with own non-household children aged 0–6.

Tables 1 and 2 show that parental respondents in the ATUS with more education and higher family incomes spend more time in developmental childcare for their children aged 0–6 and are more likely to live with a coresident partner. Further, we find that non-coresidential fathers contribute relatively little developmental childcare time to their non-coresidential children aged 0–6.

Table 1.

Developmental Childcare Time and Partnership by Parental Gender, Education, and Coresidence with Children

	% Coresident with partner	Developmental childcare (min/day)
Mothers living with own children
High school degree or less	68.0	45.8
Some college, no Bachelor’s degree	72.9	47.8
Bachelor’s degree or more	92.7	61.0
Fathers living with own children
High school degree or less	92.6	32.4
Some college, no Bachelor’s degree	94.4	35.6
Bachelor’s degree or more	98.6	41.4
Fathers with own non-household children
High school degree or less		1.7
Some college, no Bachelor’s degree		11.6
Bachelor’s degree or more		17.9

Table 2.

Developmental Childcare Time and Partnership by Parental Gender and Family Income

	% Coresident with Partner	Developmental Childcare (min/day)
Mothers living with own children
0–25th income percentile	55.1	42.9
25th–75th income percentile	83.1	53.2
75th + income percentile	93.8	58.5
Fathers living with own children
0–25th income percentile	90.3	31.9
25th–75th income percentile	95.3	36.1
75th+ income percentile	98.2	39.8

ATUS Variables Used in Matching Procedures

In all matching procedures, we account for whether the time diary is filled out on a weekend or weekday, since there are significant differences in parental childcare time between weekends and weekdays (Hill et al. 2018). Below, we describe in turn the additional variables that we use to account for matching by family structure, assortative mating, and specialization.

Family Structure

We account for the contribution of class differences in family structure by first matching parental respondents by coresident status: parental respondents with coresident partners of a given sex are matched with alters of that sex who also have coresident partners,² whereas parental respondents without coresident partners are matched with opposite-sex respondents who have a child aged 0–6 with whom they do not live. This matching imputes childcare time for all coresident partners, whether they are a biological, step-, or adoptive parent of household children. For partnered respondents, we also match based on the number of children aged 0–6 living in their household and the age of the youngest child in their household. We perform this matching based on family structure in generating each of our three sets of synthetic parental dyads.

Assortative Mating

In our second and third sets of synthetic parental dyads, we augment the matching based on family structure to account for the additional contribution of assortative mating to class differences in parenting time. We do so by matching based on age, race (coded as White/other), and education. For parental respondents with coresident partners, this matching occurs for both the respondent and their partner: the respondent’s age, race, and education are matched with the age, race, and education of alter’s partner, whereas the respondent’s partner’s age, race, and education is matched with that of the alter. For respondents without coresident partners, we match their age, race, and education to that of other respondents who have children aged 0–6 with whom they do not live. Additionally, for respondents with coresident partners, we match based on household income.³

Specialization

Finally, in our third set of synthetic parental dyads, we additionally account for specialization between partners by also matching on usual hours worked per week of each partner and logged relative wages between partners. We do not expect within-household specialization to affect the total parental childcare time received by children who live with a parent not residing with a coresident partner. As such, we match these parents with non-coresident co-parents using the same variables as when matching based on assortative mating.

Matching on usual hours worked per week allows us to test whether time spent in work crowds out time for childcare (e.g., Bianchi et al. 2006, though see Hsin and Felfe 2014). Further, the amount of time that both partners spend in work is likely to partially reflect any jointly made decisions about how to allocate childcare time within the household. We match on relative wages between partners (defined, following Pollak (2005) and Connelly and Kimmel (2009), as the wage of an individual divided by the wage of their partner) because theories of specialization between partners and of bargaining power in relationships both predict that partners who earn relatively more money in paid work will spend less time in unpaid childcare. See Appendix B for a detailed description of how we estimate relative wages between partners.

Matching Procedure

In creating each of our three sets of synthetic parental dyads, we match respondents to alters in two steps. First, we use coarsened exact matching (Iacus et al. 2012) to place each respondent in a subset with alters who have similar values as the respondent on all variables. We begin by coarsening all n matching variables (X₁, …,X_n) into categorical variables (X₁^C, …,X_n^C), so that respondents with similar values for a given matching variable are grouped into the same category in the coarsened variable. Then, each respondent is placed into a subset of all respondents with alters who share the same values of all n coarsened variables; that is, for any i and j in the same matched subset, (X_1i ^C, …, X_ni ^C) = (X_1j ^C, …, X_nj ^C).

This procedure ensures that respondents are not matched to alters who do not share similar values on covariates that are important for explaining an outcome of interest, which may occur in matching procedures (like the propensity score or the Mahalanobis distance) in which the covariates used for matching are projected from their original n-dimensional space to a unidimensional space (Iacus et al. 2012). Importantly for this project, using coarsened exact matching allows us to ensure that, when we account for assortative mating in matching, each match aligns with the attainment of educational credentials and the categories of household income that we use to estimate class gaps in childcare time.

The choice of what cut-points are used for coarsening each variable—and how many cut-points are used—affects the types of matches that are made in coarsened exact matching. Some variables are categorical and do not need cut-points; for some continuous variables, cut-points can be chosen on substantive criteria (e.g., cutting years of education to split respondents into levels of educational attainment). For other continuous variables, there is a trade-off between the closeness of the match (which rises with more cut-points) and the percentage of respondents that are matched (which declines with more cut-points). In general, we used more cut-points for variables that were more strongly correlated with childcare time, and selected cut-points so that there were roughly an equal number of respondents in the “bins” defined by the cut-points. Appendix Table 1 describes which cut-points were used for all variables. We also test the robustness of the coarsening procedure by either (a) making small changes to the cut-points used for a variable or (b) adding an additional cut-point for a variable. We perform each of these robustness checks for five variables—age of youngest child, number of household children under the age of 7, respondent age, usual weekly hours worked, and relative wages between partners—for ten robustness checks in all. Our findings are robust to these permutations of the coarsening procedure (results available upon request).

Second, to find more precise matches within the subsets defined by coarsened matching, we calculate the Mahalanobis (1936) distance based on the uncoarsened matching variables between respondents in the same subset. The Mahalanobis distance between respondents i and j is as follows:

$$\begin{gathered} d_M(i,j)= \\ \sqrt{{\left(\left(X_{1i},\dots ,X_{ni}\right)-\left(X_{1j},\dots ,X_{nj}\right)\right)}^T{\Sigma }^{-1}\left(\left(X_{1i},\dots ,X_{ni}\right)-\left(X_{1j},\dots ,X_{nj}\right)\right)}, \end{gathered}$$

where Σ is the covariance matrix of Y in the entire group of potential matches. The inclusion of the covariance matrix in the Mahalanobis distance accounts for variables having different variances as well as for the covariance between variables. We match each respondent to the alter in their subset who is closest in terms of Mahalanobis distance. If there are multiple such alters, we randomly select one alter as the match. As described in more detail in Appendix C, we use multiple imputations to account for the variance in total parental childcare time that stems from this random selection when predicting class gaps in total parental childcare time.

After matching respondents to an alter, we then calculate the total parental developmental childcare time associated with each respondent, C^T, as the sum of their own developmental childcare time, C^R, and that of their matched alter, C^P. For parents with coresident partners, C^P represents developmental childcare time of their alter (a parent with coresident children) performed for their household children aged 0–6; for parents without coresident partners, C^P represents developmental childcare time of their alter (a parent with non-household children) performed for non-household children while in the presence of their own non-household child aged 0–6.

Adjusting childcare time for parental copresence

Our main measure of total parental developmental childcare time sums up developmental childcare time of a respondent and their matched alter and does not adjust for developmental childcare time that partners may perform together. To the extent that partners do perform developmental childcare together, simply summing up developmental childcare time of a respondent and their matched alter will overestimate the childcare time actually received by young household children. Though we cannot tell when the partner of an ATUS respondent is performing childcare at the same time as they are, we can calculate from ATUS-X the amount of time ATUS respondents perform in childcare when their partner is present. As a robustness check (described in further detail in Appendix D), we calculate measures of total parental childcare time that adjust for childcare performed in the presence of one’s partner and use these adjusted measures of childcare time to estimate the effects of family structure, assortative mating, and specialization between partners on class gaps in total developmental childcare time.

Analysis

We first compare class gaps in maternal childcare time among mothers who live with their children to class gaps in our estimates of total parental developmental childcare time from our three sets of synthetic parental dyads: one in which we match based on family structure, a second in which we match based on family structure and assortative mating, and a third in which we match based on family structure, assortative mating, and specialization between partners. Here, we measure gaps in mean developmental childcare time by levels of maternal education, which has been the standard in the literature, and thus focus on the subsample of respondents who are mothers (though results are substantively similar when considering all respondents). As such, our findings generalize to the population of households where mothers live with at least one own child aged 0–6. We use recommended survey weights provided by ATUS-X, additionally adjusting these weights to account for dropping observations with missing values of variables (described in more detail in Appendix A).

Next, again focusing on maternal respondents who live with their children, we more directly test for the effects of family structure, assortative mating, and specialization between partners on class gaps in total childcare received by children using a series of ordinary least squares regressions. Formally, for a given measure of total developmental childcare C_i^T, we estimate:

$$C_i{}^T={\beta }_0+{\beta }_{1}MatE{d}_i+{\beta }_{2}FamIn{c}_i+{\beta }_3{Z}_i+{\epsilon }_i.$$

Here, MatEd_i is a three-category measure of maternal education (a high school degree or less, some college but no Bachelor’s degree, and a Bachelor’s degree or higher) and FamInc_i is a three-category measure of family income (0–25th income percentiles, 25th–75th percentiles, and 75th–100th percentiles). We include a vector of control variables, Z_i, comprised of the age of the youngest household child, the number of household children aged 0–6, maternal age, whether the mother is White, whether the mother is employed, and whether the time diary was filled out on a weekend.⁴

To test the successive effects of family structure, assortative mating, and specialization on class gaps in total developmental childcare time, we want to test whether there is a significant difference between regression coefficients β₁ and β₂ when using different estimates of developmental childcare time as dependent variables. We do so using a flexible method (Clogg et al. 1995) that does not assume equality of error variances between models. Given coefficients β^T,1 and β^T,2 estimated from regressions on two measures of developmental childcare time C^T,1 and C^T,2, respectively, we calculate a Z-test for the difference between these coefficients as follows:

$$Z=\frac{{\beta }_i^{T,1}-{\beta }_i^{T,2}}{\sqrt{S{E}_{{\beta }_i^{T,1}}{}^2+S{E}_{{\beta }_i^{T,2}}{}^2}}.$$

We estimate Z-tests between regression coefficients to make three comparisons of measures of developmental childcare time: (1) between maternal developmental childcare time and estimated total developmental childcare time that accounts for family structure, (2) between estimated total developmental childcare time that accounts for family structure and estimated total developmental childcare time that accounts for family structure and assortative mating, and (3) between estimated total developmental childcare time that accounts for family structure and assortative mating, and estimated total developmental childcare time that accounts for family structure, assortative mating, and specialization between partners. These Z-tests estimate whether the respective contributions of family structure, assortative mating, and specialization between partners toward class gaps in developmental childcare time are statistically significant. We also estimate a fourth set of Z-tests that compare estimates of class gaps in maternal developmental childcare time to those in total developmental childcare time from our most detailed matching procedure to understand how the combined effects of family structure, assortative mating, and specialization between partners affect class gaps in developmental childcare time.

We will find evidence for increased class gaps in developmental childcare time if the coefficients corresponding to maternal education being a high school degree or less (as compared to the reference group of some college but no Bachelor’s degree) and/or family income being below the 25th percentile (as compared to the reference group of family income within the 25th–75th income percentiles) significantly decrease between models, or if the coefficients corresponding to maternal education being a Bachelor’s or higher and/or family income being above the 75th income percentile (as compared to the same reference groups as above) significantly increase between models.

Results

Comparisons of Class Gaps in Developmental Childcare Time

Figure 1 shows average estimated developmental childcare time for children aged 0–6 by maternal education across ATUS maternal respondents, for our various measures of childcare (as represented by different-colored bars). Comparing across the clusters of bars (i.e., across levels of parental education), we see that accounting for family structure (in orange) certainly increases total parental time in childcare across all levels of education relative to just measuring maternal childcare time (in green). Further, accounting for family structure also appears to lead to slightly wider educational gaps in childcare in absolute terms. Additionally accounting for assortative mating (as seen in the blue bars) more clearly reveals wider educational gaps in childcare time, as estimates of childcare time for households where mothers have less than a Bachelor’s degree fall, whereas estimates of childcare time for households where mothers have a Bachelor’s degree or higher rise. Finally, Additionally accounting for specialization (red bars) appears to slightly increase the magnitude of educational gaps in total parental developmental childcare for children aged 0–6. The combined effects of family structure, assortative mating, and specialization, then, are to increase gaps in total parental childcare time by education.

Figure 1.

Developmental childcare time for children aged 0–6 by maternal education.

Figure 2 shows these same figures where class is measured by family income rather than maternal education. Here, we see that compared to the income gradient in maternal developmental childcare time (green bars), the income gradient in total parental childcare time after adjusting for family structure appears steeper (orange bars). Additionally accounting for assortative mating (blue bars) reveals moderately wider income-based gaps in total parental childcare time than only accounting for family structure: estimates of total childcare performed by parents in households below the 25th income percentile fall, whereas estimates of total childcare performed in households above the 25th percentile rise. Finally, specialization (red bars) appears to only have a minor effect on these class gaps. As with maternal childcare, more fully accounting for family structure, assortative mating, and specialization appears to reveal wider income-based gaps in total parental childcare time.

Figure 2.

Developmental childcare time for children aged 0–6 by family income.

Regression Results

In Table 3, we present the results of Z-tests that compare the coefficients on education and income in the estimates of our four regression models, each of which predicts a different measure of developmental childcare: maternal childcare time, total parental time after matching on family structure, total parental time after matching on family structure and assortative mating, and total parental time after matching on family structure, assortative mating, and specialization between partners. These Z-tests more rigorously test the contributions of the mechanisms of family structure, assortative mating, and specialization between partners toward class gaps in developmental childcare time by accounting for potential confounders that may be correlated with developmental childcare time and these mechanisms.⁵

Table 3.

Comparison of Regression Coefficients for Different DVs of Developmental Childcare Time

Model Comparison	Model 2–Model 1	Model 3–Model 2	Model 4–Model 3	Model 4–Model 1
High school degree or less	−0.52	−4.20	−1.29	−6.00+
	(−0.15)	(−1.06)	(−0.34)	(−1.87)
Some college, no Bachelor’s degree	Ref.	Ref.	Ref.	Ref.
Bachelor’s degree or higher	3.27	10.29*	7.37+	20.93***
	(0.89)	(2.47)	(1.65)	(5.22)
0–25th income percentile	−6.57+	−2.01	0.44	−8.15*
	(−1.83)	(−0.51)	(0.11)	(−2.22)
25th–75th income percentile	Ref.	Ref.	Ref.	Ref.
75th + income percentile	0.77	0.94	−4.95	−3.23
	(0.19)	(0.21)	(−1.06)	(−0.76)

Note: Z-scores in parentheses. The dependent variables in Models 1–4 are (1) maternal childcare time, (2) total parental childcare time after accounting for family structure, (3) total parental childcare time after accounting for family structure and assortative mating, and (4) total parental childcare time after accounting for family structure, assortative mating, and specialization between partners.

⁺p < .1

^*p < .05

^**p < .01

^***p < .001.

We find some evidence that accounting for family structure (Column 1) widens income-based gaps in total parental developmental childcare time between low-income and middle-income families, as compared to maternal developmental childcare time alone. This effect is likely due to the gaps in the presence of coresident partners between mothers in low- and middle-income households, as seen in Table 2. Although this effect is only marginally significant, the magnitude of this estimated increase in developmental childcare time—about seven minutes per day—is large, given that Table 2 suggests mothers in middle-income households perform about eight more minutes of developmental childcare per day than mothers in low-income households. Surprisingly, though mothers of children aged 0–6 with a Bachelor’s degree or higher are also more likely to live with a coresident partner than are developmental childcare time and these mechanisms.

Turning to Column 2, we find that additional accounting for assortative mating leads to substantively and significantly wider gaps in parental developmental childcare time based on maternal education, due to an increase in parental developmental childcare received by children aged 0–6 whose mothers have at least a Bachelor’s degree. Assortative mating increases the gap in total parental childcare time received between children of college-educated mothers and children of high school-educated mothers by almost fifteen minutes per day. To put this gap in context, the effect of assortative mating is about as large as education-based gaps in maternal childcare time per day (see Table 1). In sum, the propensity of highly educated mothers to live with partners who also provide large amounts of developmental childcare appears to drive wide inequality in total developmental childcare time. Meanwhile, assortative mating does not appear to widen income-based gaps in parental developmental childcare time.

Finally, we see in Column 3 that accounting for economic specialization between partners widens gaps in parental developmental childcare time between households where mothers have some college but no Bachelor’s degree and households where mothers have at least a Bachelor’s degree. However, this effect of economic specialization on education-based gaps in childcare—about seven minutes per day—is only marginally significant. We also find that accounting for specialization between partners does not significantly affect income-based gaps in parental developmental childcare time.

Column 4 shows the sum effect of these three processes—family structure, assortative mating, and specialization between partners—on class gaps in parental developmental childcare time for parents of young children aged 0–6, as compared to maternal developmental childcare time. These processes combine to widen gaps in parental developmental childcare time of young children aged 0–6 based on maternal education by over twenty-seven minutes per day, compared to maternal developmental childcare time alone. Meanwhile, the gap between families in the lowest income quartile and families in the middle income quartiles in our measure of parental developmental childcare time that accounts for family structure, assortative mating, and specialization between partners is over eight minutes wider than the same gap in maternal childcare time, with this difference being statistically significant.

To better place these results in context, Figures 3 and 4 present predicted class gaps in parental developmental childcare time by maternal education and family income, respectively, for each of our four dependent variables: maternal time, parental time that accounts for family structure, parental time that additionally accounts for assortative mating, and parental time that additionally accounts for specialization between partners. In Figure 3, we see that, all else equal, mothers with a Bachelor’s degree spend about seventeen more minutes per day in developmental childcare for their children aged 0–6 than do mothers with no more than a high school education. However, after accounting for family structure, assortative mating, and specialization between partners, this gap more than doubles to over forty-four minutes per day, with most of the growth in this gap due to assortative mating. Figure 4 shows that mothers in high-income households perform about twelve more minutes per day in developmental childcare for their children aged 0–6 than mothers in low-income households. This gap widens to about seventeen minutes per day after accounting for family structure, assortative mating, and specialization between partners.

Figure 3. Predicted education gaps in developmental childcare time of children aged 0–6, by matching procedure.

Note: The dependent variables in Models 1–4 are (1) maternal childcare time, (2) total parental childcare time after accounting for family structure, (3) total parental childcare time after accounting for family structure and assortative mating, and (4) total parental childcare time after accounting for family structure, assortative mating, and specialization between partners.

Figure 4. Predicted income gaps in developmentala childcare time of children aged 0–6, by matching procedure.

Note: The dependent variables in Models 1–4 are (1) maternal childcare time, (2) total parental childcare time after accounting for family structure, (3) total parental childcare time after accounting for family structure and assortative mating, and (4) total parental childcare time after accounting for family structure, assortative mating, and specialization between partners.

We also run models that predict class gaps in total parental developmental childcare time adjusted to discount childcare performed in the presence of one’s partner. The results of these regression models, and the comparison of coefficients between these models, can be found in Appendix Tables D1 and D2, respectively. We find a similar effect of assortative mating on maternal education-based gaps in parental developmental childcare time toward young children aged 0–6, and that this accounts for most of the increased education-based gap in parental developmental childcare time as compared to maternal childcare time alone. However, after discounting childcare time where both partners are present, accounting for family structure no longer widens income-based gaps in parental developmental childcare time for children aged 0–6. We interpret this finding to say that young children in middle- and high-income households receive more childcare where both parents are present than do young children in low-income households. In addition, in these adjusted models, accounting for specialization between partners no longer widens education-based gaps in parental developmental childcare time for young children.

Discussion

In recent decades, stark class divides have emerged in family structure, in the joint educational attainment and earnings of mothers and fathers, and parental developmental childcare time for young children. These features of unequal childhoods are likely to have consequences for inequality in child development. Because of the importance of child development for children’s future attainment and well-being, inequality at home stands in the way of equal opportunity.

However, prior work has treated these features of familial inequality—family structure, assortative mating, and parental developmental childcare time—largely separately. We argue that estimates of class divides in maternal or paternal childcare time understate inequality in parental investment of time in children because family structure is sharply stratified by class and because marriages are homogamous along many dimensions. Further, childcare time in two-parent households may also be affected by the degree to which one parent specializes in childcare. Yet, existing research on class gaps in parental childcare time that has relied on the high-quality time diary data in the ATUS has not investigated these interconnections, in large part because only one household member fills out an ATUS time diary.

We overcome this limitation by using matching methods to generate synthetic parental dyads in the 2003–2018 ATUS and find that previous measures of class gaps in maternal or paternal developmental childcare time understate the true extent of class inequality in total parental developmental childcare time. Further, we find that different demographic mechanisms drive income-based gaps than drive education-based gaps in total parental childcare time. There is some evidence that class gaps in family structure exacerbate income-based gaps in parental developmental childcare time as opposed to maternal childcare time, though only when considering our measure of time that is unadjusted for joint childcare time. However, we find clear evidence that assortative mating significantly and substantially widens gaps in total developmental childcare time by maternal education, whether or not we adjust for joint time in developmental childcare. Finally, specialization between partners marginally widens education-based gaps in parental developmental childcare time, though this effect also vanishes when adjusting for joint childcare time.

The sum result of accounting for these three processes is to more than double education-based gaps in total parental childcare time as compared to gaps based on maternal childcare time alone (forty-four minutes/day vs. seventeen minutes/day) and to also widen income-based gaps in total parental developmental childcare time as compared to gaps in maternal childcare time (seventeen minutes/day vs. twelve minutes/day). Notably, the size of these estimated education-based gaps is substantially larger than what might be expected from simply summing the education-based gaps in maternal and paternal childcare time presented in Table 1 under the assumption of full educational homogamy, as has been done in previous studies (e.g., Altintas 2016). We view these findings as complementary to previous research that focuses on parental motivations or beliefs about child rearing as key drivers of parental childcare time (e.g., Lareau 2003). Here, we focus instead on family composition and show that inequalities in family composition explain a substantial portion of income- and education-based gaps in total parental developmental childcare time.

Although our study focuses on parental time, children may also receive valuable investments of time from other kin. If kin support is more common for lower-SES children or in the context of non-marital households (e.g., Stack 1974), our results may overestimate class gaps in total childcare time. Because the information on family interrelationships in the ATUS only connects parents and children, we can only examine developmental childcare of household children performed by grandparents living in the household, and no other kin. As seen in Appendix Table 4, low-income and less-educated parents are more likely to live with their young child’s grandparents, but few parents are in these living arrangements overall. Further, 89% of these grandparents perform no developmental childcare, with a mean of 10.2 minutes of developmental childcare performed by all grandparents. The result is that the effect of accounting for developmental childcare provided by coresidential grandparents on class gaps in developmental childcare received by young children aged 0–6 is negligible. However, our focus on childcare provided by household adults could still miss childcare that lower-SES children receive from nearby relatives, older household children, preschools, and care workers.

Additionally, we do not directly observe total parental developmental childcare time received by household children, instead estimating this total parental developmental childcare time using matching methods. But, research that relies on the child-centered time diaries in the PSID-CDS also finds wide class gaps in the amount of time that children spend in the care of parents (Sandberg and Hofferth 2001; Hsin and Felfe 2014). Because of the difference in activity codes between the ATUS and PSID-CDS, and because the PSID-CDS time diaries are child- rather than parent-centric, direct comparison between our estimates of class gaps in total parental child care and those estimated from the PSID-CDS are problematic. That said, our results are consistent with previous work that relies on the PSID-CDS, and additionally identify the contribution of class gaps in family structure, assortative mating, and specialization between partners toward class gaps in childcare time.

We further test the validity of our matching method by applying it to the 2014–15 UK Time Use Survey (UKTUS), which collects 24-hour time diaries from both partners in a household, and thus allows us to see whether our matching method introduces bias in estimating the correlation between model variables and total parental childcare time. Although differences between the ATUS and UKTUS make an apples-to-apples comparison impossible, we find that the relationships between model variables and our synthetic estimates of total parental childcare time is not significantly different than the relationships between model variables and actual total parental childcare time (see Appendix Table 5).

Finally, it is not possible to rule out that the class gaps we observe in parental time in developmental childcare are caused by highly educated and high-income parents over-reporting their childcare time in order to more closely approximate the ideals of intensive parenting. However, time diary data are acknowledged to have high reliability and validity (e.g., Gershuny et al. 2017). Indeed, research that finds social desirability bias among high-SES parents in parental responses about how often they read to their child measures this bias by comparing these responses to benchmark time-diary data on time spent reading to children (Hofferth 2006).

In sum, we document that class gaps in family structure widen income-based gaps in total parental childcare time received by household children, whereas the presence of class-based assortative mating and specialization between partners leads to wider education-based gaps in total parental childcare time. Accounting for class gaps in family structure and specialization between partners and class-based assortative mating thus reveals larger class gaps, whether measured by education or income, in total parental developmental childcare received by children than has been reported in previous literature that relies on individual time diaries. Given the strong association between this childcare time and child development, these results suggest that class gaps in family structure and specialization between partners, and the presence of assortative mating, may contribute to inequality in child outcomes.

The policy implication of these results is by no means clear. Programs implemented to date to encourage marriage have not succeeded (Wood et al. 2014). Instead, policies designed to improve parental job quality and augment household financial resources may be more effective in helping low-income parents clear normative bars to marriage (Edin and Kefalas 2005) and/or in fostering child development directly. Further, we show that is not just that higher income and more highly educated mothers are more likely to be married, but who they are married to and how work and childcare are negotiated within these marriages, that counts for class gaps in parental investment. Although private investments in children may then be difficult to equalize, progressive public investment could serve to narrow these gaps (Jackson and Schneider 2019).

About the Authors

Joe LaBriola is a PhD candidate in sociology at the University of California, Berkeley. His research examines the causes of population-level racial and socioeconomic inequalities in the United States. One strand of his research is geared toward understanding how White and affluent households maintain their socioeconomic position and transmit their advantages to future generations. A second strand of his research focuses on causes of inequality in the experience of precarious work.

Daniel Schneider is professor of public policy and of sociology at Harvard University and co-director of The Shift Project. His research focuses on inequality, the family, and precarious work. His recent research has appeared in the American Sociological Review, Demography, and other outlets.