Raising Children, Rising Debt: Mortgage Debt Among American Families

Abstract

American households owe more than $12 trillion in mortgages, which represents the main source of a family’s debt. Scholars connect mortgages to the desire of families, especially better-off households, to seek housing in neighborhoods with good schools for their children, which tend to be more expensive. Although this perspective assumes a children–mortgage link, we do not know whether having children actually increases mortgage, nor whether and how this relationship varies by household income. To examine these issues, we use eleven waves of the Panel Study of Income Dynamics data between 1997 and 2017 and individual fixed effects, as well as propensity score matching and a quasi-experimental design. Our analyses show that generally, (1) families with children are more likely to have mortgage debt and in greater amounts; (2) it is families in the 60th to 100th income percentile who have the most mortgage debt; and (3) critically, families in the roughly 10th to 60th income percentile have more mortgage debt due to having children. These findings defy assumptions that it is well-to-do families that take on more mortgage debt as part of intensive or concerted cultivation parenting practices. Rather, our findings suggest that families who take on mortgage debt related to their children tend to be those in more economically precarious positions for whom debt for the sake of kids may be a financial burden. As such, our findings provide suggestive evidence that financially intensive parenting may contribute to growing wealth inequality among American families with children.

1. Introduction

The United States has been described as a nation of borrowers (Carruthers and Ariovich 2010; Hyman 2011). Notably, in the first quarter of 2023, total household debt in the United States reportedly surpassed $17 trillion, and mortgage balances, the biggest category of household debt, stood at more than $12 trillion (Federal Reserve Bank of New York 2023). Explanations for mounting household debt include structural constraints of stagnant real wages, rising economic insecurity (Leicht and Fitzgerald 2014; Porter 2012; Sullivan et al. 2001), or the cultures of consumerism and finance (Fligstein and Goldstein 2015; Frank 1999, 2007; Schor 1998, 2004). Studies of determinants of mortgage strain (Caputo 2012; Keene et al. 2014; Lichtenstein and Weber 2015; McCormack and Mazar 2015) examine the role of gender, race/ethnicity or immigration, and marital and disability status (Cherian 2014; Gotham 1998; Guy et al. 1982; Massey et al. 2016; Nolan 2018; Quillian et al. 2020; Squires and Kim 1995; Steil et al. 2018).1 Building on this work, we provide a complementary explanation: some families take on (more) mortgage debt due to having children.

Importantly, while mortgage debt is growing, it is not as ubiquitous as it may appear. For example, about 36% of households in America in 2021 were renting their home (Desilver 2021), and in 2022, the share of homeowners who own their home without a mortgage grew to an all-time high of 40% (Tanzi 2023). Who, then, holds mortgage debt in America? And how is it related to children? Scholars argue that families, especially those who are well-off, invest in their children by seeking housing in good-school neighborhoods (Goyette et al. 2014; Holmes 2002; Lareau and Goyette 2014; Owens 2016; Weininger 2014), which tend to be more expensive (Frank 2007) and may require more mortgage debt (Goldstein and Hastings 2019). Although scholars tend to present this argument about the link between children and mortgages, we could not find any studies that directly test these assumptions. In fact, home ownership in good neighborhoods may be beneficial for—and sought by—a range of households, not just those in which children reside. Therefore, the likelihood and size of mortgage debt to finance a home purchase may not depend on having children in the home. To seek clarity on this issue, we pursue the following research questions: (1) Are households with children (up to age 18) more likely to have mortgage debt—and more of it—compared to households without children? (2) If households with children are more likely to carry debt (and in larger amounts), is it related to having children as a driver of mortgage debt accumulation? (3) And how does the relationship between having children and mortgage debt vary by income group?

Our analyses show that, generally, families with children are indeed more likely to have mortgage debt and have a greater amount of it. More specifically, it is families in the 60th to 100th income percentile who have the most mortgage debt. In contrast, families in the roughly 10th to 60th income percentile have more mortgage debt due to having children. These findings defy assumptions that it is well-to-do families who make decisions related to mortgage debt as part of intensive parenting or concerted cultivation. Rather, our findings suggest that families who take on more mortgage debt due to their children tend to be those in more economically precarious positions for whom debt for the sake of kids may be an investment, but it may also be a financial burden.

Our results lend credence to the argument that contemporary financially intensive parenting (Bandelj and Grigoryeva 2021) is implicated in shaping the functioning of mortgage markets in the United States. To the extent that this is true, for more well-off families, taking on mortgages may be “good” debt that helps them leverage money (Charron-Chénier and Seamster 2018). In contrast, for families in more economically uncertain conditions, it appears that attempts at intensive parenting (Hays 1996; Nelson 2010; Ishizuka 2019) may result in financial burdens (Charron-Chénier and Seamster 2018; Lin and Neely 2020), including having to manage more mortgage debt.

Due to the centrality of home equity among household assets in the United States, our investigation further illuminates the importance of family dynamics in shaping wealth inequality in American society (Killewald et al. 2017). Given that “child households experienced increases in within-group wealth inequality that far outpaced” those of other households (Gibson-Davis and Percheski 2018, p. 1028), it is likely that those trends are partly due to the differential consequences of families with children holding mortgage debt.

2. Intensive Parenting and the Children–Mortgage Debt Link over the Life Course

The argument for the children–mortgage debt link is strongly connected to the imperatives of contemporary parenting. Sharon Hays (1996, p. 8) introduced the concept of “intensive” parenting to describe the “copious amount of time, energy, and material resources” that families spend to raise their children in contemporary American society. Since Hays’ groundbreaking work, scholars have provided strong empirical support for child-centered intensive parenting (e.g., Calarco 2018, 2020; Lois 2012; Reich 2014; Weininger et al. 2015) and how this norm is becoming widespread (Chin and Phillips 2004; Edin and Nelson 2013). For instance, Ishizuka (2019) presents respondents with various vignette scenarios that reflect the more-or-less intensive parenting norm. He finds that parents across class and race/ethnicity, both mothers and fathers, express similar support for intensive parenting. Speaking directly to the economic realm, Bandelj and Grigoryeva (2021) document the rise of “financially intensive parenting”, whereby over the past couple of decades, parents from across different wealth backgrounds seem to invest, save, and borrow increasingly more for the sake of their children.

One commodity where parents may be exercising financially intensive parenting may be housing. Although scholars have examined demographic determinants of home ownership, including the role of race and ethnicity (Coulson 1999; Hyde and Fischer 2023; Rucks-Ahidiana 2023), age (Fisher and Gervais 2011), or marriage (Fischer and Khorunzhina 2019), it is also possible that parents—as compared to people without children—are more likely to want to become homeowners because they want to reside in better and safer neighborhoods and to offer a sense of stability and security to their children that owning a home may imply (Pattillo 2013; Zavisca and Gerber 2016), all expressions of intensive parenting.

Specifically, due to the growing focus on the human capitalization of children (Bandelj and Spiegel 2023) parents may feel an increasing pressure to enroll their children in good schools (Lareau and Goyette 2014). Some research finds that parents are willing to pay a higher price for a home if it means their children will attend a school that is academically strong. For example, Black (1999) finds that “a 5 percent increase in elementary school test scores (approximately one standard deviation) leads to an increase in the marginal resident’s willingness to pay of approximately 2.1 percent, or $3948 at the mean house price of $188,000” (p. 578), a consequence of the positive relationship between test scores and house values (Black 1999, p. 578). Frank (2007, p. 44), similarly, argues that “the quality of public schools in the United States is closely linked to local property taxes, which in turn depend on local real estate prices”. This means that families with children have to spend more—and take out larger mortgages—to live in good-school neighborhoods (Goldstein and Hastings 2019; Owens et al. 2016). This could pertain to a large segment of American families, as more than 70% of children in public schooling attend the designated school for their neighborhood (Card and Krueger 1992; Rhodes and DeLuca 2014). In brief, if “parents are willing to pay more than non-parents for desirable schools” (Goldstein and Hastings 2019, p. 438), this may necessitate taking on more mortgage debt.

Parents may also feel like they need houses with separate bedrooms for their children, which is increasingly a norm in the United States (Moroney 2019), and that might lead them to consider bigger and, hence, more expensive houses. In addition to good schools, parents may also want to reside in safe and well-resourced neighborhoods with good access to public transportation and other shared amenities (e.g., clean parks) and where they hope their children will be exposed to quality interactions, including with peers (Holmes 2002; Goyette et al. 2014; Desmond 2016; Zavisca and Gerber 2016; Rich 2017). All these considerations are likely going to add to the cost of housing, which is often financed with mortgage debt.

Although scholars have highlighted the importance of children to home ownership, it is critical to contextualize purchasing a home within a life course framework (Elder et al. 2003; Mortimer and Shanahan 2003; Pearlin et al. 1997). As such, purchasing a home is an important life course event that fulfills contemporary normative expectations of adulthood that inextricably tie it to age, marriage, and childbirth (Arnett 2015). However, a range of contemporary dynamics, including economic uncertainty, has made life course transitions, including adulthood milestones such as marrying and having children, less standardized and progressively variegated (Shanahan 2000), decoupling marriage, childbirth, and age from each other—and from the purchase of a home (Furstenberg 2010). Rieger et al. (2019) report that the median age of first-time home buyers was 32 in 1997, dropping to 30 in 2009 and climbing to 34 in 2017. These trends show that individuals are buying their home later in the life course. Further, “as a share of all home purchases, first-time buyers accounted for 43.7 percent in 1997 and 38.5 percent in 2017” (p. 6). In other words, fewer buyers are first-time buyers, such as young people fulfilling their desire to live on their own in their own homes.

Moreover, the median age of first marriage was 28 for women and 30 for men in 2023, up from 24 and 26, respectively, in 1990 (Census 2024). The average age of a mother at first birth was 22 in the 1980s, 24 in the 2000s, and about 26 in the 2020s (Khorunzhina and Miller 2022, p. 3). These demographic trends suggest that while first birth and marriage for women are relatively close (median ages 28 vs. 26), with roughly one-half of first births occurring before marriage, partly depending on socio-economic status (Martinez and Daniels 2023), home ownership seems to occur several years after childbirth (median age 34). Thus, the purchasing of a home is increasingly decoupled from childrearing and marriage, although not entirely, as normative expectations and material necessities continue to exert converging pressure (Fischer and Khorunzhina 2019). For this reason, our analyses consider the role of age and marriage for the relationship between mortgage debt and childbearing. In addition, we examine how these relationships are impacted by a family’s income. The aforementioned trends suggest that fulfilling normative adulthood expectations, especially as they relate to home ownership for the sake of children, may be both desired and increasingly requiring of more debt, especially for more economically disadvantaged parents.

3. Who Has Mortgage Debt for Children? The Role of Economic Stratification

There are reasons to expect that the link between mortgage and children will vary across a family’s socio-economic background. Scholars have theorized parental investment through a combination of class-based parenting styles with positional competition (Bertrand and Morse 2016; Charles and Lundy 2013; Doepke et al. 2019; Frank 2007; Levey Friedman 2013; Ramey and Ramey 2010). Lareau (2003) famously argued that families in the middle and upper middle classes engaged in concerted cultivation of their children, which could include residing in good-school neighborhoods to cultivate their children’s school experience (see also Calarco 2020; Kimbro 2021). Relatedly, Owens (2016, p. 566; emphasis added) finds that “among families with children, high-income parents may… have prioritized expenditures on residence in neighborhoods seen as advantageous for their children”. Similarly, Schneider et al. (2018) find that it is high-income parents, those in the 76th to 100th income percentiles, who are investing more resources of money into their children to keep them from falling behind (cf. Kornrich and Furstenberg 2013). According to Goldstein and Hastings (2019), families in the top quintile of the income distribution are especially prone to engage in inequality-fueled position competition to reproduce or advance their children’s social class position, which is part of the reason these families are engaging in parental investment.

The alternative theoretical perspective that similarly predicts variation in the children–mortgage link across the household income distribution derives from what we call the economic sociology of financialization (Lin 2015; Lin and Neely 2020; Lin and Tomaskovic-Devey 2013; Nau 2013). Parents engage in relational work (Zelizer 2012; Bandelj 2020; Hayes and O’Brien 2021), matching economic transactions with appropriate social relations and media of exchange, and, hence, use various financial instruments in a way that they deem to fit the special child–parent bond, including taking on (more) mortgage debt. Yet, these families are also located in the broader structure of financial markets and inequality (Lin and Neely 2020). Therefore, the use of monetary instruments to engage in financially intensive parenting will vary by a family’s socio-economic background. For instance, the most enriched families may take on mortgages as financial instruments for investment and tax advantage reasons since homeowners who itemize deductions may reduce their taxable income by deducting interest paid on a home mortgage. These families have more resources and, consequently, may not need a (large) mortgage to be able to afford a home in a good-school neighborhood for their children. Moreover, higher-income groups are not bound to particular locations with regard to good public schools because they are more knowledgeable about school choice (finding opportunities in non-assigned school districts) or may opt to send their children to private schools (Murnane et al. 2018; Thomas 2019). Therefore, we would expect the high-income bracket families to have sizable mortgages but would not necessarily expect them to do so because of imperatives associated with raising children.

At the other extreme of the income distribution, due to the operation of private financial markets, poor families—who are often unbanked—may not pass qualification tests for mortgage access (Lin and Neely 2020), so they are unlikely to be able to take on mortgages, even if they subscribe to intensive parenting norms (Elliott et al. 2015; Turner 2020). Therefore, we would also not expect these families in the lowest income bracket to have mortgage debt related to children.

However, it is the lower-to-middle-income families, those with enough income to pass credit-line requirements but who cannot buy homes without substantial loan assistance, that may actually end up with more housing debt to reside in good (and, consequently, more expensive) neighborhoods. If so, for these families, mortgages are likely offered on less favorable terms, potentially increasing their inability to meet their obligations on time, which may have a negative impact on their credit scores, the cost of their other loans, and their ability to save (Lin and Neely 2020). In addition, these families may need to forgo expenditures and savings for other purposes to finance their mortgage debt. Therefore, taking on mortgage debt for lower-to-middle-income families can come at a great financial cost, turning efforts at “good parenting”—to provide home ownership in good neighborhoods—into a financial burden.

In brief, given alternative possibilities of how socio-economic background impacts the children–mortgage link, it is imperative to directly examine how the relationship between having kids and mortgage debt varies for families from different income groups.

4. The Case Against the Children–Mortgage Debt Link

It is also quite plausible to expect that having children in the home is not related to mortgage debt. First, central to the “American Dream” cultural narratives in the United States is ownership of a home (Hochschild 1995; Schor 1998). Historically, the home has been among the most important financial assets for families, often pursued (and marketed) to guarantee financial security and intergenerational wealth accumulation. The cultural imperative of home ownership is backed up by the state, which encourages home ownership through mortgages by offering generous tax benefits (Fischer and Huang 2013). A consequence of these dynamics has been a rise in the value of homes as shares of total assets and the related explosion of (household and housing) indebtedness (Wolff 2017).

Moreover, and in contrast to the perspective that parents are concerned with select schools in neighborhoods they cannot quite afford, some researchers find that school “choice” is actually not very relevant for families. Drawing on 110 interviews with parents of school-age children in Chicago, Cuddy et al. (2020) find that their interviewees did not factor schools into their house search, and if they did, schools were just one of many criteria they considered, and their considerations were not as straightforward as the literature suggests (Cuddy et al. 2020, p. 1196). For example, many working-class families were constrained in finding housing that was not only affordable but also safe—a factor that sometimes had higher priority than schools. It is important to note that these families’ “limited or lack of attention to schools in housing searches does not reflect a lower priority on schooling, but rather highlights that the ability to couple housing to schooling in a direct fashion requires resources” (Cuddy et al. 2020, p. 1196; Holmes 2002). Related to that, it is also plausible that parents rent homes in better-school neighborhoods rather than buying those with mortgage debt, given the steep costs (Coulter 2016; Lareau and Goyette 2014).

Further, even if it is the case that parents with children in the home seek good neighborhoods—whose higher prices might be driving more mortgage debt—parents may not be the only people seeking good neighborhoods because such neighborhoods may also ensure overall higher values of all homes. Given the evidence of linking house prices to good-school neighborhoods (and price stability or appreciation of homes close to good schools), it is possible that parents and non-parents alike will want to live in neighborhoods with good schools. For instance, previous research shows that elderly homeowners who have no school-age children are motivated to reside in good-school neighborhoods, mostly by a belief that this will increase the value of their homes (Bergstrom et al. 1982; Hilber and Mayer 2009). This is another reason why we may not find that having children up to age 18 in the home is related to having (more) mortgage debt.

Last but not least, it is possible that an association between mortgage debt and children exists but is driven by unobserved heterogeneity. The link could very well be due to self-selection into parenthood. It is possible, for example, that having children and taking on more mortgage debt are both a function of stable personality traits, such as risk propensity. It may also be the case that both children and mortgage are driven by parents’ efforts to diversify sources of support in older age. In short, it is possible that certain proclivities drive individuals to engage in both child-rearing and home ownership. Therefore, any investigation into the relationship between having children in the home and having (more) mortgage debt needs to address issues of selection.

5. Data

We use the Panel Study of Income Dynamics (PSID) in this study. This publicly available dataset is a nationally representative sample of American families since 1968. The sample individuals include all persons living in the PSID families in 1968 and anyone subsequently born or adopted by a sample person. This is a longitudinal dataset, and all sample members are followed, even when leaving the original household to establish separate family units. The PSID allows us to follow the same household over time. Therefore, we use panel data at the head of household level, including family characteristics, for every other year starting in 1997 to 2017. This gives us 11 waves with 72,810 heads of households. We limit the sample to household heads because, in some cases, data are collected on multiple individuals from the same households, but for the purposes of our analyses, we use data from only one representative, i.e., the head of each household. Since our goal is to examine whether mortgage debt is different over time across households with and without minor co-resident children up to age 18, we exclude households with children older than 18. (The share of those is around 16% from the initial pool of data). This way, our comparison remains only to those households truly without children (not to households with children older than 18).

Our analytic plan includes descriptive and explanatory methods to understand the phenomenon under consideration, using three main dependent variables: mortgage likelihood, mortgage real value, and logarithmic transformation of the mortgage real value to capture the percentage change in mortgage debt by having vs. not-having children up to age 18. (We also do robustness tests with a mortgage-to-income ratio dependent variable). First, we provide relevant descriptive statistics for households with and without children. We also analyze the descriptive statistics by the age of the household head and by the age of the oldest child to take into account the life course dynamics. Then, we deploy fixed-effect models that compare mortgage debt within the same families who are followed over time. We carry out all these analyses by various income brackets and test the robustness of results to different income percentile cut-offs. In addition, we perform propensity score matching to compare the most similar families with and without children, and we also leverage a quasi-experimental condition of a significant decrease in interest rates—an exogenous shock between 2001 and 2004 that made taking on mortgage debt more compelling, as it reduced the cost of loans—to examine behavior pre- and post-intervention, comparing families with and without children across income groups. We explain these in detail below.

5.1. Dependent Variables

Mortgage Likelihood. We create a binary variable, which takes a value of one if the family has a mortgage and zero otherwise. Real Value of Mortgage Debt. We generate a variable with the amount of mortgage debt at the time of the interview. This variable captures the debt in the first (and, if applicable, second) mortgage. The values for this variable represent the principal currently owed in dollars from all mortgages or land contracts on the primary residence. They do not include (potential) home equity loans on the property. To account for inflation, we use the consumer price index (CPI) to equalize the value of the home in 1997 (wave 1). Logarithm of the Real Value of Mortgage Debt. We calculate a simple logarithmic transformation of the real value of mortgage debt. Mortgage-to-Income Ratio: We divide mortgage debt with household income, adding a small constant to 596 households that report 0 income to be able to log this variable for the analyses due to its heavy skew.

5.2. Independent Variables

Children. This is our main independent variable. We create a categorical variable that takes the value of one if a household has dependent children up to and including 18 years of age, following previous research (Kornrich and Furstenberg 2013; Schneider et al. 2018). The following are considered children: son or daughter of the household head (includes adopted children), stepson or stepdaughter of the household head (children of legal wife who are not children of the head), son or daughter of the spouse but not the head, son-in-law or daughter-in-law of the head (includes stepchildren-in-law), foster son or foster daughter, not legally adopted, and children of first-year cohabitor but not of the head. Age. Following previous research and life course considerations, we account for the head of household age and age squared (Barba and Pivetti 2009). Education. We account for the head of household’s completed years of education. Number of persons in household. This variable accounts for the number of individuals living in the family unit at the time of the interview. Controlling for this variable addresses the possibility that higher prices associated with having children are related to the household size (rather than the children in the home), which is driving mortgage debt.2 Income. We use the total family income variable available in the PSID data set. This is a composite variable that adds five different sources of income: taxable income of head and spouse, transfer income of head and spouse, taxable income of other family unit members, transfer income, and Social Security income. We also scale it per capita by dividing the total family income by the number of individuals living in the family unit at the time of the interview. Net worth: We use the variable provided by PSID administrators from the “wealth module”, which comprises eight components, including nonprimary real estate holdings, business/farm equity, stocks/mutual funds, bonds, vehicles, IRA accounts and annuities, cash and/or bank account holdings, and non-housing debt. Data on housing equity are calculated as households’ (self-reported) home values minus their home mortgage debt. All of these wealth components are combined, with missing values imputed when relevant by the PSID administrators. We use the net worth variable that excludes the value of equity in a household’s primary residence. Net worth data are not available for all our waves since the wealth survey module was not consistently implemented in the earlier period, reducing our overall sample size. Our analyses with this variable provide consistent results with those presented herein; thus, to present the most parsimonious models, we do not include net worth as an independent variable in all the analyses we report. Employment: We create a categorical variable that takes the value of one if the head of household is employed and zero otherwise. Robustness checks provide similar results to our main findings; thus, to present the most parsimonious models, we do not include employment in the results we report. Home Ownership. This captures whether the household owns a home in which they live. We use this for some auxiliary analyses to check for the robustness of our results.

6. Analytic Strategies

6.1. Descriptive Statistics

Given assumptions in the literature about the prevalence of mortgages and reasons for mortgages, it is important to run basic descriptive statistics on these trends. We first determine what shares of our full analytical sample have or do not have mortgages and what shares among those have children up to age 18 or not. To examine how the children–mortgage debt link differs by income, we cut the income distribution as follows. First, we define three income groups by delineating the very bottom and the very top of the distribution, given theorizing about the role of credit access and financial leveraging at either end: (a) below the 10th percentile; (b) the 10th—90th percentile; and (c) those above the 90th income percentile. Second, we cut the 10th—90th in various ways, including the 10–50/50–90 (reported), 10–30/30–60/60–90 (reported), and 10–25/25–50/50–75/75–90 (available upon request). Because the arguments in the literature revolve around the importance of life course on home ownership and the desire to reside in good-school neighborhoods, we also present the data by, first, the age of household head in ten-year cohorts after 25, and, second, by the age of the oldest children, breaking it into age categories that roughly correspond to the levels in the U.S. education system, with 0- to 2-year-olds in infant/toddler care, 3- to 4-year-old preschool children, 5- to 11-year-olds in kindergarten through 6th grade, 12- to 13-year-olds in middle school, and 14- to 18-year-olds in high school.

6.2. Fixed Effects

Given the panel nature of the data, we deploy individual fixed effects as our main empirical strategy to capture the effects of having children on mortgage debt across our data waves, which controls for time-invariant conditions. In the first step of our analyses, we run fixed effects regressions using the full sample and various income-level groups for holding mortgage (yes/no) and size of mortgage ($ and logged $).

The full sample size is 72,810 (11 waves; 1997–2017). Note that we initially had 75,479 observations for mortgage value in the dataset, but we lost 2669 observations because of missing data on years of education completed to have a final analytical sample of 72,810 observations for regression analyses. Importantly, when running the mortgage likelihood analyses (logit), the sample size is 29,534. This is less than the full sample because, in fixed effect regressions, all observations (households) that do not vary on whether they did or did not take out a mortgage during the observation period are dropped as they experience no change in the dependent variable. Also note that when we present the analyses by income brackets, sample sizes will be different and smaller depending on how many observations fall into a specific income bracket. In addition, models with net worth have a somewhat smaller sample size because PSID data on wealth were not collected in 1997.

6.3. Propensity Score Matching

As there may be systematic differences between households who have children and those who do not, we also use a strategy that relies on propensity score matching. This is a counterfactual approach to diminish concerns about the non-random distribution of, or selection into, who ends up having children. A propensity score is the conditional probability of assignment to a particular treatment (having children in our case) given the vector of observed covariates (Rosenbaum and Rubin 1983, p. 41). Our vector of observed covariates, or variables that the groups are matched on, include age, gender, race, education completed, being married, income, net worth, and size of household.

We use the psmatch2 command in Stata, which “implements full Mahalanobis matching and a variety of propensity score matching methods to adjust for pre-treatment observable differences between a group of treated and a group of untreated. Treatment status is identified by depvar==1 for the treated and depvar==0 for the untreated observations” (Leuven and Sianesi 2003). Given the algorithm, our sample is restricted to those household heads to whom a likely counterfactual is found in the control group (as other observations are discarded).

6.4. Quasi-Experiment with Interest Rate Manipulation

We also use a quasi-experimental design that leverages the longitudinal nature of our data and an exogenous shock introduced during the observed period to further test the robustness of our results. In our case, we leverage the pronounced decrease in interest rates by the Federal Reserve in 2001. According to Ravier and Lewin (2012), the nominal long-term mortgage interest rate in the United States fell 113 basis points between 2001 and 2004. As stated in the report on “The First Global Financial Crisis of the 21st Century” (Felton and Reinhart 2008), the monetary policy of low interest rates, in response to the post-9/11 period and the collapse of the economic bubble, injected liquidity in the monetary system that was consequential for the economy. Importantly, the interest rates were reduced to their lowest level in 50 years. Consequently, lenders could be taking higher risks to get higher returns, and families with limited financial strength acquired credit more easily, allowing them, presumably, to actualize their desire for more mortgages as part of their intensive parenting strategy.

That is, if it is true that families with children in certain income brackets have more pent-up demand for mortgage debt, then we would expect to see those families with children acquire more mortgage debt after the reduction in interest rates (exogenous event) compared to others. If the propensity to take on mortgage debt is no different across income brackets, then we should generally see no difference between the pre- and post-reduction in interest rate periods, as all households would be taking advantage of lower interest rates, which make mortgages less expensive.

Therefore, we compare the same household before and after the exogenous event for those with children compared to those without and across different income groups. For this exercise, we make sure to use a sample comprising the same families who have children up to age 18 in waves 3 and 6 and the same families who have no children in the same two waves. More precisely, we use households from wave 3 (2001) and wave 6 (2007), pre- and post-reduction in interest rates. This was a time of expanding mortgage markets before the Great Recession and thus represents a natural experiment that captures a time period in which mortgages became much more accessible to all households. Our strategy evaluates the difference between the evolution of the outcome variable (mortgage size) between the pre- and post-reduction in interest rates in 2001. We compute differences in mortgage size within families (pre- and post-event) with and without children across all income groups. Specifically, we identify the difference between the following two values: [mortgage of those with children 0–18 after the exogenous shock minus mortgage of those with children 0–18 before the exogenous shock] and [mortgage of those without children after the shock minus mortgage of those without children before the shock].

7. Findings

7.1. Descriptive Statistics Results

Table 1. Share of mortgage holders by presence of children up to age 18.

	NO Children	YES Children
NO mortgage	40.02%	24.44%	64.46%
YES mortgage	16.80%	18.74%	35.54%
	56.82%	42.18%	100%

N = 72,810.

presents the cross-tabulations for our main variables of analysis. Importantly, we see here that 35.54% of the sample holds mortgage debt. We also see that the most common situation is that of households without children who do not have a mortgage, comprising 40.02% of the whole sample. The shares of those without children who have a mortgage (16.80%) and of those with children younger than 18 with a mortgage (18.74%) are relatively similar. In addition to these results, we find that within the subsample of those who have children up to age 18, 43% have a mortgage. Among other households, 40% have a mortgage. Given that the chi-square test showed that this difference is statistically significant (p < 0.001), these results begin to indicate that those with children may be more likely to have a mortgage. Yet, they do not allow us to conclude that they have a (or higher) mortgage due to having children.3

Table 2. Descriptive statistics of variables in the analysis.

	Children (Up to 18 Years Old)	No Children	T-Test (Diff in Means)
Dependent Variables
Mortgage (yes)	0.434	0.296	0.138 ***
Mortgage amount	$47,286	$24,772	$22,514 ***
Mortgage-to-income ratio (log)	−8.09	−10.06	1.97 ***
Independent Variables
Age	36.186	49.678	−13.492 ***
Male head of household	0.715	0.684	0.031 ***
Black head of household	0.374	0.292	0.082 ***
Married head of household	0.603	0.444	0.159 ***
Years of education of household head	13.162	13.235	−0.073
Total family income (real)	$52,309	$42,707	$9603 ***
Income per capita (real)	$14,736	$27,556	$−12,819 ***
Net worth	$106,537	$206,863	$−100,326 ***
Net worth per capita	$29,388	$126,814	$−97,426 ***
Employed	0.834	0.623	0.211 ***
Number of persons in household	3.730	1.559	2.171 ***
Ownership of home	0.798	0.825	−0.027 ***
N	31,442	41,368

Notes: * p < 0.05, ** p < 0.01, *** p < 0.001.

presents the descriptive statistics for the variables used in the analysis. These statistics also indicate that households with children up to age 18 have, on average, a higher proportion of those with mortgage debt (0.44 vs. 0.30) and more mortgage debt ($47,305 vs. $24,827) compared to no-child households (p < 0.001). On average, families with children also have a younger household head (38 vs. 49; p < 0.001). The proportion of married household heads is lower among families with children (0.61) compared to families without (0.82) (p < 0.001). Even though the total family income is higher ($69,074 vs. $55,480), on average, for families with children compared to families without them, the per capita income is lower ($15,073 vs. $27,521) (p < 0.001). Families with children have more persons, on average, living in the family unit (3.8 vs. 1.6) compared to those without children (p < 0.001). Families with children up to age 18 have lower net worth compared to other households ($116,570 vs. $204,496) (p < 0.001).

We conduct further descriptive analyses that consider the life course of households. First, we break up the sample by the age of the household head and then by the age of the oldest child. From the crosstabs reported in Appendix A 

Table A1. Share of households with mortgage and size of mortgage by age of household head and income percentile.

Household Head Age	<25 yrs		26–35 yrs		36–45 yrs		46–55 yrs		56–65 yrs		>66 yrs
Have children	Y	N	Y	N	Y	N	Y	N	Y	N	Y	N
All income	11%	9%	36%	26%	56%	34%	58%	41%	57%	43%	50%	22%
	7K	7K	36K	27K	65K	33K	65K	34K	63K	34K	32K	15K
0–10th income	1%	1%	3%	2%	10%	4%	17%	8%	25%	13%	25%	8%
	0.1K	0.5K	2K	1K	7K	3K	8K	5K	14K	5K	12K	3K
10–30th income	7%	4%	13%	9%	24%	15%	25%	20%	35%	23%	35%	15%
	3K	2K	7K	5K	13K	7K	12K	8K	24K	10K	28K	6K
30–60th income	20%	16%	36%	24%	48%	35%	46%	42%	55%	44%	48%	28%
	13K	10K	25K	18K	35K	24K	32K	25K	41K	25K	26K	16K
60–90th income	33%	31%	64%	48%	74%	61%	72%	66%	65%	60%	71%	37%
	28K	28K	71K	52K	82K	64K	76K	54K	61K	49K	30K	30K
90–100th income	36%	23%	80%	57%	83%	67%	84%	74%	80%	65%	75%	40%
	37K	24K	156K	105K	170K	117K	147K	117K	165K	88K	75K	52K

Notes: N = 72,810; share with a mortgage in the first line (%); amount of mortgage in italics in the second line ($ thousands).

, we see that, generally, the likelihood of holding a mortgage and holding higher mortgage amounts increases as adults age, up to the 56- to 65-year age bracket, after which it declines. However, slicing the sample by income categories is instructive. First, we notice that those in the highest income decile are the most likely to have a mortgage and the highest amount of mortgage debt. In addition, those in higher income percentile groups (60th and higher) are more likely to hold mortgages and in greater amounts earlier in the life course, that is, between ages 26 and 45. In contrast, those in lower income percentiles are more likely to hold mortgages and in greater amounts in the 56 to 65 age bracket. This suggests that economic standing significantly impacts the evolution of mortgage debt during the life course.

We see that those with children ages 14–18 are more likely to have a mortgage, at 48%, compared to those with children ages 0–2 (40%), ages 3–4 (40%), ages 5–11 (43%), and ages 12–13 (46%). Similar to analyses by age of household head, the share of those with children who hold mortgages in the 90th–100th income category is more than 80% and does not vary much by the age of children. Similarly, the share of mortgage for those in the 60–90th income category is high, at nearly 70%, and remains relatively stable across various child ages. We see that for families in these categories, mortgage values that are magnitudes larger than for the counterparts in lower income categories (with children of comparable ages) are actually decreasing as their children grow older (from birth to 18), which may signal these income classes reflect long-term mortgage holders who are paying off a greater share of their mortgage over time as their children are getting older.

On the other hand, we observe notable increases in shares of those with a mortgage and the amount of mortgage for families in the 0–10th and 10–30th income categories. For those in the 0–10th percentile, the share with a mortgage increases from 4% (with an average of $1984 in mortgage) when their children are 0–2 years old to 8% ($3823) when children are in high school (14- to 18-year-olds). Similarly, in the 10–30th income category, these figures increase from around 13% ($7930) for preschool children to 16% ($9093), 19% ($9446), and 20% ($10,400) when children are (5–11, 12–13, and 14–18 years old, respectively). There is a similar trend for the 30–60th income category, from 38% when children are preschool age to 42% and 44% when children are 12–13 and 14–18, respectively, with the amount of mortgage across the various children age categories remaining around $29,000 to $30,000. In short, we see that for those in the lower- and middle-income categories, mortgage debt increases as their children get closer to elementary and high school age, a trend that is the opposite of that for the more economically advantaged families who have the most mortgage debt when their children are not yet of school age.

7.2. Full Sample Analyses

In Table 3, we present the results of the fixed-effect estimation for each of our outcome variables. Models 1–6 in Table 3 show the results of fixed-effect models using the full analytic sample, with mortgage likelihood, real value of mortgage, and real value of mortgage logged as outcome variables, respectively. They highlight that families with children up to 18 years old are more likely to have a mortgage (Model 1, b = 0.344, p < 0.001) and have more mortgage (Model 3, b = $10,824, p < 0.001) than families without children, net of covariates. These results are consistent when economic standing is measured as net worth instead of income (Models 2 and 4). The results also show that families with children up to 18 years of age, on average, have a mortgage debt between 68 and 75% higher, on average, than families without children (Models 5 and 6, with income and net worth, respectively, p < 0.001). Analyses of the mortgage-to-income ratio show a statistically positive coefficient for children, as well. Since the models shown are fixed effects within individual households tracked over time, controls for marriage, race, and gender are dropped because of lack of variation. Similarly, propensity score matching analyses, matching on a full set of covariates (age, education, gender, race, marriage, income, wealth, and size of household) for the full sample, show a difference of 0.114 in the likelihood of having a mortgage (p < 0.001) and a difference of $17,573 in the amount of mortgage (p < 0.001) between those with and without children, with the latter group having a lower likelihood and amount of mortgage debt (Table 5).

Table 3. Mortgage debt, full sample across all income groups (comparison group “no children”).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Mortgage Likelihood, FE	Mortgage Likelihood, FE	Real Mortgage Value, FE	Real Mortgage Value, FE	Log Real Mortgage Value, FE	Log Real Mortgage Value, FE	Log Mortgage-to-Income, FE
Having children 0–18	0.344 ***	0.252 **	10,824.17 ***	9708.84 ***	0.754 ***	0.683 ***	0.909 ***
	(0.068)	(0.073)	(845.94)	(899.17)	(0.07)	(0.07)	(0.10)
Age of household head	0.010 ***	0.006	873.576 ***	716.16 ***	0.031 ***	0.022 ***	0.033 ***
	(0.00)	(0.00)	(34.84)	(39.12)	(0.00)	(0.00)	(0.00)
Age squared	−0.003 ***	−0.004 ***	−30.308 ***	−31.29 ***	−0.003 ***	−0.003 ***	−0.004 ***
	(0.00)	(0.00)	(1.16)	(1.27)	(0.00)	(0.00)	(0.00)
Years of education of household head	−0.070 **	−0.080 **	−452.102	−444.43	−0.043	−0.048	−0.107 **
	(0.03)	(0.03)	(304.86)	(315.40)	(0.02)	(0.03)	(0.04)
Number of people in household	0.542 ***	0.419 ***	5483.495 ***	4911.65 ***	0.482 ***	0.430 ***	0.452 ***
	(0.03)	(0.03)	(297.91)	(314.58)	(0.02)	(0.03)	(0.04)
Real household income per capita	0.529 ***		0.134 ***		0.216 ***
	(0.03)		(0.01)		(0.02)
Net worth (in thousands)		0.756 ***		0.529		0.580	−0.343
		(0.02)		(0.30)		(0.23)	(0.34)
Married household head	/	/	/	/	/	/	/
Male household head	/	/	/	/	/	/	/
Black household head	/	/	/	/	/	/	/
Constant			−10,463.18 **	2070.71	0.480	3.141	−9.66 ***
			(4052.34)	(4217.66)	(0.35)	(0.33)	(0.48)
N	29,534	25,722	72,810	67,288	72,810	67,288	67,288

Notes: * p < 0.05, ** p < 0.01, *** p < 0.001; No overall constant is identified in logit fixed-effects models. Marriage, gender, and race variables were dropped due to lack of variation in fixed-effects estimation.

Table 3. Mortgage debt, full sample across all income groups (comparison group “no children”).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Mortgage Likelihood, FE	Mortgage Likelihood, FE	Real Mortgage Value, FE	Real Mortgage Value, FE	Log Real Mortgage Value, FE	Log Real Mortgage Value, FE	Log Mortgage-to-Income, FE
Having children 0–18	0.344 ***	0.252 **	10,824.17 ***	9708.84 ***	0.754 ***	0.683 ***	0.909 ***
	(0.068)	(0.073)	(845.94)	(899.17)	(0.07)	(0.07)	(0.10)
Age of household head	0.010 ***	0.006	873.576 ***	716.16 ***	0.031 ***	0.022 ***	0.033 ***
	(0.00)	(0.00)	(34.84)	(39.12)	(0.00)	(0.00)	(0.00)
Age squared	−0.003 ***	−0.004 ***	−30.308 ***	−31.29 ***	−0.003 ***	−0.003 ***	−0.004 ***
	(0.00)	(0.00)	(1.16)	(1.27)	(0.00)	(0.00)	(0.00)
Years of education of household head	−0.070 **	−0.080 **	−452.102	−444.43	−0.043	−0.048	−0.107 **
	(0.03)	(0.03)	(304.86)	(315.40)	(0.02)	(0.03)	(0.04)
Number of people in household	0.542 ***	0.419 ***	5483.495 ***	4911.65 ***	0.482 ***	0.430 ***	0.452 ***
	(0.03)	(0.03)	(297.91)	(314.58)	(0.02)	(0.03)	(0.04)
Real household income per capita	0.529 ***		0.134 ***		0.216 ***
	(0.03)		(0.01)		(0.02)
Net worth (in thousands)		0.756 ***		0.529		0.580	−0.343
		(0.02)		(0.30)		(0.23)	(0.34)
Married household head	/	/	/	/	/	/	/
Male household head	/	/	/	/	/	/	/
Black household head	/	/	/	/	/	/	/
Constant			−10,463.18 **	2070.71	0.480	3.141	−9.66 ***
			(4052.34)	(4217.66)	(0.35)	(0.33)	(0.48)
N	29,534	25,722	72,810	67,288	72,810	67,288	67,288

Notes: * p < 0.05, ** p < 0.01, *** p < 0.001; No overall constant is identified in logit fixed-effects models. Marriage, gender, and race variables were dropped due to lack of variation in fixed-effects estimation.

Table 4. Mortgage debt by income groups, fixed effects (all waves, comparison group “no children”).

	Mortgage Likelihood				Mortgage Amount $
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Income Groups	0–10%	10–50%	50–90%	90–100%	0–10%	10–50%	50–90%	90–100%
Have children 0–18	−0.215	0.535 ***	0.315 **	0.577	−213.602	4420.487 ***	7314.266 ***	14,643.13
	(0.495)	(0.131)	(0.115)	(0.307)	(682.909)	(812.717)	(1720.735)	(8014.249)
Age of household head	0.003	0.012 *	−0.004	−0.105 ***	0.517	287.301 ***	848.790 ***	593.965
	(0.024)	(0.005)	(0.005)	(0.015)	(26.499)	(34.173)	(75.224)	(336.513)
Age squared	−0.003 ***	−0.003 ***	−0.003 ***	−0.005 ***	−1.301	−11.216 ***	−38.246 ***	−107.818 ***
	(0.001)	(0.000)	(0.000)	(0.001)	(0.712)	(0.994)	(2.959)	(14.670)
Years of education of household head	−0.117	−0.031	−0.104 *	−0.033	−71.998	−48.198	−38.421	−4840.008
	(0.112)	(0.047)	(0.048)	(0.139)	(178.441)	(276.719)	(715.685)	(3684.093)
Number of people in household	0.499 **	0.520 ***	0.414 ***	0.352 **	358.055	3061.279 ***	7369.983 ***	12,125.100 ***
	(0.160)	(0.049)	(0.053)	(0.132)	(216.680)	(286.062)	(760.004)	(3292.332)
Real household income per capita	0.000	0.000 ***	0.000 ***	0.000 *	−0.124 *	0.264 ***	0.235 ***	0.076 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.056)	(0.035)	(0.046)	(0.018)
Married household head	/	/	/	/	/	/	/	/
Male household head	/	/	/	/	/	/	/	/
Black household head	/	/	/	/	/	/	/	/
Constant					4033.796	−8080.081 *	−5697.436	142,382.200 *
					(2319.569)	(3725.275)	(10,365.84)	(56,212.65)
N	614	8476	9951	1605	10,146	31,827	25,194	5643

Notes: * p < 0.05, ** p < 0.01, *** p < 0.001; no overall constant is identified in logit fixed-effects models. Marriage, gender, and race variables were dropped due to lack of variation in fixed-effects estimation.

Table 4. Mortgage debt by income groups, fixed effects (all waves, comparison group “no children”).

	Mortgage Likelihood				Mortgage Amount $
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Income Groups	0–10%	10–50%	50–90%	90–100%	0–10%	10–50%	50–90%	90–100%
Have children 0–18	−0.215	0.535 ***	0.315 **	0.577	−213.602	4420.487 ***	7314.266 ***	14,643.13
	(0.495)	(0.131)	(0.115)	(0.307)	(682.909)	(812.717)	(1720.735)	(8014.249)
Age of household head	0.003	0.012 *	−0.004	−0.105 ***	0.517	287.301 ***	848.790 ***	593.965
	(0.024)	(0.005)	(0.005)	(0.015)	(26.499)	(34.173)	(75.224)	(336.513)
Age squared	−0.003 ***	−0.003 ***	−0.003 ***	−0.005 ***	−1.301	−11.216 ***	−38.246 ***	−107.818 ***
	(0.001)	(0.000)	(0.000)	(0.001)	(0.712)	(0.994)	(2.959)	(14.670)
Years of education of household head	−0.117	−0.031	−0.104 *	−0.033	−71.998	−48.198	−38.421	−4840.008
	(0.112)	(0.047)	(0.048)	(0.139)	(178.441)	(276.719)	(715.685)	(3684.093)
Number of people in household	0.499 **	0.520 ***	0.414 ***	0.352 **	358.055	3061.279 ***	7369.983 ***	12,125.100 ***
	(0.160)	(0.049)	(0.053)	(0.132)	(216.680)	(286.062)	(760.004)	(3292.332)
Real household income per capita	0.000	0.000 ***	0.000 ***	0.000 *	−0.124 *	0.264 ***	0.235 ***	0.076 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.056)	(0.035)	(0.046)	(0.018)
Married household head	/	/	/	/	/	/	/	/
Male household head	/	/	/	/	/	/	/	/
Black household head	/	/	/	/	/	/	/	/
Constant					4033.796	−8080.081 *	−5697.436	142,382.200 *
					(2319.569)	(3725.275)	(10,365.84)	(56,212.65)
N	614	8476	9951	1605	10,146	31,827	25,194	5643

Notes: * p < 0.05, ** p < 0.01, *** p < 0.001; no overall constant is identified in logit fixed-effects models. Marriage, gender, and race variables were dropped due to lack of variation in fixed-effects estimation.

7.3. Income Group Analyses

In light of our theoretical framework on the role of economic stratification, we conduct further analyses to examine whether the children–mortgage debt link we found in the entire sample is driven by families in specific locations of the income distribution. We examine separately those between (0–10th], (10th–50th], (50th–90th], and (90th–100th] income groups. We also break it down further into (10th −30th], (30th–60th], and (60th–90th] income groups (available upon request). Table 4 shows that the children–mortgage debt link is not robust in the income groups below the 10th income percentile and above the 90th income percentile. In contrast, for the (10th–50th] and (50th–90th] income groups, fixed effects show that families with children are more likely to hold a mortgage and are more likely to have more mortgage debt compared to households without children in those income groups.

The propensity score matching analyses reported in Table 5, matching on a full set of covariates (age, education, gender, race, marriage, income, wealth, and size of household), show consistent results to these findings. That is, differences between households with and without children are not statistically significant across all income brackets. Specifically, for the mortgage amount, we see a difference of $3943 (p < 0.01) and $10,812 (p < 0.01) for the households with and without children in the (10th–50th] and (50th–90th] brackets, respectively. However, the p-values are outside the bounds of statistical significance for the differences among these households that fall into the (0–10th] and (90th–100th] percentile income groups.

Table 5. Mortgage debt, propensity score matching.

Mortgage Likelihood	Difference	se
All income groups	0.114 ***	(0.01)
0–10%	−0.042	(0.03)
10–50%	0.031 +	(0.02)
50–90%	0.096 **	(0.03)
90–100%	0.051	(0.11)
Mortgage Amount $	Difference	se
All income groups	17,572.91 ***	(1660.77)
0–10%	−2109.29	(1644.75)
10–50%	3943.14 **	(1234.12)
50–90%	10,811.63 **	(3267.99)
90–100%	48,041.62	(38,746.85)

Notes: + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001; Ns vary across different income groups, as the sample is restricted to those household heads to whom a likely counterfactual is found in the untreated group (no children group). The vector of observed covariates on which the treated/untreated groups are matched includes age, gender, race, education completed, being married, income, net worth, and size of household.

Table 5. Mortgage debt, propensity score matching.

Mortgage Likelihood	Difference	se
All income groups	0.114 ***	(0.01)
0–10%	−0.042	(0.03)
10–50%	0.031 +	(0.02)
50–90%	0.096 **	(0.03)
90–100%	0.051	(0.11)
Mortgage Amount $	Difference	se
All income groups	17,572.91 ***	(1660.77)
0–10%	−2109.29	(1644.75)
10–50%	3943.14 **	(1234.12)
50–90%	10,811.63 **	(3267.99)
90–100%	48,041.62	(38,746.85)

Notes: + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001; Ns vary across different income groups, as the sample is restricted to those household heads to whom a likely counterfactual is found in the untreated group (no children group). The vector of observed covariates on which the treated/untreated groups are matched includes age, gender, race, education completed, being married, income, net worth, and size of household.

7.4. Pre-/Post-Interest Rate Manipulation

Table 6. Real mortgage debt by income groups, pre-/post-interest rate manipulation.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Income Group	0–10%	10–50%	50–90%	90–100%	10–30%	30–60%	60–90%
Have children 0–18	7483.17	14,859.12 ***	7140.68	−37,960.44	9593.81 *	14,789.54 **	6553.58
	(4594.59)	(3796.49)	(6493.74)	(25,037.04)	(3923.90)	(5511.45)	(8604.02)
Age of household head	−122.59	−458.41 ***	−1275.67 ***	−3198.26 ***	−414.79 ***	−772.56 ***	−1419.66 ***
	(121.03)	(108.06)	(219.50)	(824.21)	(108.36)	(170.56)	(298.25)
Demeaned age squared	2.05	5.10	24.82 *	76.08	4.99	12.46 +	29.47 *
	(4.40)	(4.43)	(10.15)	(42.16)	(4.40)	(7.42)	(13.65)
Married	12,373.07	18,184.49 ***	7986.85	−2221.83	13,145.09	18,219.84 ***	5775.48
	(4039.13)	(3106.82)	(5769.53)	(23,740.05)	(3298.94)	(4550.14)	(7764.89)
Male household head	875.59	−2578.71	−7498.19	18,834.57	−505.74	−8853.87	−6823.34
	(3525.98)	(3219.79)	(9376.23)	(55,919.67)	(3141.09)	(5628.44)	(13,183.97)
Black household head	−5620.76 +	−8699.35 ***	−1064.91	40,923.53	−6896.68 **	−7027.76	−920.87
	(3077.71)	(2907.63)	(5601.49)	(35,471.85)	(2963.14)	(4355.92)	(7819.69)
Years of education of household head	700.59	940.68 +	2003.82	2386.90	872.83	1325.15	959.24
	(589.45)	(537.44)	(970.48)	(4087.13)	(544.30)	(856.77)	(1239.11)
Number of people in household	−1346.16	−2230.70 *	−4417.76	−3422.95	−1990.92 +	−2761.35 +	−4628.22 +
	(1184.08)	(1076.60)	(1986.24)	(7619.99)	(1089.22)	(1647.59)	(2629.69)
Net worth	−0.009	0.009	−0.010 *	−0.004	0.014 +	0.005	−0.011
	(0.01)	(0.01)	(0.00)	(0.00)	(0.01)	(0.01)	(0.01)
Constant	2369.81	14,485.96	52,784.35	148,795.3	12,616.43	27,988.76	77,298.38
	(11,486.52)	(10,283.42)	(19,278.24)	(88,591.6)	(10,499.94)	(16,115.92)	(25,439.28)
N	525	1984	1527	260	1077	1365	1069

Notes: The dependent variable is a difference in mortgage size ($) among those with and without children in the pre- and post-event conditions; + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001.

reports the results of the quasi-experiment where we compare outcomes of households pre- and post-interest rate manipulation. Remember that the lowering of interest rates by the Federal Reserve after 2001 made loans less expensive and thus may have released a pent-up demand for mortgages by households. We are comparing whether the response to this exogenous shock is similar or different across households with and without children, as well as across different income groups.

Specifically, models in Table 6 show that households with children in the 10th–50th income percentile (Model 2, b = $14,859, p < 0.001), as well as the 10th–30th (Model 5, b = $9593, p < 0.05) and 30th–60th (Model 6, b = $14,790, p < 0.01) income percentiles, accrue more mortgage debt than those without children post-intervention. Model 3 in Table 6 shows that the difference in mortgage amount between households with and without children up to age 18 (among those in the (50th–90th] percentile in the income distribution, b = $7141) is not statistically significant. Similarly, Model 7 shows that among those in the (60th–90th] percentile in the income distribution (b = $6554), the difference between those with and without children pre- and post-intervention is not statistically significant. (As in the fixed-effects results reported in Table 4, those in the lowest and highest income deciles do not show statistically significant differences).

On the whole, the pre-/post-intervention difference results show that the link between having children and mortgage debt is the most robust for families in the range between the 10th and the 60th income percentiles. This range suggests that taking on more mortgage debt when a family has children up to age 18 in the home is not a distinct feature of what are traditionally considered to be upper-middle class or upper-class families, as for those income groups, the with/without children differences do not remain robust when tested under the conditions of the interest rate manipulation.

7.5. Robustness Checks

We ran several tests to check the robustness of the results reported in the findings section. First, we ran all regressions without the education control because that somewhat increased our sample size. Second, we ran regressions including the married variable but using random effects models due to little variation across cases in fixed-effects models. Third, we ran the income group regressions using net worth instead of income. Fourth, we ran regressions that included the employment variable. Fifth, we ran regressions that included the home ownership variable. We also ran regressions with time fixed effects.

Furthermore, we ran regressions using various subsamples to perform additional sensitivity analyses (some suggested by reviewers). We ran the analyses on (a) homeowners only; (b) a subsample where household heads are ages 25–55, the ages most prone to having children; (c) a sample that included only those without children and without a mortgage in 1997 at the start of our observation period; thus, only those who were “eligible” to yet have children or to yet take on a mortgage during our period of analysis; (d) a sample that removes those who own their houses without mortgage debt, so the comparison was closer to just those who remain renters; (e) a sample that includes only the period before the Great Recession (1997 through 2007); (f) a sample that drops the years 2005 through 2007 due to sensitivity to the Great Recession; and (g) a sample that drops the years 2008 through 2010 because of the proximity to the aftermath of the Great Recession. Finally, we also checked the robustness of our results when using the mortgage-to-income ratio (leverage) as a dependent variable across all income groups. Notably, since in many of the above-listed specifications we are dividing our sample into smaller and smaller slices, we do run into power issues, in that some Ns are too small to meaningfully compare income tercile groups. However, the variety of sensitivity checks most consistently supports the conclusion that it is families with children up to age 18 in the income group between roughly the 10th and 60th percentiles that have more mortgage debt than other households (available upon request).

8. Discussion and Conclusions

American households hold nearly $12 trillion in mortgage debt. Because mortgage debt constitutes the largest portion of American households’ indebtedness, it is critical to understand which families are more likely to take on mortgage debt and why. In this paper, we examine whether having children up to age 18 is related to mortgage debt and how this varies across income groups. To do so, we rely on eleven waves of the Panel Study of Income Dynamics data, deploying a range of analytic methods. We find that, in general, families with children up to age 18 are more likely to have a mortgage and tend to have more mortgage debt than others. However, analyses by income percentiles reveal important nuances. First, households located between the 60th and 100th percentiles of the U.S. income distribution are the most likely to have a mortgage and hold the highest amounts of mortgage debt. However, despite common assumptions that rich households would be the most likely to have more mortgage due to the intensive parenting of their children, we find that this association is not consistent across different analytic techniques (including fixed effects for various subsamples, propensity score matching, and pre-/post-intervention difference) for the households in the upper third of the income distribution. Therefore, we do not find strong support for the claim that rich households take on more mortgage debt for the sake of children. In contrast, we find that the children–mortgage debt link is the most robust and consistent for those in the 10th to 60th income percentiles. These findings indicate that it is families with lower-to-middle income who end up taking on more housing debt in association with parenting their children.

Abundant research documents the myriad reasons parents prefer to reside in good neighborhoods (Lareau and Goyette 2014; Owens 2016; Goldstein and Hastings 2019). This includes school academic performance (Bast and Walberg 2004; Erickson 2004; Weininger 2014) but also neighborhood safety, access to public transportation and other shared amenities, peer-to-peer interactions, racial homogeneity, and other things parents want for their children (Holmes 2002; Goyette et al. 2014; Desmond 2016; Zavisca and Gerber 2016; Rich 2017). Given the structure of financing public education in the United States, better-school neighborhoods are also more expensive (Frank 2007). Our analyses show that this exerts financial pressure on families in the lower- and middle-income brackets who aspire to abide by the contemporary imperative of intensive parenting for the sake of their children by taking on more mortgage debt. In gist, in a valiant attempt to be “good parents”, families from lower socio-economic backgrounds are exposed to heavier financial burdens related to mortgage debt.

It is important to acknowledge that the exact income percentile cut-off identifying which families take on more debt because of children might depend on methodological decisions and that some thinner income percentile slices (such as quartiles and terciles) run into issues of model power for various robustness checks on smaller subsamples. Nevertheless, our multiple sensitivity analyses with various income bracket categories and many subsamples provide compelling evidence that when it comes to the lowest decile and the upper third of the income distribution, having children in the home does not directly influence the mortgage debt amount.

Importantly, this does not mean that families who are better off do not have mortgage debt. In fact, households in our dataset from the upper end of the income distribution have mortgages and the most sizable mortgage debt. This aligns with the finding by Lin and Neely (2020, p. 123) that in the United States, the top 20th percent of income distribution holds nearly half of the country’s residential debt, and this figure has not changed much since the early 1990s. Still, as our findings show, these well-off families’ financial decisions related to mortgages are less connected to their children. One reason is that these families do not need to take on debt to reside in preferred locations but take on mortgages anyway because it is financially advantageous for them due to the generous tax benefits on mortgage interest and because not locking equity in their home frees them to invest it elsewhere. It is also possible that, for these families, the neighborhood of residency and the choice of school for children are not as tightly linked, as they are likely more knowledgeable about school choice options and may have the economic resources to choose private schools for their offspring (Murnane et al. 2018). We hope that future research will provide a more direct test of these conjectures.

Finally, our study can inform current debates about the dynamics of economic inequality in American society, particularly the role and character of debt and inequality among families with children, where inequality has grown the fastest (Western et al. 2008; Gibson-Davis and Percheski 2018). Our results show that taking on mortgage debt by families with children can have starkly uneven consequences across the socio-economic spectrum. These conclusions also have significant repercussions for understanding the major milestones in the life course and the transition to adulthood, in which the normative expectations of marriage, childbirth, and home ownership are becoming increasingly impacted by one’s economic standing.

On the one side, families with children who possess sizable assets take on the largest amounts of mortgage debt earlier in the life course, but they are likely able to benefit from this debt (i.e., they can leverage it) because it is usually granted on favorable terms, has tax advantages, and these families are also able to invest money not spent on house equity into other wealth-generating opportunities. Basically, for these families, mortgage debt confers financial advantages and likely contributes positively to the intergenerational accumulation of wealth. By contrast, lower- and middle-income families are more likely to take out (bigger) mortgages later in the life course that are offered on less favorable terms, and/or they need to forgo expenditures and savings for other purposes to finance their mortgage debt, increasing the likelihood of not being able to meet their obligations on time, which will all have a negative impact on their credit scores, the cost of their other loans, and their ability to save. For lower- and middle-income families, such debt represents a financial burden, not leverage. Servicing this debt—for longer—might diminish resources and, therefore, negatively impact these families’ ability to accumulate wealth across generations.

It is important to note that our investigation does not negate the importance of stagnant real wages, rising economic insecurity, consumerism, or the rise of the culture of finance in shaping housing, specifically mortgage decisions. These have been researched extensively and are a part of the story. Yet, our investigation highlights another social force that significantly influences mortgage markets—contemporary parenting. By focusing on the likely financial consequences of intensive parenting, our investigation underscores how structural constraints and cultural imperatives interact—in this case, to boost the wealth of the already well-off and to further deepen the economic insecurity of the lower and middle classes. To the degree that taking on mortgage debt by these less well-to-do families is related to providing for and investing in children, taking on debt carries a strong moral force. Those with heavier debt burdens will likely increase their economic uncertainty, but changing course is difficult given the moral force behind mortgage debt decisions. It is, after all, for the children.