Income Inequality and Class Divides in Parental Investments

Daniel Schneider UC Berkeley Department of Sociology

Orestes P. Hastings UC Berkeley Department of Sociology

Joe LaBriola UC Berkeley Department of Sociology∗

∗Daniel Schneider (Corresponding author): UC Berkeley, Department of Sociology, 480 Barrows Hall, Berkeley, CA 94720; [email protected]. The authors thank the Institute for Research on Labor and Employment (IRLE) at UC Berkeley and the Russell Sage Foundation for research support. Income Inequality and Class Divides in Parental Investments

Abstract

Historic increases in income inequality have coincided with widening class divides in parental investments of money and time in children. These widening class gaps are significant because parental investment is one pathway by which advantage is transmitted across generations. But, is the apparent association with income inequality coincident or causal? Using over three decades of micro-data from the Consumer Expenditure Survey and American Heritage Time Use Survey linked to state-year measures of income inequality, we test the effects of inequality on class gaps in parental investment. We find that rising income inequality increased class gaps in parental expenditures on investment in children, but did not affect class gaps in childcare time. We further explore mechanisms that may drive the relationship between rising income inequality and widening class gaps in parental expenditures on investment in children. This relationship is partially explained by the increasing concentration of income at the top of the income distribution in times of higher inequality, which gives higher-earning households more money to spend on parental investment. In addition, we find evidence that rising income inequality reshaped parental preferences towards investment in children differentially by class. Introduction The last forty years have witnessed historic increases in income inequality in the United States.

Over the same period, existing class divides - by household income and by parental educational attainment - in how much parents spend on children and in how much time parents spend with children have widened considerably. These increasingly evident class divides in parental investments of time and money spark concern because parental investment is seen as an important factor in the intergenerational perpetuation of advantage. If aﬄuent families are increasingly able to transmit their advantages to children, that bodes poorly for an open opportunity structure.

Scholars have often observed the correspondence in the timing of these two trends and have suggested that the rise in income inequality may be causally implicated in the growing class divide in parental investment. However, scholars have yet to actually test the empirical relationship between income inequality and class-gaps in parental investment or to investigate the pathways by which any such relationship might be brought about. We would expect rising income inequality to increase class gaps in parental expenditures on children mechanically if rising income inequality simply means that the aﬄuent have more to spend. But, income inequality might also widen class-gaps in investments in children if income inequality changed parental preferences and practices.

It is also possible that income inequality is not related to class-gaps in parental investment.

Indeed, recent work suggests a narrowing of gaps in early achievement by family income and a narrowing or arrested divergence in some gaps in parenting practices even as income inequality has continued to rise, raising questions about this often assumed empirical relationship.

We empirically investigate these questions. First, we merge thirty-five years worth of detailed micro-data from the Consumer Expenditure Survey on parental expenditures on children with area- level annual measures of income inequality from the IRS and Census. Using a set of regression models with state and year fixed effects, we show that the gaps by family income in expenditures on investment in children are wider in more unequal contexts. Any such inequality effects could derive from the concentration of income among high income households such that they had more to spend or from contextual effects of income inequality on parents’ decision-making about allocations of money. We develop and conduct a set of tests to adjudicate between these mechanisms and find that this gap is both the result of the mechanical concentration of income and changing parental preferences.

1 We then investigate the broader distributional implications of rising income inequality for class gaps in parental investment by using the American Heritage Time Use Survey to examine how income inequality shapes class gaps in parental time with children. Here we ﬁnd a null result – high income parents did not increase their investments of time with children with rising income inequality, but neither were the widening class gaps in expenditures oﬀset by narrowing class gaps in childcare time.

Background Parenting and Child Wellbeing Social scientists have long been concerned with how contextual factors shape economic and social attainment and mobility. While institutions of higher education, the labor market, and the criminal justice system powerfully bear on these processes, early life conditions and contexts also appear to be enormously consequential (i.e. Heckman, 2006). Children’s environments are a product of the neighborhoods they live in, the schools they attend, but also, crucially, the families in which they grow up (Duncan and Murnane, 2011; McLanahan, 2004).

A large body of research documents how parenting practices - time and money spent - have important eﬀects on child wellbeing and later life attainment. This literature ﬁnds that more involved parenting, including providing educational materials, enrolling students in activities, and spending time with children is positively related to child test scores and cognitive development

(Bodovski and Farkas, 2008; Greeman, Bodovski, and Reed, 2011; Del Boca et al., 2012; Carneiro and Rodriguez, 2009). These dynamics are evident in US data, and also in samples from Denmark

(Thomsen, 2015), Australia (Fiorini and Keane, 2014), and the UK (Del Bono et al., 2014). While much of this research is observational, Price (2010) and Villena-Roldan and Rios-Aguilar (2012) instrument for parental time and also ﬁnd positive eﬀects on child cognitive test scores.

Further, these eﬀects appear to be substantively important. For instance, decomposing income- related gaps in achievement scores at Kindergarten entry, Waldfogel and Washbrook (2011) ﬁnd that parenting style and home learning environment are together more important than maternal education in explaining income-related gaps in scores on language, math, and literacy assessments.

2 Class Divides in Parental Investment Existing research also documents stark class diﬀerences in parenting practices. Examining parental investments of money and time along the axes of education and income shows clear stratiﬁcation.

There are substantial diﬀerences in parental expenditures on children by parental income group

(Kornrich and Furstenberg, 2013; Kaushal, Magnusen, and Waldfogel, 2011). For example, Kornrich and Furstenberg (2013) show that parents in the top decile of earners spent ﬁve times what parents at the median household income spent on children in the period 2006-2007 - $11,000 to $2,220.

That higher income households spend more on children is not surprising. But, notably, these class-gaps in parental investment appear to be widening over time. Using CEX over forty years,

Kornrich and Furstenberg (2013) ﬁnd that the gap in expenditures on children under age 25 between parents in the top 20% of families by income and those making less widened in the 1990s and 2000s.

This widening gap appears to be the product of increases in spending by households in the top two income deciles. Some of that increase may be driven by rising college costs. But, in recent work,

Kornrich (2016) ﬁnds a similar widening gap in expenditures for households with children under age 6 over the period 1980-2010, driven by the top 10% of households by income.

Parental investments of time in children are also strongly patterned by socio-economic status

(Bianchi et al., 2004; Phillips, 2011) with more educated parents and higher income parents (Guryan,

Hurst, and Kearny, 2008) spending more time in childcare and with more educated parents more eﬀectively targeting age-appropriate developmental care to children (Kalil et al., 2012).

There also appear to be widening gaps by class in parental investments of time. Charting childcare time, Ramey and Ramey (2010) ﬁnd that the gap between college-educated and less educated mothers widened substantially after the mid-1990s. These increases appear most pronounced for time in teaching and activities (Ramey and Ramey, 2010) and for time spent by mothers with children under the age of 5 (Hurst, 2010; Sacks and Stevenson, 2010). Extending the time series,

Altintas (2016) ﬁnds that the gap in time spent in developmental childcare between college-educated mothers of young children and mothers who had only a high school degree or less grew signiﬁcantly from the mid-1970s through 2013, with wide gaps emerging in the recent period 2003-2013.

However, the trends in class-gaps in parental investment in children are not without nuance. Re- search harmonizing multiple data sets (including NLSY-CS, PSID-CDS, and ECLS) ﬁnds widening class-gaps in formal childcare and book ownership (Bassok et al, 2016), as well as in daily reading,

3 frequent teaching of letters, words, or numbers, frequent story telling, visits to the zoo, a museum, and in going to a play or concert (Kalil et al, 2016), but narrowing gaps in computer use, learning activities at home, and out-of-home activities (Bassok et al, 2016) and library visits (Kalil et al,

2016), with some suggestion that class-gaps stopped increasing by 2005 or 2007 (Kalil et al, 2016).

A Backdrop of Rising Income Inequality It is diﬃcult to overlook the fact that these generally widening class gaps in parental investment have played out over exactly the same decades during which income inequality has increased dramatically.

Figure 1 presents the time trend in national income inequality from 1975-2013 with the top 10% income share on the left and the Gini index on the right. There is a large increase in inequality between the mid-1970s and the present, with the top 10% share rising from about 33% to 48%.

Indeed, scholars have noted this correspondence and alluded to the possibility of a causal relationship between income inequality and widening class gaps in parental investment (Kaushal,

Magnuson, and Waldfogel, 2011; Duncan and Murnane, 2011; Kornrich and Furstenberg, 2013;

Kalil, 2014). Rising income inequality could cause wider gaps in parental investments of money and time by both increasing the amount of disposable income available to high income households to spend and/or by changing high or low SES parents’ decision making about their allocations of money and time.

And yet, even as income inequality has continued to rise through to the present, there is some evidence that class-gaps in certain parenting behaviors have stopped widening (Kalil et al, 2016;

Bassok et al, 2016) and recent work suggests a narrowing of the class gap in math and reading scores at Kindergarten entry between 1998 and 2010 (Reardon, 2011; Reardon and Portilla, 2016).

We cannot then so easily assume that rising income inequality is responsible for widening class- gaps in parental investment in children. Yet, no research has actually examined the empirical relationship between income inequality and class-gaps in parental investment. Further, research on the eﬀects of aggregate income inequality on health and wellbeing should lead us to be cautious about connecting inequality and class-divides in parenting. While there has been a great deal of research showing that income inequality is associated with worse social outcomes (i.e. Wilkinson and Pick- ett, 2009), other scholars caution that this work may suﬀer from problems of omitted variables bias and have suggested more robust analysis strategies that rely less on cross-national cross-sectional

4 comparison and more on within-state over-time changes in income inequality (Kenworthy and Mc-

Call, 2008; Evans, Hout, and Mayer, 2004; Deaton and Lubotsky 2003), with particular attention to how income inequality may widen class-gaps or disparities in important behaviors (Truesdale and

Jencks, 2016; Neckerman and Torche, 2007).

This research notes the possibility that income inequality could be connected to class divides in parenting. However, we are aware of only one study that has attempted a more rigorous identiﬁca- tion of the contextual eﬀects of income inequality on class divides in human capital development, though this study focuses on educational attainment rather than parenting. Mayer (2001) uses data from PSID merged with state-level data on income inequality to show that income inequality widens the class gap in attainment, by both increasing the attainment of high-income children and reducing that of low-income children.

Mechanisms Connecting Income Inequality and Parental Investment Evidence of a relationship between income inequality and class-gaps in parental investment would naturally raise the question of what mechanisms might underlie such a relationship. In the broader literature on inequality eﬀects, scholars have outlined a set of likely mechanisms by which income inequality might aﬀect social outcomes (Evans et al., 2004). Two general classes of these mechanisms could explain a relationship between income inequality and the class gap in parental investment.

Rising Income Eﬀect One pathway by which inequality could shape class divides in parental investment is through rising income for the highest SES families. Rising inequality serves to concentrate income among the highest earners, which means they simply have more to spend generally and on children (Furstenberg and Kornich, 2013). Prior to about 1980, economic growth led to substantial and fairly evenly distributed income gains across the income distribution. However, since 1980, income growth has been concentrated among households at the top of the distribution. Rising income inequality has meant that those at the top income ranks have more in absolute dollars to spend. This pathway is akin to what Evans, Hout, and Mayer (2004) term the mechanical consequence of income inequality.

Higher income rank households might then increase spending across the board, both on investments in children, but also on consumption goods and on services. Could this aspect of rising income

5 inequality explain rising class gaps in parental expenditures on children? While Mayer (2001) finds little evidence that higher household income explains the relationship between state-level income inequality and educational attainment, Kornrich (2016) decomposes the increase in expenditures on children over the period 1980-2010 within the top income decile and shows that a third of the increase is due to higher incomes within the decile. That this effect is in some ways mechanical is not to discount it. If rising income inequality means that affluent households have still more money to spend and spend it in ways that widen the gap in parental expenditures, that is notable and important.

Concerted Cultivation and Preferences A second pathway by which income inequality might shape class gaps in parental expenditures on investment in children is through changes in attitudes, beliefs, expectations, or culture. Here, income inequality shapes class gaps not because of the mechanical concentration of income, but rather through “contextual eﬀects” on how parents think about the world (Evans et al., 2004;

Neckerman and Torche, 2007). There are good reasons to suspect that rising income inequality may have reshaped parental calculus and that this process may have played out diﬀerently by class.

Increasing returns to education with rising income inequality combined with increasing compe- tition for college entry could induce parents to invest more of their available resources in children

(Ramey and Ramey, 2010). This reaction could be more pronounced among higher SES parents if the returns to their investment of time and money are larger (Guryan et al., 2008), which could be the case given empirical literature that shows the strongest connections between parental time and child outcomes for parents with higher verbal skills (Hsin, 2009) or college education (Villena-Roldan and Rios-Aguillar, 2012). Additionally, parental investment by high-SES parents might be more re- sponsive to inequality if high-SES parents are more informed about the beneﬁts of such investment, which Kalil, Ryan, and Corey (2012) note is possible because the beneﬁts of such parenting have been widely touted in academic research and in the popular press.

Empirical evidence supports the idea that some parents are engaging in a style of parenting that prioritizes the careful creation of cultural and human capital in young children. Ethnographic research has shown that high-SES parents directly connect their intensive parenting with their desire to increase their children’s chances of attending a selective college and achieving occupational

6 success (Levey Friedman, 2013; Lareau, 2002). This set of behaviors, what Lareau (2002) calls

“concerted cultivation,” reﬂects changing parental preferences for child rearing.

This pathway is qualitatively distinct from the “rising income” explanation described above.

Rather than simply restructuring income such that high income rank households have more money to spend (in general and so also on children), here income inequality has reshaped preferences around parental investment in a class-biased manner. Both mechanisms can be understood as expressions of the effect of income inequality on class-gaps in parenting and both may lead to the same increasing stratification in parental investment in children, but the social processes are quite different.

Time and Money Tradeoﬀs? Whichever the mechanism, if higher SES-households increase spending on investment in children when income inequality is higher, they potentially act to more strongly transmit economic advantage across generations. In this way, a link between class gaps in investment and income inequality could provide a plausible pathway connecting income inequality and intergenerational mobility (Corak,

2013).

We have focused so far on how income inequality shapes class gaps in parental expenditures on investment in children. But, parents also invest in children by spending time with them (Kalil et. al., 2012; Ramey and Ramey, 2010; Guryan et al., 2008). The implications of any inequality eﬀects on gaps in expenditures on investment for intergenerational mobility could either be strengthened or weakened depending on how inequality also shaped gaps in parental time spent with children.

There is good reason to suspect that class gaps in time with children would widen with income inequality just as class gaps in expenditures on investment in children might widen with inequality.

First, parents may consider time spent with children as a normal good – that is, as household income increases, parents will desire to “consume” more time with their children. Though it is diﬃcult to cleanly identify whether increases in household income are associated with increases in childcare, the observed positive association between household income and time spent in childcare (Zick and

Bryant 1996) may indicate that parents do not view time spent with children as a good that is easily substitutable. Because increasing income inequality in the United States has had the eﬀect of increasing the incomes of households at the top of the income distribution, if time spent with children is a normal good, we should see greater class gaps in time spent with children in areas

7 with greater income inequality. This would imply that increases in income inequality would have an even larger eﬀect on class gaps in total investment in children than could be seen through parental expenditures on children. Furthermore, just as rising income inequality might change parental preferences and cultures to encourage greater expenditures on investment in children, so too might rising income inequality change higher-SES parents’ preferences for spending time with children if parents view such time as investment in children.

However, it is also certainly possible that class gaps in parental time with children might not widen with increasing income inequality. First, we might expect that parents who spend more money on their children would spend less time with their children if these parents are spending money in order to outsource time spent with children. In general, rising incomes at the top mean that higher income rank households are better able to hire lower-income workers to perform services. Jencks et. al. (1972) makes this point explicitly, remarking that a deﬁning feature of income inequality is that it enables the rich to hire the labor of the less aﬄuent. In recent work, Schneider and

Hastings (2016) ﬁnd that higher-SES women do less housework than lower-SES women and that the gap in housework time between them is wider in more unequal contexts, a phenomenon they attribute to inequality allowing high-SES women to outsource undesirable housework. For similar outsourcing to occur for childcare time, we would have to assume that parents prefer to substitute external childcare for their own time with children. This could occur if some parents prefer to avoid some aspects of childcare or if, given relatively limited time in children’s schedules, some parents substitute expert developmental care for their own care.

Second, class gaps in time with children might decline with income inequality if the driver of income inequality, and so of income gains for higher income rank households, is that parents are working more. There is good evidence that labor income makes up a large share of the income of top earners (Picketty and Saez, 2003), that married couples are increasingly homogenous with respect to socio-economic status (Mare and Schwartz, 2006), and that many high income rank households are composed of dual earners (PRC 2015) who may work long hours (Guryan et al., 2008; EPI,

2012). If income inequality is in part the result of higher income rank households working more, those households may spend more on investment in children, but may necessarily spend less time with their children themselves.

8 Hypotheses We ﬁrst more formally state our general hypothesis on the relationship between income inequality and gaps in parental expenditures on investment in children by household income rank. We then generate a set of hypotheses that allow us to adjudicate between the rising income eﬀect pathway and the concerted cultivation pathway. Finally, we introduce our hypothesis regarding how class gaps in parental time interact with income inequality.

Main Eﬀects We begin by examining if income inequality is related to class gaps in parental investment in children.

We focus on the relationship between parental expenditures on children and the interaction of parental income-rank and area-level income inequality. If the time trends in widening class-gaps in parental expenditures and rising income inequality are not simply coincident, then we would expect to see an empirical relationship between the two. Speciﬁcally:

Hypothesis #1: The gap in parental expenditures on investment in children is wider between high and lower income rank households in contexts of higher income inequality than in contexts of lower income inequality.

We next turn to the question of why income inequality might be related to class divides in parental investment. Here, we aim to adjudicate between the rising income effect that comes with higher income households having more income in higher inequality regimes and the contextual effect of changing preferences and cultures of concerted cultivation. We first describe hypotheses about the existence and strength of the rising income pathway, to be tested by: (1) controlling for household income in dollars and (2) examining expenditures on investment as a share of income. We then describe hypotheses designed to test the concerted cultivation pathway in two ways: (1) examining differences between investment and consumption goods and (2) examining stratification by parental education rather than income.

Rising Income Pathway If rising income for high income-rank households partially explain a relationship between area-level income inequality and widening class gaps in parental investments, then we would expect that simply controlling for household income would attenuate any signiﬁcant interaction of area-level income inequality and household income rank.

9 Hypothesis #2A: The gap in parental expenditures on investment in children is wider between high and lower income rank households in contexts of higher income inequality than in contexts of lower income inequality, but signiﬁcantly attenuates after controlling for household income in dollars.

Further, this attenuation should be most likely to occur in the relationship between income inequality and parental expenditures for the highest income households, since increased income inequality has resulted in higher incomes at the very top of the income distribution.

Hypothesis #2A identifies a prediction that results from rising income effects partially explaining the relationship between area-level income inequality and widening class gaps in parental investments. However, it is possible that rising income effects fully explain this relationship. That is, it is not so much high-income households’ position in the distribution of incomes that matters for their spending on children, but the sheer amount of income that they have. If true, we would expect that:

Hypothesis #2B: The gap in parental expenditures on investment in children is not wider between high and lower income rank households in contexts of higher income inequality than in contexts of lower income inequality after controlling for household income in dollars.

The implication of the rising income pathway is that higher-income rank households are simply spending more in general, on children included. A second test of this pathway then is to substitute the share of household income spent on children for the amount of household income spent on children. Here, if the rising income pathway fully explains the relationship between income inequality and parental investment, we would expect that:

Hypothesis #2C: The gap in parental expenditures on investment in children as a share of income is not wider between high and lower income rank households in more unequal contexts versus more equal contexts.

Testing the Concerted Cultivation Pathway The concerted cultivation pathway would lead us to expect that higher income parents are allocating resources speciﬁcally towards investment in children in higher inequality contexts, but not towards expenditures on other goods and services. We construct a conservative test of this proposition.

We expect that the interaction of household income rank and area-level income inequality predicts

10 parental expenditures on investment in children (Hypothesis #1). Here, we test if that interaction also predicts parental expenditures on consumption for children. More speciﬁcally, we contrast our key measure - parental expenditures on daycare, lessons, classes, and schooling - with parental expenditures on children’s clothing and furniture. If the concerted cultivation pathway matters, then we would expect:

Hypothesis #3: The gap in parental expenditures on children’s consumption goods is not wider between high and lower income rank households in contexts of higher income inequality than in contexts of lower income inequality.

So far, we have focused on the relationship between income inequality and parental household income rank. Though the literature on class gaps in parental expenditures on children almost always operationalizes class in terms of parental income,1 the concerted cultivation pathway would lead us to expect that parental investments might also be shaped by the interaction of income inequality and parental education. Indeed, Lareau (2002), defines class as a function of educational attainment and finds stark differences in adherence to the concerted cultivation versus natural growth models of parenting between parents of different educational (and occupational) levels. We test whether income inequality and parental education interact to predict parental expenditures on investment in children. If the concerted cultivation pathway matters, then we would expect:

Hypothesis #4: The gap in parental expenditures on investment in children is wider between households with more and less educated parents in contexts of higher income inequality than in contexts of lower income inequality.

Further, this relationship should be robust to controlling for household income:

Hypothesis #4A: The gap in parental expenditures on investment in children is wider between households with more and less educated parents in contexts of higher income inequality than in contexts of lower income inequality, controlling for household income.

And, if educational attainment and area-level income inequality really interact to shape parental preferences, then we should expect that

1This literature on class and parental investment does not engage with the classic sociological operationalizations of class as measured by occupational prestige, type, or SEI (e.g. Hodge et. al., 1964; Blau and Duncan, 1967; Wright, 1997). Rather, as described above, this research generally measures socio-economic status or class by either measuring parental income or education. As Hout (2008) notes, these variables are highly correlated with Americans’ own self-reported subjective class status and with other markers such as occupation and wealth.

11 Hypothesis #4B: The gap in parental expenditures on investment in children as a share of income is wider between households with more and less educated parents in more unequal contexts versus more equal contexts.

Testing Time and Money Tradeoﬀs Finally, we turn from these tests of the pathways by which income inequality might shape class gaps in parental expenditures on investment in children to examining how income inequality might shape class gaps in time spent with children as a way of understanding the broader distributional consequences of income inequality for investment in children.

Theory is ambiguous with respect to how income inequality might shape class gaps in parental time with children. There is good reason to suspect that these gaps would widen with income inequality, but also reason to expect that gaps might narrow if time and money are substitutes and if parents’ increased work eﬀort is the driver of high income rank in high income inequality contexts.

We propose that if these substitution or work eﬀort eﬀects dominate, then:

Hypothesis #5: The gap in parental time with children is narrower between high and lower income rank households in more unequal contexts versus more equal contexts.

Data and Analysis We conduct these tests by leveraging geographic and temporal variation in income inequality within the United States to estimate how class gaps in parental expenditures are affected by local contexts of inequality. Our overall approach is to merge area-level data on income inequality from the IRS and Census at the area-year level with data from the Consumer Expenditures Study (1980-2014) and the American Heritage Time Use Survey. We then use state and year fixed effects models to estimate if class gaps in parental investments are larger in contexts of higher aggregate income inequality. Below, we begin by describing the Consumer Expenditures Study data that we use to test most of the hypotheses and then more briefly describe the American Heritage Time Use Survey data that we use to test Hypothesis #5.

Data: Consumer Expenditures Study We use data from the Consumer Expenditure Study (CEX) from 1980-2014 to assess how class gaps in parental expenditures have changed with rising inequality. The CEX collects information on the expenditures, income, and characteristics of a nationally representative sample of households in

12 the U.S. We use data from the Interview surveys which are collected quarterly for each household

(technically, a consumer unit) for 12 consecutive months. We organize the data into a household- quarter structure (a household could be present between 1 and 4 times in the ﬁnal dataset). The start-month for a quarter is determined by the ﬁrst month of interview. Consequently, quarters can span more than one calendar year. In such cases, we assign the quarter to the year during which most of the quarter occurred (e.g., a quarter of Nov 2011/Dec 2011/Jan 2012 would be assigned to

2011, but a quarter of Dec 2011/Jan 2012/Feb 2012 would be assigned to 2012).2

State-level identiﬁers were not available for 1995 and this year is not included in our analysis.

State-level identifiers were also missing for first quarter of 1984, but we are able to identify the household’s state if they appeared in a later quarter in 1985. In addition, the CEX did not provide state identifiers on some respondents due to concerns that state identifiers could be used in con- junction with the other geographic characteristic variables (population of sampled area, urban vs rural, and census region) to identify geographic areas with populations less than 100,000 – violating standards mandated by the Census Disclosure Review Board. As a result we are missing state-level identifiers on about 15% of observations. We further limit our sample to households where at least one parent is over the age of 25 (to allow for normative age completion of schooling) and neither parent over the age of 65 (to exclude those who are retired). In all, we analyze expenditures for children from 221,959 household-quarters.

CEX Measures of Parental Investments We focus on three forms of parental investments: (1) Lessons - fees for recreational lessons and other instruction, (2) Schooling - student room and board; school meals; books, supplies, and equipment for school; tuition; and any other Pre-K through 12th grade school-related expenses, (3) Childcare

- all costs for babysitting, nannies, day care centers, and nursery schools.

Our preferred measure is the total sum of the three categories above. We divide the expenses per household by the number children in the household between the ages of 0 and 18, generating a per child expenditure measure. We exclude expenditures on behalf of those outside the home. To

2A Consumer Unit is “(1) all members of a particular household who are related by blood, marriage, adoption, or other legal arrangements; (2) a person living alone or sharing a household with others or living as a roomer in a private home or lodging house or in permanent living quarters in a hotel or motel, but who is ﬁnancially independent; or (3) two or more persons living together who use their income to make joint expenditure decisions. Financial independence is determined by the three major expense categories: Housing, food, and other living expenses.” Source: http://www.bls.gov/cex/csxgloss.htm

13 avoid unduly inﬂuential outliers, for each expenditure category we drop the top 1% of expenditures

(among those with any expenditures in that category). All expenditures and income measures are adjusted to 2014 real dollars using the CPI-U-RS series.

We also create a measure of parental expenditures on investments in children as a share of income that we use to test Hypothesis #2B. Here, we divide our preferred measure of expenditures on investments by the our measure of household income, described below.

Other expenditures recorded in the CEX are also targeted to children including spending on clothes, children’s furniture and equipment, toys, books, games, electronic equipment, travel, and sporting goods. These expenses are sometimes considered as “enrichment” (i.e., Kornrich, 2016;

Kaushal et al., 2011) and some, such as toys, books, and games, could reasonably be seen as an

“investment” in children. But, others, such as clothes and furniture, are less likely to be and, beyond a certain level of necessity, seem more like “consumption” by adults. Given these considerations, we do not include this class of expense items in our preferred measure. However, we do create a separate measure of parental expenditures on the “consumption goods” of children’s clothing and furniture and also examine the relationship between class divides in expenditures in these categories to test Hypothesis #3.

CEX Measures of Parental Income Rank and Education Household income is only measured in the CEX after the first and fourth quarters. After the second and third quarters, income is carried over from the first quarter, unless there have been changes to the household composition (e.g., a family member leaves the household), in which case the first quarter value is updated. In both the first and fourth quarters, income is asked about the previous

12 months. As such, the income reported after the ﬁnal quarter is a measure of the household’s income over the same period that the expenses occurred. So, we use the income reported in the fourth quarter. If households did not appear in the last quarter, we use the average income from the quarters that we observed.

We categorize households as: those in the 25th percentile of income and below, those in the

26th to 75th percentiles, those in the 76th to 90th percentiles, and those above the 90th percentile.

Prior work on income and expenditures on children shows a particularly large separation between the expenditures of the top income decile and the remaining groups (Kornrich and Furstenberg,

14 2013; Kornrich, 2016), so this measure is designed to capture similar dynamics.

Figure 2 shows trends over time in our measure of household expenditures on investment in children by household income rank. The gap in parental expenditures on investments in children between households in the top decile of income and households in the lower quartile of income has dramatically increased from less than $200 per quarter in 1980 to over $500 per quarter in 2014.

Notably, this increase is almost entirely due to an increase in investment spending by parents in top- decile households. This widening income gap in household expenditures on investment in children is consistent with expenditure gaps found in Kornrich and Furstenberg (2013) and Kornrich (2016), and occurred at the same time as the national increases in income inequality seen in Figure 1.

For Hypothesis #4, #4A, and #4B, we test how class gaps by parental education, rather than household income rank, may vary by contexts of income inequality. Our outcome, parental expenditures on investment in children, is a household level behavior. While income can be measured at the household-level as well, educational attainment is an individual-level attribute. Further, since some households contain two parents and some only one, measuring education is somewhat diﬃcult.

Our preferred measure takes the educational attainment of the most highly educated parent in the household. We code this measure as a BA or more, high school or some college, or less than high school. We also test the robustness of using measures coding (1) education of parent in household with lowest educational attainment and (2) number of parents with a college degree in the household

(coded as 0, 1, or 2). Results from these models, which are very similar to results from our preferred models, are included in the Appendix.

CEX Control Variables We also construct a set of controls: total household size, family structure, age of the oldest parent and age-squared, and race of each parent with flags for single parent households. We include the educational attainment of the most highly educated parent in the household as a control in models in which class is defined by income rank. In addition, we also create a measure of household income in thousands of dollars. We introduce this measure of income into the main model to assess if any effects of income inequality on expenditures operate through a purely rising income effects channel

(Hypothesis #2A). We also use this measure as a control in the model that interacts education and income inequality (Hypothesis #4A).

15 Data: American Heritage Time Use Study Second, we use data from the American Heritage Time Use Study (AHTUS) to estimate how class- divides in parental time spent with children are affected by state contexts of inequality. The AHTUS combines data from seven major time-diary studies conducted over the past seven decades and fielded in the years 1965, 1975, 1985, 1992-1993, 1994-1995, 1998-2001, and 2003-2014. We do not use data from either the 1965 survey, since it was fielded mostly in Michigan, or the 1985 survey, since it does not contain any geographic identifiers. We also drop the data from the 1992-1993 surveys since no data on household income was collected. Each of the other studies fielded from 1975 to

2001 assembles a nationally representative sample and then collects a single day comprehensive time-diary from a chosen adult. The 2003-2014 data is drawn from the American Time Use Survey

(ATUS) which provides nationally representative estimates of Americans’ time use using a detailed time diary methodology. The ATUS selects a sample of the respondents each year from the CPS and has each respondent use detailed time diaries to track their time use for a full day.

We combine data from these sources, as harmonized by the Centre for Time Use Research at

Oxford and accessed using the AHTUS Extract Builder (Fisher et. al., 2015). Dropping cases that are ﬂagged as “low quality time diaries” or have missing values of variables used in our model leaves an analysis sample of 19,947 mothers and 14,446 fathers between the ages of 25 and 65 surveyed in the years 1975, 1994-1995, 1998-2001, and 2003-2014.

AHTUS Measures of Parental Time We measure total maternal and paternal time with children aged 0-18. We construct this measure by summing total time spent on primary childcare by mothers or fathers, including time in basic care

(care of infants, general care of older children, medical care of children), play, teaching (supervising children, helping with homework, reading to/ talking with child), and management (other child care including coordination of activities and travel relating to child care).

Additionally, following Kalil et al. (2012), we also construct measures of time in age-appropriate childcare activities by age of the youngest child. We construct a measure of time spent in basic child care and a measure of time spent in play for respondents with a child under age 2 in the household.

We construct a measure of time spent on learning activities for respondents with a child age 3-5 in the household and a measure of time spent on management for those with a child age 6-13.

16 AHTUS Measures of Parental Income Rank The data on household income is more limited in the AHTUS than in the CEX. Because the AHTUS surveys collected income data in categories and used somewhat different categories over time, the harmonized measure of income divides respondents into income quartiles. While in the CEX we are able to isolate those above the 90th percentile by income, in the time-diary data we are limited to the 75th percentile and cannot include a control for dollars of household income (which is less important than in the expenditure models because rising income effects are less likely to play a central role for parental time). We check the robustness of the results by limiting the sample to the 2003-2014 American Time Use Surveys, which have more fine-grained income data, and re- estimating the models using the same percentile cut-points as in the CEX. We access ATUS data using the ATUS Extract Builder (Hofferth, Flood, and Sobek 2013).

Figure 3 shows trends over time in the AHTUS in maternal and paternal time spent in child care activities by household income rank. We average values across each time use survey used in our models, with the ATUS grouped here into three four-year time spans (2003-06, 2007-10, 2011-14) to smooth out time use trends. In 1975-76, there is no class gap in maternal time spent in child care; starting in 1994-95, mothers in households in the top income quartile report spending more time in child care activities than other mothers. But, there is no consistent gap in maternal time spent in child care between mothers in households in the bottom income quartile and mothers in households in the middle income quartiles. The gap in maternal child care between mothers in households in the top income quartile and other mothers appears to be at its largest in the late 1990s and early

2000s, and then slightly narrows in the late 2000s and early 2010s. Meanwhile, fathers in households in the top income quartile have had larger increases in their time spent in child care since 1975-76 than fathers in households in lower income quartiles. Until 2000, fathers in the top income quartile spent less time in child care than other fathers; since 2003, they have spent more time in child care than other fathers. Overall, there appear to be increases in gaps in parental child care between households in the top income quartile and other households, but these trends are nowhere near as pronounced as the increases in class gaps in parental expenditures on investment in children.

17 AHTUS Control Variables We also construct controls for education of the diarist (less than a high school degree, some college but no college degree, and college degree or higher), family structure (married or cohabiting versus other), number of children in the household under the age of 18, number of adults in the household, age of the respondent and age-squared, race of the diarist (white versus other), time spent working at main job outside of the home, and a ﬂag indicating whether the diary day is a weekend.

Data: Income Inequality Each set of micro-data is matched to year-speciﬁc area-level measures of income inequality based on the year of interview and the place in which the CEX or AHTUS respondent resides. Two important considerations guide the choice of measure: the level of aggregation and the metric.

There is real ambiguity in the literature on the eﬀects of income inequality with respect to what level of aggregation should be used to measure inequality. Most studies that seek to estimate the eﬀects of income inequality on social outcomes take the nation as the unit of aggregation and then generally make cross-sectional cross-national comparisons. However, there is little theoretical basis for using the nation as the unit of aggregation and there is a serious risk that other national-level variables confound any relationship between income inequality and outcomes of interest.

An alternative is to use state-level measures of income inequality. States are appealing because they represent a useful sub-national aggregate that may inform parents’ reference groups and purchasing markets. Compared to the nation, states provide more variation in the independent variable and allow for stronger causal inference. Additionally, unlike smaller geographic areas (such as counties or MSAs), the geographic boundaries of states are consistent over time.

While intuition may suggest that smaller geographic areas might better proxy for the reference groups that drive many contextual eﬀects of income inequality, in fact the empirical literature suggests that income inequality at the sub-state level is more weakly associated with social outcomes than at larger levels of aggregation (Wilkinson and Pickett, 2009).

Further, from a purely practical standpoint, there is no consistent sub-state annual time series data on income inequality. Such measures must be created by linearly interpolating from decennial census data for the years before the ACS. More to the point for this project, no sub-state geographic identiﬁers are available in the AHTUS or the CEX, either in public or restricted data.

18 In terms of metric, the Gini assesses inequality across the entire income distribution and is the most commonly used measure of inequality. However, we also use the top 10% income share.

Theoretically, it appears that much of the separation in parental spending over time has been driven by the top-end of earners, making it desirable to isolate this portion of the income distribution.

Thus, for both theoretical and practical reasons, we make use of state-level measures of the Gini and the top 10% income share. The inequality measures are constructed using data published in the IRS’s Statistics of Income. Thus, they are measures of the inequality of Adjusted Gross Income

(including wages, salaries, capital income, and entrepreneurial income) of tax ﬁlers. The Top 10% measure comes from the Frank-Sommeiller-Price Series (Frank et al 2015) and the Gini from the a separate series provided by Frank (2014). Both series are annual and extend up to 2013. In Figure

4, we plot the annual series of state top 10% income shares and the Gini from 1975-2014. There is considerable variation between states and over time.

As a robustness check, we also estimated our same models using the Top 1% income share,

Atkinson index, and Theil’s entropy index from the IRS data. Further, while a key strength of the IRS data is that it much more accurately accounts for top incomes, which are both top-coded out and susceptible to greater reporting error in the ACS, Census, and CPS, we also check the robustness of our results to using the Gini constructed by merging the results from Census and ACS and ﬁlling in missing years with linear interpolation. All of these models produced substantively similar results.

Area-Level Control Variables We also include a several time-varying state-level independent variables in our models: the unemployment rate, median income, percent black, and percent foreign born - all lagged one-year.

The state-level unemployment rate comes from the Bureau of Labor Statistics, and the remaining measures come from the Census and American Community Survey.

Analytical Methods We estimate linear regression models that examine the eﬀects of area-level income inequality on class divides in parental investments of money (and time) while accounting for other individual- and state-level covariates that may confound this relationship. Formally, consider person i living in

19 state s surveyed in year t. The individual-level regression equation is

Yist = β0 + β1Inequalitys,t−1 + β2Classist + β3Inequalitys,t−1 × Classist (1) +β4Z + γs + θt + ǫist.

In our baseline model, Yist is the measure of parental expenditures on investments in children,

Classist is a categorical measure of household income rank, and Inequalitys,t−1 is the lagged state- level top 10% income share or Gini coeﬃcient. Z is a vector of individual and state-level controls,

γs and θt specify full sets of indicators for the state and the year (i.e., two-way ﬁxed eﬀects), and

ǫist is the error. All models adjust the standard errors for clustering within states. In general terms, our analytic approach is very similar to that of Mayer (2001), Bloome (2015), and Kearney and

Levine (2014).

Given our interest in the class divide in parental investments, our focus is on β3, the coeﬃcients of the interaction term between income inequality and the parental income rank indicators. As inequality increases, we expect the class divide in parental investment to increase.

Equation (1) describes our empirical test of Hypothesis #1, that gaps in parental expenditures on investments in children will be wider between higher and lower income rank households in contexts of higher income inequality. In Table 1 we note how this equation (1) is modiﬁed to test each of the hypotheses designed to adjudicate between the two possible pathways - rising income eﬀect or concerted cultivation - that could explain why income inequality is related to wider class divides in parental investment.

Results We begin by showing the association between the household income rank gap in parental expenditures on investment in children and state-level income inequality. We then present the results from our models that more rigorously test Hypothesis #1 – that these gaps are wider in more unequal contexts. Third, we turn to our tests of the rising income eﬀect pathway, reporting the results of the models that test Hypotheses #2A, #2B, and #2C. Fourth, we report on the evidence for the concerted cultivation pathway, discussing the results of the models that test Hypotheses #3, #4,

#4A, and #4B. Finally, we discuss how income inequality shapes class gaps in parental time with

20 children and present the results of our tests of Hypothesis #5.

Figure 5 plots our measure of household expenditures on investment in children across observed levels of state income inequality, by household income rank. The left panel shows the Gini on the x-axis and the right panel shows the top 10% share on the x-axis. Each line plots the relationship between expenditures on investment and inequality for a diﬀerent household income rank group.

The blue line plots expenditures for households in bottom quartile of household incomes and the red line plots expenditures for the middle two quartiles of household income. In both plots and for both of those groups - for the bottom 75% of households by income - expenditures on investment in children are essentially ﬂat if not declining across increasing levels of income inequality. The green line plots expenditures for households between the 76th and 90th percentiles and the yellow line for those in the top decile of incomes. Expenditures are slightly upwardly sloping for those between the 76th and 90th percentiles with inequality. There is a more pronounced upward slope for the top decile of households by income. For these most aﬄuent parents, expenditures on investment in children rise with income inequality, whether measured by the Gini or the top 10% share.

Main Effects While these figures suggest a positive relationship between income inequality and the income gap in household expenditures on expenditures in children, these plots do not account for the demographic composition of households, for area-level confounders, or for time. Models 1 and 2 of Table 2 present the estimates from our first regression models (Equation 1). These models adjust for household characteristics, area-level confounders, and include state and year fixed effects. The estimates of key interest are the interaction terms between household income rank and state-level income inequality. The other coefficients from the models are reported in Appendix Table 1.

We see that relative to households in the middle two quartiles of income, those in top ten percent of households by income increase their spending on investment in children as income inequality rises. This coeﬃcient is large, positive, and signiﬁcant when we measure income inequality with the top 10% share (β = 1351.4, p < 0.001) and when we use the Gini (β = 1617.4, p < 0.001).

There is also a positive signiﬁcant coeﬃcient on the interaction between households in the 76th -

90th percentiles of household income and state-level income inequality, both when using the Gini

(β = 594.1, p < 0.001) and the top 10% share (β = 541.2, p < 0.001) to measure inequality. The

21 predicted eﬀect of increased inequality on the gap in parental expenditures between these households and the baseline group is notably smaller than for households in the top income decile. The class divide in investment with inequality extends downward as well. Households in the bottom quartile of income spend signiﬁcantly less relative to those in the middle quartile as inequality increases

(β = −420.4, p < 0.001; β = −339.7, p < 0.001).

Figure 6 plots the predicted expenditures on investment in children from Models 1 and 2 for each household income rank group across at the 5th percentile, median, and the 95th percentile of the level of the Gini (left panel) and the top 10% income share (right panel). These figures show that the effects are not just statistically significant, but are also substantively large. Whatever the level of income inequality, higher income rank households spend more than lower income rank households. For example, when the top 10% share is lowest (33% of income), those in the top decile spend a less than $400 per quarter on investment on average as compared to a bit more than $200 by those in the bottom three quartiles. At very high levels of income inequality, when the top 10% have 54% of income, parents in the top decile spend almost $700 as compared to, again, about $200 by those in the bottom three quartiles. In sum, the gap in expenditures on investment in children is about 150% larger in contexts of high income inequality as compared to contexts of low income inequality. This widening gap is not simply between the most affluent and the poorest (though there certainly is a widening gap between those groups). Rather, in contexts of higher income inequality, the highest income households diverge from the entire bottom 75% and, to some extent, even from those in the 76th - 90th percentiles of household income. In sum, the evidence for Hypothesis #1 is quite strong. Household income rank gaps in expenditures on investment in children are wider when income inequality is higher.

Rising Income Eﬀects This relationship between gaps in parental expenditures on investment in children and income inequality could come about through two possible pathways. Income inequality could concentrate income such that high income rank households have more to spend - what we have called the rising income pathway – or by changing parental preferences or cultures – what we have called the concerted cultivation pathway.

We now turn to our tests of the rising income pathway hypotheses. The four right-hand side

22 columns of Table 2 (models 3-6) present the first set of tests of this pathway. We first discuss how controlling for household income in dollars (Hypotheses #2A and #2B) affects our results and then how taking the share of income expended on investment in children as the left-hand side variable affects our results (Hypothesis #2C). We expect that if the relationship between income inequality and household income rank gaps in expenditures on investment in children is at least partially driven by the greater availability of income for those at top rank, then the key coefficients should significantly attenuate after controlling for household income. If the rising income pathway in fact fully explains this relationship, then the key coefficients should no longer be significant after controlling for household income or when using parental expenditures on investment in children as a share of income as the dependent variable.

Models 3 and 4 present the key coeﬃcients after controlling for household income in dollars.

We see the interaction of household income rank and state-level income inequality remains highly significant in both models, with the direction of the coefficient the same as in Models 1 and 2 for each household income rank. This suggests that controlling for household income does not fully explain the relationship between state-level income inequality and class gaps in parental expenditures, as proposed by Hypothesis #2B. For both models, the coefficients of this interaction are attenuated for high-income households - by just under 50% for households in the top income decile, and by roughly

30% for households between the 76th and 90th percentiles of income - but not for households in the lowest income quartile. That this attenuation only occurs for higher-income households makes sense, since the rising income pathway suggests that increased income inequality may increase the income gap in parental expenditures because higher-income households have more money to spend.

If the rising income pathway has a significant effect on the relationship between inequality and gaps in parental investment, then there should be a statistically significant difference in the magnitude of the coefficients corresponding to the interaction between income rank and inequality between

Models 1 and 3 (as well as between Models 2 and 4).3 We find that there is a statistically significant difference between the coefficients that correspond to the interaction between the top decile of

3We use a z-test (Clogg, Petkova, and Hatirou 1995; Paternoster et. al. 1998) to compare the signiﬁcance of the change in regression coeﬃcients between nested models:

β1 − β2 z = . 2 2 (2) p(SEβ1 ) + (SEβ2 )

23 income and state-level income inequality in Models 1 and 3 (p < .05), and between the same coef-

ficients in Models 2 and 4 (p < .05). However, controlling for household income in dollars does not significantly change the magnitude of the coefficients corresponding to the interactions between the

76th-90th percentiles of income and income inequality, and between the bottom income quartile and income inequality. We thus ﬁnd support for Hypothesis #2A: the rising income pathway partially explains the relationship between state-level income inequality and gaps in parental expenditures between households in the top income decile and households in the middle two income quartiles.

Models 5 and 6 present the tests of Hypothesis #2C, which predicts that, if all income inequality effects flow through the rising income pathway, then there would be no significant interaction between household income rank and income inequality in predicting the share of income spent on investment in children. Here, the logic is that if income inequality only widens class gaps by shifting more resources, and so enabling more spending of all kinds, to high income rank households, then we should not see any widening of expenditures as a proportion of income, which would reflect allocation decisions rather than simply more spending in general. We find no evidence for this proposition. In Models 5 and 6, the interaction between household income rank and state-level income inequality is significant for all income rank groups, and the coefficients have the same sign as in Models 1 and 2. In short, the gap in the proportion of income spent on investment in children, not just the gap in total amount spent, is wider in more unequal contexts.

Concerted Cultivation The main model results presented in Table 2 (models 1 and 2) show that household income rank gaps in expenditures on investment in children are wider in more unequal contexts. The additional results in Table 2 (models 3-6) show that rising income pathway partially explains the relationship between income inequality and parental expenditures by households at the top of the income distribution, but the relationship between state-level income inequality and class gaps in parental expenditures is robust to various ways of controlling for the rising income pathway.

In Table 3, we report the results of our tests of Hypothesis #3, that if concerted cultivation is the primary pathway connecting income inequality to widening gaps in parental investment, then we should not see any relationship between income inequality and income rank gaps in expenditures on children’s consumption goods. Models 1 and 2 of Table 3 test this proposition, substituting a

24 measure of expenditures on consumption for children for our measure of investment in children.

As predicted by Hypothesis #3, we do not see a significant interaction. In fact, the coefficient is negative and actually significant for households in the p76-p90 rank group as well as for households in the top decile of income. Rather than gaps in expenditures on consumption widening with rising income inequality, as we saw for gaps in expenditures on investment, gaps in consumption actually narrow with rising income inequality. These results show that income inequality specifically affects parental investment in children, rather than simply expenditures more generally.

Hypotheses #4, #4A, and #4B turn our attention from household income rank gaps in expenditures on investment in children to gaps by household educational attainment, as measured by the educational attainment of the more highly educated parent. These models allow us to test if educational gaps in expenditures also widen with income inequality. Such widening would be significant evidence for the concerted cultivation pathway, particularly if such widening appeared in models that also adjust for household income in dollars (Hypothesis #4A) or that take expenditures on investment as a share of income as the dependent variable (Hypothesis #4B) since both tests effectively separate a correlation between education and income that could be picking-up a rising income effect from the role of preferences that is in line with inequality driving concerted cultivation.

Table 4 presents six models that test these three hypotheses. In models 1 and 2, we see that the interaction between households in which the most highly educated parent has at least a bachelor’s degree (relative to the reference group of households where the most highly educated parent has high school but not a bachelors’ degree) and state-level income inequality is positively and signiﬁcantly related to parental investment in children, whether inequality is measured by the Gini

(β = 709.5, p < .001) or the top 10% income share (β = 593.7, p < .01). Further, the coeﬃcient on the interaction between education and income inequality is negative for households in which the most highly educated parent does not have a high school degree (β = −266.6, p < .001 when using the Gini; β = −240, p < .01 when using the top 10% income share).

Models 3 and 4 include a control for household income in dollars in an effort to further separate any effects along the pathway of rising income differences from those of concerted cultivation. After controlling for household income, there still remains a significant and positive relationship between state-level income inequality and wider class gaps (as measured by parental education) in parental investment in children. The coefficients on the interaction between parental education and state-

25 level income inequality do attenuate in Models 3 and 4 by roughly 30% for households in which the most highly educated parent has a bachelor’s degree and for households in which the most highly educated parent does not have a high school degree. However, the differences between the coefficients of the interaction in Models 1 and 3 and in Models 2 and 4 are not statistically significant.

Finally, Models 5 and 6 of Table 4 take expenditures on investment in children as a share of household income as the dependent variable. In these models, testing Hypothesis #4B, we test whether educational gaps in the allocation of expenditures towards investment in children are wider in more unequal contexts. This is in some ways the most rigorous test of the concerted cultivation pathway. Notably, we see that the interaction of having attained at least a bachelor’s degree with both the top 10% share (β = 0.026, p < 0.01) and the Gini (β = 0.026, p < 0.01) is positive and statistically signiﬁcant. Highly educated parents allocate a greater share of income towards expenditures on investment in children in more unequal contexts.

We also test the robustness of these six models to coding parental education by taking the education of the least educated parent in the household and by coding the number of parents with a college degree in the household. These results are presented in Appendix Tables 2 and 3.

Here we ﬁnd even more evidence of a widening gap in investment spending with income inequality.

The interaction term between inequality and the most educated parents is positive and at least marginally signiﬁcant (p <0.1) in all robustness models.

Time is Money? Hypothesis #5 turns our attention from class-gaps in expenditures on investment in children, to class-gaps in time spent with children. Time is an important form of parental investment in of itself.

If the wider income rank gaps in parental expenditures on investment in children are accompanied by wider income rank gaps in parental time with children, that could compound eﬀects on the transmission of advantage across generations. However, if class gaps in parental time with children narrow with rising income inequality, then the wider gaps in expenditures on investment might be of less concern.

Table 5 presents the results of our tests of Hypothesis #5 using the American Heritage Time

Use Surveys data. Models 1 and 2 present the coeﬃcients on the interaction of household income rank and our two measures of income inequality for maternal time and Models 3 and 4 do the same

26 for paternal time. We see no evidence of substitution effects. But, we also see little evidence that high income rank parents “double down” and increase investments of both time and money. The coefficients are non-significant and relatively small. There does not appear to be any meaningful variation in the class-gaps in parental time with children across differing levels of income inequality.

We also tested the robustness of these results to focusing on the American Time Use data

(2003-2014) which has more ﬁne-grained measures of income and so permits us to isolate the top decile of households by income. We examine our key interactions for both mothers and fathers and

ﬁnd no evidence of eﬀects on income inequality on the class gap in parental time with children.

Coefficients for the interaction between top decile households and state-level income inequality are not significantly different from zero, suggesting that income inequality does not induce parents in the highest-income households to spend significantly more (or less) time with children relative to middle-income households.

We also test disaggregating parental time into time spent on age-appropriate childcare activities

(as deﬁned by Kalil et al., 2012) by age of the youngest child. We substitute (1) time spent in basic child care and a measure of time spent in play for respondents with a child under age 2 in the household, (2) time spent on learning activities for respondents with a child age 3-5 in the household, and (3) time spent on management for those with a child age 6-13 for our main dependent variable and assess if household income rank interacts with income inequality to reduce any of these dimensions of childcare, by either mothers or fathers. We ﬁnd little evidence that income inequality is associated with increased class gaps in age-appropriate childcare activities.

Discussion Over the same decades that income inequality has increased dramatically in the United States, class gaps in parental investment in children have also widened. Observers have often attributed a causal role to income inequality in the widening of these gaps. However, it is essential to actually empirically test this assumption, so that we can assess if this apparent correlation is simply coincident or might be causal.

We drew on data from the Consumer Expenditures Survey and from IRS statistics on income inequality to show that in more unequal contexts, household income rank gaps in expenditures on investment in children are indeed signiﬁcantly and substantively wider. By exploiting over time

27 and place variation in income inequality within the United States, we isolate the role of income inequality in driving these class divides from time-invariant unobserved characteristics of places and from period eﬀects. We also control for a rich set of individual and area-level confounders. The gap in investment in children between the rich and rest is wider when income inequality is higher.

We next investigated the mechanisms that might explain this relationship between income inequality and class gaps in parental investment. Perhaps the most natural reason why these gaps would be larger in more unequal contexts versus more equitable contexts is that households of a given high income rank in unequal contexts will have more dollars available to spend than household of that same rank in more equitable contexts. Such eﬀects would be very “real,” even though there are essentially mechanical. Our tests of this “rising income pathway” show that higher income explains some, but by no means all of the relationship. The widening of the class gap with income inequality appears even after controlling for household income in dollars and we also see evidence of re-allocations of spending by higher income rank households in more unequal contexts.

We then tested several hypotheses relating to the “concerted cultivation” explanation - that in contexts of higher income inequality, the preferences or parenting cultures of higher-SES parents might change to prioritize investment in children. We ﬁnd that, in accord with this explanation, the gap in spending on investment in children is wider in more unequal places, but the gap in spending on children’s consumption is not. High income rank parents are not simply spending more in general, but appear to be targeting their expenditures in line with changes to parenting expectations and preferences. Further, these gaps in investment are also evident by parental education, controlling for income, which also accords with the “concerted cultivation” pathway.

Finally, we assess the broader distributional implications of this widening gap in expenditures on investment. If high income rank parents also increased their investments of time with children with rising income inequality, then the propagation of advantage would be increased. On the other hand, expenditures on children might effectively outsource some childcare time which could narrow the class gap in time spent with children, as could parents’ increased work hours in more unequal contexts. This kind of offsetting effect would imply that investment was simply reallocated from time to money so our results might be less troubling from a social equity standpoint. Empirically, we find no significant change in the class gap in parental time with rising income inequality. That is, even as high income rank parents increase their expenditures, they keep their time investments constant,

28 suggesting a net widening of class gaps in investment in children with rising income inequality.

Our work is subject to some important qualifications. First, even as we have homed in on the relationship between income inequality and class-gaps in parental investment by using state and year fixed-effects models with rich controls, we do not have a way of more cleanly identifying the effects of income inequality. This distributional parameter is not randomly assigned and arises from the complexity of economic, political, and social processes, factors that are difficult to measure well and could confound the relationships that we observe. However, scholars are yet to identify a convincing instrument for income inequality and it seems unwise to rule out any investigation of these questions on those grounds.

Second, while we observe household expenditures on investment in children and see that income rank gaps in spending rise with income inequality, we cannot observe what those investments actually yield. The assumption is that more money spent on lessons or daycare or school translates to some developmental beneﬁt. But, we are not able to observe that directly. It is possible that higher income rank households simply waste their money more in more unequal places, over-paying for child-oriented services of the same quality that they could pay less for in more equitable contexts.

Third, we draw on data from the AHTUS to assess whether class gaps in parental investments of time widen in line with those in money. We ﬁnd no evidence of widening gaps in time, but no evidence of narrowing gaps either. We suggest that the conclusion is that the net investment gap in children then widened with income inequality. However, the estimates for time and money are based on separate data sets and diﬀerent people. We do not, however, know of any data source with enough substantive, temporal, and geographic coverage to permit the joint evaluation of these two processes.

Our ﬁndings have implications for several areas of sociological research. First, scholars of strati-

ﬁcation have long examined processes of intergenerational mobility and trends and levels in income inequality. In more recent research, given the sharp rise in income inequality, scholarship has examined the association between income inequality and inter-generational mobility. Cross-national research suggests the presence of a strong correlation, the so-called “Great Gatsby Curve,” in which contexts of higher inequality are associated with lower intergenerational mobility (Corak, 2013).

While there is some evidence of a similar relationship when comparing between U.S. Commut- ing Zones (Chetty et al, 2014), other work suggests no relationship when comparing across states

29 (Bloome, 2015). Our work suggests that income inequality may indeed lay a foundation of unequal investment in children that could be expected to manifest in diﬀerent adult attainment.

Second, sociologists of the family have charted just how historically contingent popular defini- tions of “good” parenting are (i.e. Hays, 1998; Wrigley, 1989). Recent influential work on parenting practices observes a class-specific manifestation of “good parenting” in the contemporary period in which high-SES parents make extraordinary investments of money and time in their children

(Lareau, 2003). However, scholars disagree about the origin of this parenting schema (Sherman and Harris, 2012). While some situate the motivation for these practices in parental education and culture (Kalil, 2015), others point to more structural economic drivers (Sherman and Harris, 2012).

Our work shows how intertwined these explanations really are - income inequality powerfully shapes how high-SES parents invest in their children. But, these eﬀects are not simply economic. Rather, contexts of higher income inequality really do seem to reshape parental preferences for investment in children.

Finally, as Cherlin, Ribar, and Yasutakea (2016) note, there is a surprising dearth of studies that examine the relationship between income inequality and outcomes of interest using individual-level data. Yet, these designs are perhaps best suited to approximate causal eﬀects of income inequality

(Evans et al, 2004). Our research contributes to this body of evidence. Further, we answer the recurrent calls of those who have synthesized the literature on income inequality to both investigate how income inequality shapes class disparities and the mechanisms by which income inequality may aﬀect behavior (Neckerman and Torche, 2007; Truesdale and Jencks, 2016).

In sum, we provide new evidence that rising income inequality is reshaping the parenting practices of Americans along class lines. Rising income inequality has increased the class gap in parental expenditures on investment in children - money spent on daycare, lessons and classes, and schooling.

This is the result both of the mechanical concentration of income, but also of changing parental preferences. Our evidence for the latter helps to tell the origin story of the much discussed rise in concerted cultivation among high-SES American parents. We ﬁnd that these widening class gaps in expenditures were not oﬀset by narrowing class gaps in childcare time. As inequality rose, the class gap in parental developmental investment in American children also rose, with troubling implications for intergenerational mobility.

30 References

Altintas, Evrim. 2016. “The Widening Education Gap in Developmental Child Care Activities in the United States, 1965-2013.” Journal of Marriage and Family 78: 26-42. Bassok, Daphna, Jenna Finch, RaeHyuck Lee, Sean F. Reardon, and Jane Waldfogel. 2016. “So- cioeconomic Gaps in Early Childhood Experiences: 1998 to 2010.” AERA Open 2(3):1-22. Bloome, Dierdre. 2014. “Income Inequality and Intergenerational Income Mobility in the United States.” Social Forces 93(3): 1047-1080. Bodovski, Katerina and George Farkas. 2008. “ ‘Concerted Cultivation’ and Unequal Achievement in Elementary School.” Social Science Research 37(3), 903-919. Carneiro, P. and M. Rodriguez. 2009. “Evaluating the Effect of Maternal Time on Child Devel- opment Using the Generalized Propensity Score.” Institute for the Study of Labor, 12th IZA European Summer School in Labor Economics. Clogg, Clifford C., Eva Petkova, and Adamantios Hatirou. 1995. “Statistical Methods for Comparing Regression Coefficients Between Models.” American Journal of Sociology 100(5):1261-93. Corak, Miles. 2013. “Income Inequality, Equality of Opportunity, and Intergenerational Mobility.” Journal of Economic Perspectives 27(3): 79-102. Chetty, Raj, Nathaniel Hendren, Patrick Kline and Emmanuel Saez. 2014. “Where is the Land of Opportunity? The Geography of Intergenerational Mobility in the United States.” Quarterly Journal of Economics 29 (4): 1553-1623. Cherline, Andrew, David Ribar, and Suzumi Yasutake. 2016. “Nonmarital First Births, Marriage, and Income Inequality.” American Sociological Review 81(4) 749?770. Deaton, Angus and Darren Lubotsky. 2003. “Mortality, Inequality, and Race in American Cities and States.” Social Science and Medicine (March): 1139-53. Del Bono, Emilia, Marco Francesconi, Yvonne Kelly, and Amanda Sacker. 2014. “Early Maternal Time Investment and Early Child Outcomes.” IZA Discussion Paper No. 8608. Del Boca, Daniela, Chiara Monfardini, and Cheti Nicoletti. 2012. “Self Investments of Adolescents and their Cognitive Development.” IZA Discussion Paper 6868. Duncan, Greg and Richard Murnane. 2011. “Introduction: The American Dream, Then and Now.” In Whither Opportunity: Rising Inequality, Schools, and Children’s Life Chances. Eds. Murnane and Duncan. New York: Russell Sage Foundation Press. Employment Policy Institute. 2012. The State of Working America. (Online Link) Evans, William, Michael Hout, and Susan Mayer. 2004. “Assessing the Effect of Economic Inequal- ity." In Social Inequality. Ed. Katherine Neckerman. New York: Russell Sage Foundation. Fiorini, Mario and Michael Keane. 2014. “How the Allocation of Children’s Time Affects Cognitive and Noncognitive Development.” Journal of Labor Economics 32(4): 787-836. Fisher, Kimberly, Jonathan Gershuny, Sarah M. Flood, Joan GarcÃa Roman, and Sandra L. Hofferth. 2015. “American Heritage Time Use Study Data Extract System: Version 1.0.” Min- neapolis: University of Minnesota. (Online Link) Frank, Mark W. 2014. “A New State-Level Panel of Annual Inequality Measures over the Period 1916 - 2005" Journal of Business Strategies 31(1):241-263.

31 Frank, Mark, Estelle Sommeiller, Mark Price, and Emmanuel Saez. 2015. “Frank-Sommeiller-Price Series for Top Income Shares by US States since 1917." The World Top Incomes Database Methodological Notes. July. Greeman, Emily, Katerina Bodovski, and Katherine Reed. 2011. “Neighborhood Characteristics, Parental Practices and Children’s Math Achievement in Elementary School.” Social Science Research 40(5): 1434-1444. Guryan, Jonathan, Erik Hurst, and Melissa Kearney. 2008. “Parental Education and Parental Time with Children.” Journal of Economic Perspectives 22(3): 23-46. Hays, Sharon. 1996. The Cultural Contradictions of Motherhood. New Haven: Yale University Press. Heckman, James. 2006. “Skill Formation and the Economics of Investing in Disadvantaged Chil- dren.” Science 312: 1900-1902. Hsin, Amy. 2009. “Parent’s Time with Children: Does Time Matter for Children’s Cognitive Achievement?” Social Indicators Research 93:123-126. Jencks, Christopher, Marshall Smith, Henry Acland, Mary Jo Bane, David Cohen, Herbert Gintis, Barbara Heyns, and Stephan Michaelson. 1972. Inequality: A Reassessment of the Eﬀect of Family and Schooling in America. New York: Basic Books. Kalil, Ariel. 2014. “Addressing the Parenting Divide to Promote Early Childhood Development for Disadvantaged Children.” The Hamilton Project, Brookings Institution. Kalil, Ariel, Rebecca Ryan, and Michael Corey. 2012. “Diverging Destinies: Maternal Education and the Developmental Gradient in Time With Children.” Demography 49:1361-1383. Kalil, Ariel, Kathleen Ziol-Guest, Rebecca Ryan, and Anna Markowtiz. 2016. “Changes in Income- Based Gaps in Parent Activities with Young Children from 1988 to 2012.” AERA Open 2(3): 1-17. Kalil, Ariel. 2015. “Inequality Begins at Home: The Role of Parenting in the Diverging Destinies of Rich and Poor Children.” (Pp. 63-83) In Families in an Era of Increasing Inequality Eds. Amato et al. Zurich: Springer International. Kaushal, Neeraj, Katherine Magnuson, and Jane Waldfogel. 2011. “How Is Family Income Related to Investments in Children’s Learning?” In Whither Opportunity: Rising Inequality, Schools, and Children’s Life Chances. Eds. Murnane and Duncan. New York: Russell Sage Foundation. Kearney, Melissa and Phil Levine. 2014. “Income Inequality and Early Nonmarital Childbearing.” Journal of Human Resources 49(1): 1-31. Kenworthy, Lane and Leslie McCall. 2008. “Inequality, Public Opinion, and Redistribution.” Socio- Economic Review 8:35-68. Kornrich, Sabino and Frank Furstenberg. 2013. “Investing in Children: Changes in Parental Spend- ing on Children, 1972-2007.” Demography 50:1-23. Kornrich, Sabino. 2016. “Inequalities in Parental Spending on Young Children: 1972 to 2010.” AERA Open 2(2):1-12. Lareau, Annette. 2002. “Invisible Inequality: Social Class and Childrearing in Black Families and White Families.” American Sociological Review 67(5): 747-776. Levey Friedman, Hilary. 2013. Playing to Win: Raising Children in a Competitive Culture. Berkeley: University of California Press.

32 Mare, Robert D. and Christine R. Schwartz. 2006. “Educational Assortative Mating and the Family Background of the Next Generation: A Formal Analysis.” Riron to Hoho (Sociological Theory and Methods) 21:253-77. Mayer, Susan. 2001. “How Did the Increase in Economic Inequality between 1970 and 1990 Affect Children’s Educational Attainment?” American Journal of Sociology 107(1): 1-32. McLanahan, Sara. 2004. “Diverging Destinies: How Children Are Faring under the Second Demo- graphic Transition?” Demography 41(4): 607-27. Paternoster, Raymond, Robert Brame, Paul Mazerolle, and Alex Piquero. 1998. “Using the Correct Statistical Test for the Equality of Regression Coefficients.” Criminology 36(4): 859-866. Pew Research Center. 2015.Raising Kids and Running a Household: How Working Parents Share the Load. (Online Link) Picketty, Thomas and Emmanuel Saez. 2003. “Income Inequality in the United States, 1913-1998.” Quarterly Journal of Economics 118(1): 1-39. Price, Joe. 2010. “The Effect of Parental Time Investments: Evidence from Natural Within-Family Variation.” Working Paper. Ramey, Garey and Valerie Ramey. 2010. “The Rug Rat Race.” Brookings Papers on Economic Activity Spring: 129-176. Reardon, Sean. 2011. “The Widening Academic Achievement Gap Between the Rich and the Poor: New Evidence and Possible Explanations.” In Whither Opportunity: Rising Inequality, Schools, and Children’s Life Chances. Eds. Murnane and Duncan. New York: Russell Sage Foundation. Reardon, Sean and Ximena Portilla. 2016. “Recent Trends in Income, Racial, and Ethnic School Readiness Gaps at Kindergarten Entry.” AERA Open 2(3): 1-18. Sacks, Daniel and Betsey Stevenson. 2010. “Comment: The Rug Rat Race.” Brookings Papers on Economic Activity Spring: 185-199. Schneider, Daniel and Orestes P. Hastings. “Income Inequality and Household Labor.” Working paper presented at the 2016 Population Association of America conference. Sherman, Jennifer and Elizabeth Harris. 2012. “Social Class and Parenting: Classic Debates and New Understandings.” Sociology Compass 6(1): 60-71. Thomsen, Mette Kjaargaard. 2015. “Parental Time Investments in Children: Evidence from Den- mark.” Acta Sociologica. Truesdale, Beth and Christopher Jencks. 2016. “The Health Effects of Income Inequality: Averages and Disparities.” Annual Review of Public Health 37: 413-430. Villena-Roldan, Benjamin and Ceclilia Rios-Aguilar. 2012. “Causal Effects of Maternal Time- Investment on Children’s Cognitive Outcomes.” IZA Discussion Paper. Waldfogel, Jane and Elizabeth Washbrook. 2011. “Early Years Policy.” Child Development Research vol. 2011, Article ID 343016, 12 pages, 2011. Wilkinson, Richard and Kate Pickett. 2009. “Income Inequality and Social Dysfunction.” Annual Review of Sociology 35: 493-511. Wrigley, Julia. 1989. “Do Young Children Need Intellectual Stimulation? Experts’ Advice to Par- ents, 1900-1985.” History of Education Quarterly 29(1) 41?75. Zick, Cathleen and Keith Bryant. 1996. “A New Look at Parents’ Time Spent in Child Care: Primary and Secondary Time Use.” Social Science Research 25(3): 260-280.

33 Table 1. Hypotheses Mapped to Data, Outcome, and Predictors

Hypothesis Micro-Data Yist Classist Inequalitys,t−1 State Gini #1 CEX $ Investment in Children <18 Household income rank State Top 10% Share

Household income rank, State Gini #2A CEX $ Investment in Children <18 controlling for HH income State Top 10% Share

Household income rank, State Gini #2B CEX $ Investment in Children <18 controlling for HH income State Top 10% Share

$ Investment in Children <18 State Gini #2C CEX Household income rank as%ofincome StateTop10%Share

State Gini #3 CEX $ Consumption for Children <18 Household income rank State Top 10% Share

State Gini #4 CEX $ Investment in Children <18 Parental education State Top 10% Share

Parental education, State Gini #4A CEX $ Investment in Children <18 controlling for HH income State Top 10% Share

$ Investment in Children <18 State Gini #4B CEX Parental education as%ofincome StateTop10%Share

Maternal Time Caring State Gini #5 AHTUS Household income rank forChildren<18 StateTop10%Share

Paternal Time Caring State Gini #5 AHTUS Household income rank forChildren<18 StateTop10%Share

34 Table 2. Relationship between State-Level Income Inequality and Gaps by Household Income Rank in Expenditures on Investment in Children (CEX 1980-2014) Main Control for Income Dollars Investment/Income (1) (2) (3) (4) (5) (6) ∗∗∗ ∗∗∗ ∗∗∗ Income group 0-25p × Gini index -420.4 -414.2 -0.095 (61.2) (61.8) (0.015) Income group 26-75p × Gini index 0 0 0 (.) (.) (.) ∗∗∗ ∗∗∗ ∗ Income group 76-90p × Gini index 594.1 408.3 0.012 (80.6) (80.4) (0.0052) ∗∗∗ ∗∗∗ ∗∗ Income group 91-100p × Gini index 1617.6 876.7 0.028 (226.5) (249.6) (0.0080) ∗∗∗ ∗∗∗ ∗∗∗ Income group 0-25p × Top10%incomeshare -339.7 -323.2 -0.079 (56.5) (56.5) (0.014) Income group 26-75p × Top10%incomeshare 0 0 0 (.) (.) (.) ∗∗∗ ∗∗∗ ∗∗ Income group 76-90p × Top10%incomeshare 541.2 382.4 0.013 (94.3) (88.3) (0.0041) ∗∗∗ ∗∗∗ ∗∗∗ Income group 91-100p × Top10%incomeshare 1351.4 700.1 0.030 35 (216.4) (187.6) (0.0062) ∗∗ ∗∗ Gini index 298.6+ 421.5 0.044 (159.6) (156.6) (0.014) ∗ Top 10% income share 76.4 167.9+ 0.029 (99.1) (93.8) (0.013) ∗∗∗ ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Incomegroup0-25p 178.4 82.8 238.3 139.7 0.068 0.048 (35.3) (24.2) (35.8) (22.9) (0.0085) (0.0060) Incomegroup26-75p 0 0 0 0 0 0 (.) (.) (.) (.) (.) (.) ∗∗∗ ∗ ∗∗ ∗ ∗∗ ∗∗∗ Incomegroup76-90p -192.0 -82.4 -159.7 -89.7 -0.010 -0.0092 (45.0) (38.9) (45.6) (36.8) (0.0031) (0.0018) ∗∗∗ ∗∗ ∗∗ ∗∗ ∗∗∗ ∗∗∗ Incomegroup91-100p -639.0 -288.8 -416.1 -214.6 -0.022 -0.019 (125.7) (84.8) (130.0) (68.2) (0.0046) (0.0028) ∗∗∗ ∗∗∗ Income dollars (in thousands) 1.43 1.45 (0.088) (0.087) Individual controls Yes Yes Yes Yes Yes Yes State Controls Yes Yes Yes Yes Yes Yes State FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Observations 221959 221959 221959 221959 207016 207016 Standard errors in parentheses ∗ ∗∗ ∗∗∗ + p < .1, p < .05, p < .01, p < .001 Table 3. Relationship between State-Level Income Inequality and Gaps by Household Income Rank in Expenditures on Consumption for Children (CEX 1980-2014) (1) (2) Income group 0-25p × Gini index 23.4 (29.1) Income group 26-75p × Gini index 0 (.) ∗∗∗ Income group 76-90p × Gini index -163.1 (38.1) ∗∗ Income group 91-100p × Gini index -183.0 (53.4) Income group 0-25p × Top10%incomeshare 28.9 (24.6) Income group 26-75p × Top10%incomeshare 0 (.) ∗∗∗ Income group 76-90p × Top10%incomeshare -146.2 (30.3) ∗∗∗ Income group 91-100p × Top10%incomeshare -206.4 (51.6) Gini index 42.8 (52.9) Top 10% income share 76.3 (47.1) ∗ Incomegroup0-25p -28.4 -27.4 (17.4) (11.1) Incomegroup26-75p 0 0 (.) (.) ∗∗∗ ∗∗∗ Incomegroup76-90p 127.8 96.6 (22.7) (13.7) ∗∗∗ ∗∗∗ Incomegroup91-100p 171.6 155.0 (31.0) (22.5) Individual controls Yes Yes State Controls Yes Yes State FE Yes Yes Year FE Yes Yes Observations 221867 221867 Standard errors in parentheses ∗ ∗∗ ∗∗∗ + p < .1, p < .05, p < .01, p < .001

36 Table 4. Relationship between State-Level Income Inequality and Gaps by Parental Education in Expenditures on Investment in Children (CEX 1980-2014) Main Control for Income Dollars Investment/Income (1) (2) (3) (4) (5) (6) ∗∗∗ ∗∗ No HS × Gini index -266.6 -192.2 0.0060 (70.7) (66.4) (0.0069) HS no BA × Gini index 0 0 0 (.) (.) (.) ∗∗∗ ∗∗ ∗∗ BA+ × Gini index 709.5 498.6 0.026 (168.5) (159.8) (0.0078) ∗∗ ∗ No HS × Top10%incomeshare -240.0 -163.7 0.0080 (71.1) (63.0) (0.0056) HS no BA × Top10%incomeshare 0 0 0 (.) (.) (.) ∗∗ ∗∗ ∗∗ BA+ × Top10%incomeshare 593.7 412.9 0.026 (169.3) (153.8) (0.0081) Giniindex 178.1 282.2+ 0.017 (148.6) (140.3) (0.013) Top10%incomeshare 21.5 85.4 0.010 (98.1) (92.3) (0.010) ∗ ∗∗ ∗∗∗ No HS 95.3 44.9 56.0 15.7 -0.013 -0.013 (38.9) (31.5) (36.4) (28.3) (0.0041) (0.0026) HSnoBA 0 0 0 0 0 0 (.) (.) (.) (.) (.) (.) ∗∗ BA+ -257.6 -103.6 -146.5 -36.5 -0.0068 -0.0029 (93.8) (68.0) (88.9) (62.1) (0.0044) (0.0034) ∗∗∗ ∗∗∗ Incomedollars(inthousands) 1.55 1.55 (0.081) (0.082) Individual controls Yes Yes Yes Yes Yes Yes State Controls Yes Yes Yes Yes Yes Yes State FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Observations 221959 221959 221959 221959 207016 207016 Standard errors in parentheses ∗ ∗∗ ∗∗∗ + p < .1, p < .05, p < .01, p < .001

37 Table 5. Relationship between State-Level Income Inequality and Gaps by Household Income Rank in Time Spent with Children (AHTUS 1975-2014) Mothers Fathers (1) (2) (3) (4) Income group 0-25p × Giniindex 17.45 218.4 (218.7) (281.0) Income group 26-75p × Gini index 0 0 (.) (.) Income group 76-100p × Giniindex 17.73 48.54 (160.2) (127.5) Income group 0-25p × Top10%incomeshare -3.392 177.6 (177.1) (226.7) Income group 26-75p × Top10%incomeshare 0 0 (.) (.) Income group 76-100p × Top10%incomeshare 1.989 94.09 (120.5) (97.18) Gini index 522.4 -746.3 (1007.9) (1090.5) Top 10% income share -299.6 -263.6 (447.5) (430.1) Incomegroup0-25p -22.99 -10.45 -136.9 -86.82 (138.2) (87.46) (172.9) (107.6) Incomegroup26-75p 0 0 0 0 (.) (.) (.) (.) Incomegroup76-100p 14.06 24.23 -16.47 -31.88 (101.0) (59.08) (79.87) (47.16) Individual controls Yes Yes Yes Yes State Controls Yes Yes Yes Yes State FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes Observations 19947 19947 14446 14446 Standard errors in parentheses ∗ ∗∗ ∗∗∗ + p < .1, p < .05, p < .01, p < .001

38 Figure 1: National Gini and Top 10% Income Share by Year (1975-2013) .65 50 .6 45 .55 40 National Gini Index National Gini .5 National Top 10% IncomeShare Top National 35 .45 39 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 Year Year Figure 2: Parental Expenditures Per Quarter By Income Group 800 600 400 200 Parentalexpenditures ($/quarter) 0 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014

Year

Income 0-25p Income 26-75p Income 76-90p Income 91-100p

40 Figure 3: Maternal and Paternal Time Spent on Child Care by Year 70 120 60 100 50 40 80 30 Minutes/day of child care, fathers Minutes/day of child care, mothers 60 20

1975−76 1994−1995 1998−2000 2003−2006 2007−2010 2011−2014 1975−76 1994−1995 1998−2000 2003−2006 2007−2010 2011−2014

Year Year 41 Income 0−25p Income 26−75p Income 76−100p Income 0−25p Income 26−75p Income 76−100p Figure 4: Box Plot of State-Level Gini and Top 10% Income Share by Year (1975-2013) 60 .7 50 .6 40 State GiniIndex State .5 State Top 10% IncomeShare Top State 30 .4 42 2011 2011 2012 2013 2012 2013 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 excludes outside values excludes outside values Figure 5: Binned Scatter of Household Expenditures on Investment in Children by Household Income Rank and by State-Level Gini and Top 10% Income Share 800 800 600 600 400 400 200 200 Parentalexpenditures ($/quarter) Parentalexpenditures ($/quarter) 0 0 .45 .5 .55 .6 .65 .7 .3 .4 .5 .6 Gini (lagged) Top 10% Income Share (lagged) 43 Income 0-25p Income 26-75p Income 0-25p Income 26-75p Income 76-90p Income 91-100p Income 76-90p Income 91-100p Figure 6: Predicted Household Expenditures on Investment in Children by Household Income Rank and by State-Level Gini and Top 10% Income Share 800 800 600 600 400 400 LinearPrediction LinearPrediction 200 200 0 0 .48 .58 .67 .33 .42 .54 Gini index Top 10% income share 44 Income group 0-25p Income group 26-75p Income group 0-25p Income group 26-75p Income group 76-90p Income group 91-100p Income group 76-90p Income group 91-100p

Note: Points at 5th percentile, median, and 95th percentile of inequality with 95% conﬁdence intervals Appendix Table 1. Relationship between State-Level Income Inequality and Gaps by Household Income Rank in Expenditures on Investment in Children, Full Model Results (CEX 1980-2014) (1) (2) ∗∗∗ Income group 0-25p × Gini index -420.4 (61.2) ∗∗∗ Income group 76-90p × Gini index 594.1 (80.6) ∗∗∗ Income group 91-100p × Gini index 1617.6 (226.5) ∗∗∗ Income group 0-25p × Top10%incomeshare -339.7 (56.5) ∗∗∗ Income group 76-90p × Top10%incomeshare 541.2 (94.3) ∗∗∗ Income group 91-100p × Top10%incomeshare 1351.4 (216.4) Gini index 298.6+ (159.6) Top 10% income share 76.4 (99.1) ∗∗∗ ∗∗ Incomegroup0-25p 178.4 82.8 (35.3) (24.2) ∗∗∗ ∗ Incomegroup76-90p -192.0 -82.4 (45.0) (38.9) ∗∗∗ ∗∗ Incomegroup91-100p -639.0 -288.8 (125.7) (84.8) Income dollars (in thousands)

∗∗∗ ∗∗∗ HS no BA 57.8 58.2 (6.99) (7.10) ∗∗∗ ∗∗∗ BA+ 206.3 207.1 (10.8) (10.9) ∗∗∗ ∗∗∗ Household size -56.9 -57.0 (2.17) (2.20) ∗∗ ∗∗ Age -5.94 -5.79 (2.05) (2.05) Age × Age 0.037 0.036 (0.023) (0.023) ∗∗∗ ∗∗∗ Female non-white -31.6 -31.4 (7.99) (8.12) ∗∗∗ ∗∗∗ Female not present -54.1 -54.8 (8.42) (8.41) ∗∗ ∗∗∗ Male non-white 32.8 33.2 (9.42) (9.40) ∗∗∗ ∗∗∗ Male not present -25.4 -25.6 (5.33) (5.29) ∗ ∗ Statemedianincome(lagged) 0.0020 0.0017 (0.00083) (0.00080) Stateunemploymentrate(lagged) 0.25 2.03 (2.85) (2.96) Statepercentblack 211.4 264.9 (381.3) (344.1) Statepercentforeignborn -215.0 -234.0 (220.0) (220.0) State FE Yes Yes Year FE Yes Yes Observations 221959 221959 Standard errors in parentheses ∗ ∗∗ ∗∗∗ + p < .1, p < .05, p < .01, p < .001 45 Appendix Table 2. Relationship between State-Level Income Inequality and Gaps by Parental Education using Education of Lowest Educated Parent in Expenditures on Investment in Children (CEX 1980-2014) Main Control for Income Dollars Investment/Income (1) (2) (3) (4) (5) (6) ∗∗∗ ∗∗∗ No HS × Gini index -352.6 -271.8 -0.0042 (58.1) (53.1) (0.0054) HS no BA × Gini index 0 0 0 (.) (.) (.) ∗∗ ∗ ∗ BA+ × Gini index 691.7 482.6 0.030 (239.6) (232.5) (0.011) ∗∗∗ ∗∗∗ No HS × Top10%incomeshare -279.1 -199.6 -0.00034 (57.2) (49.3) (0.0046) HS no BA × Top10%incomeshare 0 0 0 (.) (.) (.) ∗ ∗∗ BA+ × Top10%incomeshare 548.1 372.8+ 0.029 (225.4) (214.3) (0.010) ∗ Gini index 327.2+ 391.0 0.022 (167.9) (159.1) (0.014) Top10%incomeshare 156.6 183.4+ 0.015 (100.5) (98.0) (0.012) ∗∗∗ ∗∗ ∗∗∗ No HS 127.7 44.8+ 86.6 16.1 -0.0059+ -0.0081 (31.4) (22.6) (28.8) (20.0) (0.0032) (0.0020) HSnoBA 0 0 0 0 0 0 (.) (.) (.) (.) (.) (.) BA+ -228.8+ -64.1 -120.9 -1.85 -0.0086 -0.0037 (134.7) (92.3) (130.1) (87.6) (0.0066) (0.0045) ∗∗∗ ∗∗∗ Incomedollars(inthousands) 1.49 1.49 (0.082) (0.083) Individual controls Yes Yes Yes Yes Yes Yes State Controls Yes Yes Yes Yes Yes Yes State FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Observations 221959 221959 221959 221959 207016 207016 Standard errors in parentheses ∗ ∗∗ ∗∗∗ + p < .1, p < .05, p < .01, p < .001

46 Appendix Table 3. Relationship between State-Level Income Inequality and Gaps by Parental Education using Number of College-Completed Parents in Expenditures on Investment in Children (CEX 1980-2014) Main Control for Income Dollars Investment/Income (1) (2) (3) (4) (5) (6) No BA × Gini index 0 0 0 (.) (.) (.) ∗∗ ∗ 1 BA × Gini index 395.9 261.4 0.013+ (121.8) (119.6) (0.0067) ∗∗∗ ∗∗ ∗∗ 2 BAs × Gini index 1115.6 789.8 0.035 (280.8) (273.5) (0.012) No BA × Top10%incomeshare 0 0 0 (.) (.) (.) ∗∗ ∗ ∗ 1 BA × Top10%incomeshare 374.1 259.9 0.016 (119.3) (112.3) (0.0068) ∗∗ ∗ ∗∗ 2 BAs × Top10%incomeshare 844.1 568.3 0.032 (253.4) (236.4) (0.010) Gini index 148.5 262.2+ 0.019 (153.8) (146.8) (0.013) Top10%incomeshare 9.18 80.7 0.012 (97.5) (91.9) (0.010) NoBA 000 0 00 (.) (.) (.) (.) (.) (.) 1BA -106.9 -39.7 -35.3 3.50 0.00064 0.0015 (68.8) (49.4) (67.9) (47.2) (0.0038) (0.0029) ∗∗ 2BAs -425.1 -142.2 -254.6+ -41.5 -0.0093 -0.0025 (156.3) (101.4) (151.3) (94.3) (0.0071) (0.0043) ∗∗∗ ∗∗∗ Incomedollars(inthousands) 1.48 1.49 (0.084) (0.084) Individual controls Yes Yes Yes Yes Yes Yes State Controls Yes Yes Yes Yes Yes Yes State FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Observations 221959 221959 221959 221959 207016 207016 Standard errors in parentheses ∗ ∗∗ ∗∗∗ + p < .1, p < .05, p < .01, p < .001