• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptNIH Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Popul Res Policy Rev. Author manuscript; available in PMC Oct 1, 2013.
Published in final edited form as:
Popul Res Policy Rev. Oct 1, 2012; 31(5): 703–726.
Published online Jul 14, 2012. doi:  10.1007/s11113-012-9249-2
PMCID: PMC3432984

A Sudden Transition: Household Changes for Middle Aged U.S. Women in the Twentieth Century


Between 1900 and 1990, the percentage of U.S. white women aged 40–69 living with a child of their own fell from 63% to 27%, with three fourths of that change occurring between 1940 and 1960. Historical census data from the Integrated Public Use Microdata Series and longitudinal data from the Panel Study of Income Dynamics allow an historical and contemporary examination of co-residence patterns among these women. Analysis reveals three eras: a system of co-residence in the early twentieth century, a sudden transition toward separate households at mid century, and the maintenance of that separate household system thereafter. The scholarly literature features cultural, demographic, and economic explanations for the long-term decline in co-residence, but has given little attention to the rapid mid-century shift. Analysis of IPUMS data confirms the long-term effects of declines in mortality and fertility, and concomitant declines in the age of mothers at last birth, but also points to a sharp drop in the age of children at marriage in the mid-twentieth century. These factors raised the potential for the formation of separate households, but this historical era was also a propitious one for separation: income gains for young workers were unprecedented, the labor force participation of married women rose, and immigration fell. Analysis of PSID data from 1968 to 2009 confirms the salience of children’s socioeconomic circumstances—particularly their marriage and employment prospects but also the increasing availability of higher education—in maintaining the separate household system. While the data analyzed allow only inferences about cultural factors, the resiliency of the new household system, even in periods of economic decline, suggests that it is now likely buttressed by strong normative views.

Keywords: household, co-residence, home leaving, motherhood, aging

In 1900, most white U.S. women who had children continued to live with at least one of those children until the end of their lives. By 1990, such co-residence was the exception rather than the rule. Historical analyses have often argued that 19th century cultural norms drove mothers and children to stay together, while changing norms in the 20th century encouraged them to live apart. Research on the period since World War II—when the separate household system became firmly established—favored the explanatory role of demographic change and economic affluence, particularly declining fertility and generous public old age benefits. Moreover, such studies began to highlight the role of adult children’s opportunities, modeling the simultaneous decision-making of both generations.

Scholars have yet to link historical patterns to more recent conditions. This article addresses that gap, using both long-term and contemporary data to periodize the transition to a new household system for middle-aged women, to weigh the factors that were most likely to have created that system, and to assess which have tended to maintain it in recent decades. Using cross-sectional data that cover the twentieth century, we link the characteristics of mothers to the probability of living away from all of their children. With longitudinal data for 1968 to 2009, we then assess the influence of characteristics of both mothers and children on this outcome. While these data do not include direct measures of cultural effects, they provide strong evidence of demographic and economic foundations for the establishment and maintenance of the separate household system.

Analysis reveals the long-term secular trend away from intergenerational co-residence among middle-aged mothers, with a particularly spectacular change between 1940 and 1960, and demonstrates that the separate household system has been maintained since then. The standard model of generationally distinct households for older and younger adults in the late 20th century now appears to be well embedded, even as the economic conditions that helped create it have faltered. Only in the first decade of the 21st century has declining economic capacity appeared to have had even a marginal effect in undermining this model; that the current “boomerang” or “accordion” phenomenon is discussed as a social pathology further attests to the normalization of generationally separate households (Newman 2012).

Previous Research and Theory

For much of U.S. history, black and white households have formed under very different conditions of possibility. De jure structures of slavery and Jim Crow legislation, accompanied and followed by forms of discrimination that ranged from de jure to de facto have long limited the household forms available to African American families. As a result, black and white households followed quite distinct historical trajectories, particularly apparent in patterns of intergenerational co-residence and young adult home leaving that at times ran in opposite directions, as demonstrated by several studies (Goldscheider and Bures 2003; Ruggles and Goeken 1992; Kramarow 1995). For this reason, scholarship generally treats the groups separately. The literature discussed here is, for the most part, confined to white households, as is our own analysis.

Historical accounts of the white nineteenth-century household in the United States portray a familial landscape in which aging parents expected to live with their children. At least one child remained in the parental home, or aging parents (usually widows) moved in with their children after a period of living apart. By the mid-twentieth century, however, the incidence of co-residence had declined visibly. Historians initially argued that this change was prompted by cultural transformation. Smith (1986) and Ruggles (1994) found that nineteenth-century families with greater means were more likely to co-reside, indicating a preference for this arrangement. Had this preference remained intact, rising incomes should have increased the rate of co-residence. Since co-residence instead began to decline in the early twentieth century, Ruggles (1994) and Kramarow (1995) argued that a cultural change must have occurred. In his more recent work, however, Ruggles (2003, 2007) attributes the decline in co-residence to structural economic factors, specifically the demise of agriculture and other forms of household production.

Mid-twentieth-century social scientists (e.g. Parsons 1949) also described the intergenerational co-residence norm as dominant in the past, but, unlike historians, they attributed its fall to long-term demographic and economic changes rather than cultural ones. Declines in fertility, age at last birth, mortality, and age at marriage combined to lengthen the portion of a mother’s life between the marriage of her last child and the death of her husband, reducing the risk of widowhood that had made residential companionship likely in the past (Glick 1977); Glick and Parke 1965; Kertzer 1995; Weiss 2000). Secular increases in divorce rates may have undermined protection for older women, but the move to separate households continued nonetheless.

Levy (1965) and Kobrin (1976) argue that, as older couples lived longer and had fewer children, the probability of bearing the burden of co-residence rose for any given child, undermining an ideal more easily held when risk was small. Despite some caution, most scholars consider the reduction in the number of children available to support parents a critical factor in levels of co-residence (Treas and Bengston 1982; cf. McGarry and Schoeni 2000). Even increases in fertility during the baby boom had the seemingly paradoxical effect of accelerating rather than reversing the trend toward the empty nest (Weiss 2000: baby boom couples had more children but at younger ages and in quicker succession, so were younger when their last child reached adulthood than their own mothers had been.

Other scholars give more weight to steady increases in average income and the expansion of private and public pension programs—most notably the establishment of Social Security—in the formation of separate households after 1940 (Michael et al. 1980; Pampel 1983; Costa 1999; Bethencourt and Rios-Rull 2009; McGarry and Schoeni 2000). Gratton and Gutmann (2010), however, demonstrate that Social Security cannot fully explain a shift toward the empty nest, since many men who did not receive benefits moved toward that household form. The postwar period witnessed an unprecedented expansion in the employment of married women, providing them considerable economic independence in late middle age.

Recent research has examined the resources of children, suggesting that separate residence may turn on adult children’s opportunities for employment and college education (Smits et al. 2010; Crimmins and Ingegneri 1990; Ward et al. 1992; Kotlikoff and Morris 1990). Using cross-sectional data for unmarried women 65 and over and their children, Wolf and Soldo (1988) demonstrated the negative impact of children’s marriage and employment on the probability of co-residence. Several studies have indicated that the typical age at which children left home dropped in the mid-twentieth century (Gutmann et al. 2002; Stanger-Ross et al. 2005; Buck and Scott 1993). Mobilization for the Second World War moved adult children out of their parents’ homes, while sharp income gains for young workers during and after the war and subsidized home financing allowed them to marry at younger ages. College educational opportunities spread across more of the population in the postwar period, maintaining the new pattern of early home leaving, even as age at marriage increased. After 1960, though children began marrying at older ages, they also became more likely to leave home prior to marriage (Gutmann et al. 2002; Stanger-Ross et al. 2005; Buck and Scott 1993).

Cultural histories of the recent period confirm a change in values regarding co-residence (Plant 2010; Feldstein 1998). Americans now believe that children ought to move out of the parental home by their early twenties, and children hold this view more rigidly than do parents (Stettersten 1998; Veevers et al. 1996). Ethnocultural norms have also been linked to persistent co-residence. As an example, complex households among Mexican Americans, the largest twentieth-century immigrant group, are said to be buttressed by familistic values (Wilmoth 2001; Burr and Mutchler 1999; Gonzales 2007; Sarkisian et al. 2007). Still, ethnic values have been difficult to separate from demographic and economic conditions; the shock of immigration itself—rather than the peculiarities of an ethnic culture—may explain differences in living arrangements (Glick and Van Hook 2002; Gratton et al. 2007).

The historical and social scientific literature thus presents two worlds: one a past in which co-residence was normal, the expected living arrangement, and the other a present in which co-residence is abnormal, avoided by both generations whenever possible. Few have examined the transition, as white American families moved from one world to the next. Gratton and Gutmann (2010) have shown that long-term trends did play a part, but that a rapid change in the mid-twentieth century demands particular attention. They and other scholars suggest that demographic and economic factors are more important than cultural ones. Finally, if a standard of separate residence has been established in post Second World War America, what maintains that standard? Have the conditions that favored its creation faltered and has it begun to weaken?


Explanation of the marked transformation in older women’s living arrangements requires data that cover the long period over which the shift occurred—the entirety of the twentieth century—and that allow us to identify the effects of both mother and child decision-making on household separation. In order to achieve these desiderata, we use two sources of data, one offering the necessary chronological scope and the other providing information on children as well as mothers. Individual-level census data for 1900 through 1990 from the Integrated Public Use Microdata Series (IPUMS) (Ruggles et al. 2010) provide evidence of the broad contours of change over time. Long-term demographic shifts apparent in these cross sectional data leads to multivariate logistic regression models for particular years, assessing which variables were most determinative of whether a woman lived apart from her children at each census date. We then use longitudinal data about both mothers and children from the Panel Study of Income Dynamics (PSID) for 1968 to 2009—the contemporary period of relative stability—in event history models of the emptying of the parental nest.

IPUMS provides an indispensable source of historically broad, nationally representative individual-level data, standardized across time, but has two critical limitations. First, the data are cross-sectional: we cannot observe shifts in an individual’s household arrangements over time, and when we find an older parent with a co-resident child, it is unclear whether the two had always lived together, whether the child had moved back in with the parent, or whether the parent had moved in with the child. Second, the data describe household relationships rather than familial ones: we observe jointly only mothers and children who live together; data are missing for non-co-resident children.

The PSID, while not as broad historically, solves some of these limitations. Begun in 1968, it is one of the oldest nationally-representative longitudinal surveys of U.S. families. The initial sample of 4,800 families has been re-interviewed annually through 1997 and biennially thereafter; data are currently available through 2009. When family members move out, their new households become part of the sample. While this structure provides the data necessary to assess factors related to children who have left home, the sample becomes increasingly less representative over time. Weighting corrects much of this bias, but the discrepancy between population and sample is particularly apparent in the low proportion of immigrants in the PSID core sample, drawn just before the revival of mass immigration after 45 years of stringent restrictions. Additional families were added in 1997 to address this issue.



We begin with a descriptive analysis of IPUMS samples to answer three questions that have received relatively little attention in the literature. First, did generationally-distinct households become the norm because more women experienced all of their children leaving home or because fewer women reintegrated households with their adult children after living separately? Second, what was the chronology of the shift toward separate households? Third, how did long-term changes in mortality, fertility, and nuptiality contribute to the observed change in co-residence? To answer these, we construct synthetic cohorts of women from the IPUMS samples and employ historical life tables. We compare the timing of changes in living arrangements to the timing of changes in the age of a mother at the birth and marriage of her last child and at widowhood.

Multivariate Logistic Regression

We then fit multivariate logistic regression models to assess the factors determining the log odds that a mother aged 40–69 would live apart from all children in each census year (Model 1). Because mothers can only be identified in the censuses in which women were asked how many children they had borne—1900, 1910 and 1940 through 1990—we limit multivariate analysis to those years; given largely consistent results for all census years, we report models only for 1900, 1940, 1960, and 1990.

Event-History Models

Our cross-sectional analysis of the IPUMS data identifies factors correlated with having had all children already leave home; we follow it with event-history analysis of longitudinal data from the PSID to assess the factors that increase the probability that all of a woman’s children will leave home. Our dependent variable is whether or not a woman lives away from all children in the current wave of the PSID, conditional on the presence of at least one child in her household at the previous wave. Each analytic episode therefore begins with at least one child in the home and represents one wave of the PSID. To assess change over time and accommodate the shift from annual to biennial interviewing in the PSID after 1997, we divide our analysis into three periods, 1968–1979, 1980–1996, and 1997–2009. Each wave or analytic episode is therefore one person-year in the first two periods and two person-years in the third.1 A woman enters our sample in the first episode during which she is at least 40 years old and her youngest co-resident child is at least 16.2 She leaves our sample after the episode during which the last child leaves home, and is otherwise censored when she reaches age 70, dies, or is lost to follow-up.3 Because we are interested only in initial home-leaving, we consider the emptying of the nest an absorbing state: women do not re-enter the analysis if a child returns home.

Our longitudinal analysis employs two approaches. Model 2 estimates mothers’ hazard of having the last co-resident child leave home for any reason, while Model 3 estimates the competing risks of a mother experiencing her last co-resident child leave for college, marriage, or employment. The competing risks model assumes that the three exit options for children are mutually exclusive and absorbing: a mother whose last co-resident child leaves for one destination is no longer at risk of having that child leave for any other. Computationally, this procedure is comparable to multinomial logistic regression (Allison 1995).

Our multivariate logistic and event-history analyses focus on the demographic and economic factors identified in the literature as having the most bearing on a mother’s living arrangements in late middle age: age, childbearing history, marital status, nativity, education, employment status, and the sex and marital and employment status of her children.

Demographic Variables: Age, Childbearing History, Marital Status, and Nativity

In both IPUMS and PSID analyses, age is recorded in 5-year groups, with 55–59 as the reference level. However, age has different impacts in the two analyses, given the differences in the structure of the models. In the IPUMS analysis, we expect age to have a more linear effect, with advancing age increasing the likelihood that a mother’s children will have already left home. In the PSID analysis, age will not have a linear effect, as women leave the risk set when their children leave home. Therefore, after a certain age—to be determined empirically—women who remain in the risk set are those whose children plan to stay, so higher age is associated with a reduced hazard of having children leave home.

Childbearing history is operationalized differently in the two sets of models because different measures are available in the IPUMS and PSID datasets. IPUMS provides the number of children ever born, which we include categorically, with levels of one, two (reference), three, and four or more (alternate specifications did not change the results). In the PSID analysis, we use age at last birth for childbearing history because complete histories were available only for women who were in the sample as of 1985, and only these women could be reliably linked to their biological children. In order to avoid limiting the sample to these women, we instead included all women who were heads of household or wives of heads, assumed that the youngest child of head was their youngest child, and calculated the mother’s age at last birth from the age of her youngest child. Children ever born and age at last birth are correlated with one another—all else equal, women with more children will be older at their last birth. These variables are also related to important secular changes in fertility over the twentieth century: the general trends toward fewer children and an earlier cessation of childbearing. To control for these trends, analyses limit the “risk” of living away from children to women who do not have children younger than 18 in IPUMS and women who do not have children younger than 16 in PSID.

In both datasets, marital status is identified as married with husband present (reference), widowed, or divorced/separated.

Nativity in the IPUMS analysis simply indicates whether the woman was born abroad (foreign born = 1). The PSID did not collect place of birth; the closest proxy is the region where the head grew up (and that not available in 1994–2001). Therefore, if a woman lives in a household with a head raised in a foreign country, or if she is in the immigrant sample in the 1997–2009 period, we treat her as foreign born. Between 1994 and 1996, we have no indication of foreign birth and so treat all women as native born, contributing to noise in this variable in the second period.

Socioeconomic Variables: Farm Residence, Education, Employment, and Income

Farm residence, high school completion, and employment are dichotomous variables in both datasets. In IPUMS, high school completion is available only beginning in 1940, and in the PSID it is available only in the second and third periods. In IPUMS employment status is indicated directly; in the PSID a woman is considered employed if she reports at least 500 hours of work in the previous year (10 hours per week). In both datasets, income is a continuous variable measured in units of $1,000, and includes husband’s income if he is present. In the IPUMS analysis, income is indexed to 1999 dollars and is available only for the years 1940 and 1960–90. In the PSID analysis, income is indexed to 1980 dollars and is available for all years.

Child Variables: Sex, Marital Status, and Employment

The longitudinal structure of the PSID dataset and our event history models allow us to test the effects of child characteristics on the likelihood that a mother will experience all of her children leaving home and the effects of both mother and child characteristics on the likelihood of different modes of departure. While reasons for leaving may be complex, we make commonsense assumptions about principal causation: if a child is in the mother’s home at one wave and living in an educational institution at the next, we assume he or she left for college. If a child is unmarried and living in the mother’s home at one wave, and married and living in a different household at the next, we assume he or she left home for marriage. Over 80% of children who leave for destinations other than college or marriage are employed either in the last year they are observed in the mother’s home or in the first year they are observed on their own; in our analysis, we refer to this residual group as having left for “employment.” This last category includes military service, which may have accounted for a larger proportion of home leaving during the first few years of our PSID analysis, when young men were being drafted into the Vietnam War.

In addition to mother characteristics, our event history models include the sex, marital status, and employment status of the youngest co-resident child, who is not necessarily the mother’s youngest child overall. His or her identity can change between episodes if, for example, the youngest of three moves out.4

Tables 1 and and22 give summary measures for the dependent and independent variables in our IPUMS and PSID analyses respectively. In the event-history models, time-varying characteristics of mother and child are updated at the beginning of each episode. “Outcome” indicates the percentage of episodes ending with a child still at home, or with the last child departing for marriage, college, or employment or other destinations.

Table 1
Summary measures for IPUMS analysis, dependent and independent variables.
Table 2
Summary measures for PSID analysis, dependent and independent variables for all episodes and for episodes in which last co-resident child leaves home.


Figure 1 uses IPUMS data to illustrate the living arrangements of four synthetic cohorts of women born between 1870 and 1900. It indicates the percentage of women in each cohort who lived with a child of their own at ages 30, 40, 50, 60, 70, and 80. The top line is the 1870 birth cohort, aged 30 in 1900, 60 in 1930, and 80 in 1950. The last cohort, displayed in the bottom line, was born in 1900 and reached 50 in 1950 and 80 in 1980. The figure demonstrates a dramatic fall in the proportion of women who had at least one child remain in the home, with the most rapid decline occurring between 1940 and 1960. These findings confirm the prevalence of joint living arrangements for late 19th century women noted by many (for example, Ruggles (2003, 2007)) and replicate the rapid fall in co-residence with children that Gratton and Gutmann (2010) have found for older men between 1940 and 1970.

Figure 1
Women living with own child by age and cohort, IPUMS

In each cohort, women who had children tended to live with at least one of them through age 40. Between ages 40 and 70, many experienced all children leaving home; some reintegrated households with their children after age 70, as can be seen in the modestly rising proportion with a child after that age in each cohort. Reintegration increased slightly in more recent cohorts, so a lower rate of reintegration does not explain the secular decline in co-residence. More women live away from their children today because more women now experience all of their children leaving home.

The percentage ever living without children increased across cohorts. Of women born in 1870 (solid black line), just about half lived away from their children at age 70, while over 80% of those born in 1900 did at that age. However, the figure also demonstrates that, for all cohorts, the most dramatic shift occurred at mid century. The line for women born in 1880 (dashed gray line) is roughly parallel to that for women born in 1870 until age 70, reached in 1950, when co-residence with children was suddenly much lower than for women born in 1870 at the same age. Similarly, women born in 1890 (dotted black line) had levels of co-residence equivalent to those of women born in 1880 at each age until 1950 (age 60 for women born in 1890), when the proportion sharply declined. Women born in 1900 exhibited co-residence at levels lower than those of any previous cohort at all ages, but particularly at age 50, reached at mid-century. The figure therefore demonstrates three historical characteristics of intergenerational co-residence: first, the majority of women born in 1870 or before never experienced all of their children leaving home; second, the majority of women born in 1900 or later had all children leave home before they reached age 70; third, the women in the transitional cohorts experienced the bulk of the shift toward generationally-separate living between 1940 and 1950, regardless of their age in that decade.5

The sharp mid-century shift in living arrangements is linked to changes in the timing of reproduction and the timing of children’s home-leaving, and may also be related to changes in male mortality. Analysis of IPUMS samples reveals that women born between 1831 and 1860 were, on average, 35 years old at the birth of their last child, those born between 1881 and 1910 were 33, and those born between 1921 and 1950 were 31. Meanwhile, the singulate mean age at marriage (calculated from IPUMS for men and women together) dropped from 25 in 1900 to 21 in 1950, implying that the typical age of the mother at her last child’s marriage fell from 61 to 55. Increases in male life expectancy meant that a woman in the earliest cohort would, on average, have been widowed at age 60 in 1900, before the expected marriage of her last child, but one in the middle cohort would have been widowed at age 64 in 1950, well after the expected marriage of her last child.6 Therefore, if widowhood had once deterred the last child’s departure, it now occurred after that last child had left.

Moving beyond these long-term demographic shifts, a series of multivariate logistic regression models assesses which factors determined whether a mother whose children had all reached age 18 lived apart from them at the 1900, 1940, 1960, and 1990 censuses. Results of Model 1 (Table 3) indicate that in 1900 (and in results of a model not shown for 1910) advancing age fails to increase the likelihood that a woman will live without any child; by 1940, however, older women are, in general, more likely to live away from their children. The IPUMS model also confirms the expected strong negative relationship between fertility and likelihood of living away from all children. Women with more children are always less likely to have had all of them leave home, and the magnitude of the coefficients increases somewhat over time. Relative to the reference woman—for whom all variables are at their reference level (age 55–59, two children, married, native-born, does not live on a farm, has not completed high school, is not employed, and reports no income)—increasing the number of children from two to four or more always decreases the probability of living away from all children by at least 0.11.

Table 3
Model 1, IPUMS analysis—results of logistic regression analysis.

Widows are always less likely to live apart from their children than are married women, as are divorced women through 1960 (and in a 1970 model not shown). Foreign-born women are consistently less likely to live apart from children. Women on farms are less likely to live away from their children in the early 20th century, but the effect diminishes over time; by 1970 they have become more likely to live away from their children. Higher socioeconomic status, as indicated by education, income, and employment, is generally linked to an increased likelihood that a mother will live away from all children. By 1990, however, income and employment were no longer statistically significant, and the effect of high school completion had diminished.

PSID data for 1968 to 2009 permit a closer, longitudinal analysis of household arrangements in the period in which the standard of separate households has been firmly established. Table 4 displays the results of Model 2, the single-risk event-history model. This analysis demonstrates that a mother’s age has very little effect on the likelihood that her last child will leave home in any wave of the survey. However, as suggested above, women at the oldest ages experience the lowest hazard of having their last child leave, since most are women who will never have all children leave. Age at last birth also has very little effect.

Table 4
Model 2, PSID—results of single-risk discrete-time event-history analysis.

Widowhood does not affect the likelihood of having the last child leave, and divorced women have an increased risk only in the second time period. Foreign women exhibit a lower risk in the final period. High school completion has no significant effect, while higher household income increases the probability of the last child’s departure in the first and third periods. In general, few of the variables that show an effect in the IPUMS analysis reach significance in the event history model, as the PSID analysis includes children’s characteristics, which have a much stronger bearing on mothers’ living arrangements. Women whose youngest co-resident children are daughters are more likely to have them leave in the first time period only, and those whose youngest co-resident children are employed are more likely to have them leave in the first two time periods. Though few mothers have married children in their homes at any wave of the PSID, those who do are much more likely to have that child leave by the next wave in all three time periods.

In Model 3, we examine simultaneously the effects of mother and child characteristics on the likelihood that a woman’s last child will leave home for one of three destinations—college, marriage, and employment. Results are displayed by departure type in Tables 5a–c for each time period.7 The only destination for which mothers’ characteristics have consistent influence is college. Mothers who are younger at the beginning of the episode but were older at last birth are more likely to see their last child leave home for college in all three time periods. Moreover, across the three time periods, higher socioeconomic status of mothers—as indicated by education and income—is generally linked to having the last child depart for college. Children’s characteristics have no bearing on this outcome.

When the destination of interest is marriage, the significance of children’s characteristics returns. There is no consistently significant effect of mothers’ characteristics across the three time periods. However, having one’s youngest co-resident child employed consistently predicts that the mother will no longer have a child in the home at the next wave of the survey, and that the last child will have departed for marriage. In the first two time periods, having a daughter as the youngest child also increased the likelihood that a mother would experience this outcome.

Modeling the final destination, employment, reveals that children’s characteristics again dominate, with employment and/or marriage in the episode before departure appearing as significant factors in each time period. For the mother, a foreign background does make it less likely that she will experience this outcome in the first and last periods, and divorce makes it more likely in the first and second.


Our results suggest that, while many of the factors discussed in the literature had bearing on the twentieth-century change in older women’s living arrangements, they were intertwined in ways not addressed by studies that deal with each factor individually. In census data between 1900 and 1990, having fewer children, being married, being native born, and having higher socioeconomic status were consistently associated with living away from all children. At the same time, broad demographic, social, and economic trends resulted in lower fertility, reduced incidence of widowhood, reduced rates of immigration, higher educational attainment, increased labor force participation, and higher incomes for middle-aged women, suggesting that much of the aggregate change in living arrangements during the 20th century resulted from these shifts in the characteristics in the population of mothers aged 40–69.

In this group, the proportions married, with two children, and of native birth increased gradually and modestly over the twentieth century. Shifts were more imposing and more concentrated at mid-century for proportion employed, high-school education, and higher-income, pointing to these as the best candidates to explain the sharp upward tick in women living away from children at mid century. Between 1940 and 1960 the proportion of women with a high school diploma increased from 18% to 29%, the proportion employed increased from 11% to 47%, and mean husband-wife joint income rose in real terms from just under $7,000 to $26,000. These abrupt changes likely drove much of the overall increase in the proportion of middle-aged women living away from all of their children.

Although secular change in the demographic and economic characteristics of mothers is clearly at work, it does not tell the whole story. Controlling for all of the factors in the IPUMS data still leaves a large unexplained increase in the propensity toward separate households at mid-century. Though the reference woman’s characteristics are held constant over the period, the probability that she will live away from all of her children (calculated from the exponentiated model intercept) rises dramatically from .54 in 1940 to .66 in 1950 and to .78 in 1960. Thereafter, it remains at about that level for the rest of the century. This finding indicates that even women who did not share in the rapid increases in employment, education, and income became more likely to live away from their children during this transitional period.

The sharp discontinuity at mid-century suggests that factors not accounted for in the cross-sectional logistic regression models are at play. We argue that drops in the age of mothers when their last child comes of age, resulting both from long-term decreases in the age at last childbirth and an abrupt mid-century decline in the age at which children married, provided the demographic foundation for separate households. And we infer, from other studies and from our analysis of the PSID, that economic conditions at mid-century made such separation feasible. Increasing income for older persons is one part of a general story of rapidly rising affluence for all in the postwar period.8 Analysis of IPUMS data reveals that, in real (1999) dollars, the wages of men aged 18–29 reporting income more than doubled during the critical period, rising from $9,000 in 1940 to over $19,000 in 1960, while those of their female counterparts rose from $6,500 to $10,000.9 Such income increases facilitated younger and more universal marriage and home leaving. While 36% of men and 53% of women in this age group were married in the 1940 census, 55% of men and 68% of women were married in 1960.

Writing in 1964, demographer Clyde Kiser pointed to earlier and more universal marriage, concentration of childbearing in the first decade of marriage, and the re-entry of wives into the labor force after family completion as important and widespread postwar changes in the white American family.10 Though Kiser did not comment on the effects of these changes on the living arrangements of older women, the earlier marriages of young adults at mid-century would have had consequences both for their parents’ generation and for their own: first, as children, they would have left their parents’ homes earlier in the parents’ lives, producing the sudden transition in living arrangements observed at mid-century; second, their early childbearing meant that, as adults, their children reached home-leaving age earlier in their own lives, perpetuating the separate household pattern. Since PSID data show that the employment of a child strongly predicts his or her marriage, it seems evident that the availability of higher-wage jobs after the Second World War promoted early marriage and contributed to adult children leaving home.

Event-history models with PSID data allow us to explore longitudinally and in more detail the experience of these later cohorts—those marrying at and after mid-century. Mother’s characteristics had few consistent effects on the overall risk of the last child leaving home. Higher income made a woman more likely to experience her last child leave in any given wave of the survey, but the magnitude of the effect is small. Instead, children’s characteristics dominated: mothers whose youngest co-resident children were employed or married at one wave of the PSID were much more likely to live away from all children by the next wave. A child’s marriage appears to have strongly motivated parents and children to seek separate living arrangements, and a child’s employment facilitated the move out of the mother’s house.

Our competing risks model adds nuance to this story, demonstrating that although child characteristics are the most important determinants of when a woman’s last child will leave home, her own characteristics have bearing on why that child leaves home, and specifically whether her last child will leave for college, an outcome more commonly experienced by women of higher socioeconomic status. Younger mothers are also at a higher risk of experiencing this outcome because children leaving for college do so at younger ages than those destined to go elsewhere.

Child characteristics have no bearing on whether a mother’s last child will leave for college, but the employment and marital status of a woman’s youngest child strongly predict the other outcomes. Mothers whose youngest co-resident child was a daughter were more likely to have their last child leave home for marriage in the first two time periods and less likely to have her leave for employment in the second, suggesting that, even into the 1990s, marriage was still the most readily-accessible ticket out of the parental home for daughters.


This paper has examined the dramatic change in the living arrangements of middle aged white women across the 20th century, asking why women who have children now spend their later years living away from them, reversing the conditions that obtained a century ago. Analysis of microdata samples from the decennial censuses of 1900–1990 compared the characteristics of mothers who did and did not live with at least one adult child; longitudinal data for the years 1968–2009 permitted analysis of both mothers and children. This series also allowed analysis of the competing risks that a child would leave home for college, marriage, or employment and other destinations.

Together, the two sources of data reveal the intersection of demographic and economic change across the twentieth century in this major shift in older women’s living arrangements. Demographic conditions are ascendant in the analysis: the potential for separate households was clearly enhanced by the concentration of childbearing earlier in a mother’s life. It was increasingly likely that a woman’s last child would reach adulthood well before she was widowed, and between 1940 and 1960 there was a particularly sharp drop in the singulate mean age at marriage for this woman’s children, further increasing the likelihood that children would marry and leave the household well before their mothers were widowed. Census data from across the century and PSID data for 1968–2009 indicate that mothers were always very unlikely to live with married children. A decline in immigration over the period also had an effect on living arrangements, as both datasets indicate the propensity for mothers with foreign backgrounds to continue to live with their children. While this may be a cultural effect, other studies of immigrant household arrangements have found economic factors to be at least as relevant as cultural ones (Gratton and Gutmann 2010).

In general, demographic markers were not independent of economic ones. The evidence for 1968 to 2009 demonstrates the salience of children’s employment for both their marriages and their departure from their mothers’ homes. This result strongly suggests that children’s economic opportunities had an effect in earlier periods, though we cannot measure them in the long-term census data. The remarkable expansion of the American economy between 1940 and 1970 had a particularly positive effect for young workers, providing new employment and higher income. Children could afford to marry earlier and set up new households. Their parents’ rising income and access to public benefits also meant less demand on children to contribute to their parents’ maintenance. Mothers’ labor force participation likely sustained separate households, an economic shift that was itself facilitated by a demographic one: earlier cessation of childbearing.

Although we cannot test cultural factors directly, normative change may play a role in perpetuating generationally separate households. Throughout the century, children nearly always left their parents’ homes upon marriage if they had not already done so; after 1960, studies show that young adults became much more likely to leave home prior to marriage (Gutmann et al. 2002). Therefore, even as the singulate mean age at marriage increased rapidly toward the end of the century, men and women continued to leave their mothers’ homes in early young adulthood. The proportion of young persons attending college has steadily increased since 1950 and continuing, if now faltering, employment opportunities for both mothers and children also help make separate residence likely and possible. These changes now undergird a system in which a mother can expect that all of her children will leave home by the time she reaches middle age. Though the social scientists who examined this trend away from co-residence in the early postwar period initially anticipated that the loss of residential companionship may have a negative impact on the well-being of older women, this study indicates that the trend was instead motivated by the increased well-being of both mothers and their children, and by the erosion of the need of either generation for residential support. The findings of this study therefore suggest that only severe economic distress, such as a steep decline in work opportunities for young adults, is likely to undo this now-expected separation of the generations.


This research has been supported in part by Grant Number R01AG020705 from the United States National Institute on Aging, Principal Investigator, XXX. The National Science Foundation of the United States provided support to XXX for work on this article. Any opinion, finding, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. We thank J. David Hacker for assistance with the cohort fertility analysis, Catherine Fitch and Marianne Wanamaker for valuable comments and insights offered at the 2010 meeting of the Population Association of America, Steve Ruggles for critique and helpful suggestions at the 2010 meeting of the Social Science History Association, Jeremy Albright for modeling advice, Glenn Deane for guidance in interpreting the event history models, and Peter Van Cleave for assistance with the text.


1Episodes represent one person-year in the 1968–1979 and 1980–1996 panels and two person-years in the 1997–2009 panel. Were the effects of the independent variables to remain constant between panels, doubling the risk time in the third panel would change the intercept (roughly doubling the odds, or the exponentiated intercept), but not the coefficients. However, doubling the risk time also doubles the time between when independent variables are measured and when outcomes are measured, which may bias coefficient estimates, particularly for the time-varying covariates (e.g, child employment and marital status). Changes in the magnitude of coefficient estimates from the first and second periods to the third are likely an artifact of the doubling of the risk time rather than a real change in magnitude.

2While certainly possible, we assume that mothers are not at risk of having the last child leave as part of the conventional home-leaving process before the youngest child is 16.

3Censoring means that the woman is removed from the risk set despite not having experienced the event in question (see Alison [1982] and Alter et al. [2012]). A woman is included in the analysis for the episode between t0 and t1 only when she is observed with a child in the home at t0, and observed in the PSID—either with a child in the home or not—at t1. If a mother has a child in her house at one wave (t0), but has turned 70, died, or been lost to follow-up before the next wave (t1), then she is not included in the analysis during the episode between t0 and t1. We do not expect attrition from the sample to bias our study because it is unlikely that the loss of an entire family to follow up would be related to the departure of the last child from the parental home.

4We do not include separate variables for more than one co-resident child or any variables that might be correlated with number of children in the household (for example, sex balance of co-resident children). Women with more than one co-resident child at the beginning of an episode are much less likely to end the episode with no children in the house, and coefficients for such variables as sex balance would proxy number of children. We assume that the youngest child will likely, though not necessarily, be the last to leave.

5In 1950, the Census stopped including children away at college as part of their parents’ households (Ruggles and Brower 2003). The sudden mid-century transition in living arrangements, however, is not simply an artifact of this change in reporting. Rates of intergenerational co-residence for women aged 50–69 in 1880–1940 would decrease by no more than two percentage points in any census year if children aged 18 or older who reported attending school were removed from the households.

6Since we are dealing only with men and women who lived long enough to marry and bear children, we calculate husbands’ expectation of life as ex from the appropriate life table, where x is the age men were, on average, when their wives were 30 years old for each cohort, as estimated from IPUMS samples (ages 35, 34, and 30, respectively). We calculate women’s expectation of life as 30+e30 from life tables as follows: 1830–1900 Hacker (2010), 1909–1931 Hill (1936), 1939–1941 Greville (1947), and 1949–1981 U.S. Department of Health (1949–1951; 1959–1961; 1969–1971; 1979–1981). All except Hacker are available at http://www.cdc.gov/nchs/products/life_tables.htm.

7Models were estimated simultaneously for all destinations and separately for each time period; for ease of discussion, results for each destination in all time periods are displayed in separate tables.

8Changes in the income of mothers are, admittedly, imperfectly measured: the model for 1940 uses wage income (including that of husband, if married), that for 1950 omits income altogether, and those for 1960 and later use total income (including that of husband, if married). These discrepancies may account for some of the unexplained change between models.

9Although incomes increased for all workers at mid-century, those for younger workers increased most dramatically: “The real median income of families across the 13-year period between 1947 and 1960 rose by 38%…but those headed by persons aged 25 to 34 increased by nearly 50%.” http://www.census.gov/hhes/www/income/data/historical/families/

10American Philosophical Society, records of the American Eugenic Society, Box 16, Clyde Kiser, “Types of Demographic Data Useful for Genetics.”

Contributor Information

Emily R. Merchant, ICPSR and Department of History, University of Michigan.

Brian Gratton, Faculty of History, Arizona State University.

Myron Gutmann, ISR, Department of History, and School of Information, University of Michigan, National Science Foundation.


  • Allison PD. Discrete-time methods for the analysis of event histories. Sociological Methodology. 1982;13:61–98.
  • Allison PD. Survival analysis using the SAS system: A practical guide. Cary: SAS Institute; 1995.
  • Alter GC, Gutmann MP, Leonard SH, Merchant ER. Introduction: Longitudinal analysis of historical-demographic data. Journal of Interdisciplinary History. 2012;42(4):503–517.
  • Bethencourt C, Rios-Rull JV. On the living arrangements of elderly widows. International Economic Review. 2009;50(3):773–801.
  • Buck N, Scott J. She’s leaving home: But why? An analysis of young people leaving the parental home. Journal of Marriage and Family. 1993;55(4):863–874.
  • Burr JA, Mutchler JE. Race and ethnic variation in norms of filial responsibility among older persons. Journal of Marriage and Family. 1999;61(3):674–687.
  • Costa D. A house of her own: Old age assistance and the living arrangements of older nonmarried women. Journal of Public Economics. 1999;72:39–59.
  • Crimmins EM, Ingegneri DG. Interaction and living arrangements of older parents and their children: Past trends, present determinants, future implications. Research on Aging. 1990;12:3–35. [PubMed]
  • Feldstein Ruth. Motherhood in black and white: Race and sex in American liberalism, 1930–1965. Ithaca: Cornell University Press; 2000. [PubMed]
  • Glick JE, Van Hook J. Parents’ coresidence with adult children: Can immigration explain racial and ethnic variation? Journal of Marriage and Family. 2002;64:240–253.
  • Glick PC. Updating the life cycle of the family. Journal of Marriage and Family. 1977;39(1):5–13.
  • Glick PC, Parke R. New approaches in studying the life cycle of the family. Demography. 1965;2:187–202.
  • Goldscheider FK, Bures RM. The racial crossover in family complexity in the United States. Demography. 2003;40(3):569–587. [PubMed]
  • Gonzales AM. Determinants of parent-child coresidence among older Mexican parents: The salience of cultural values. Sociological Perspectives. 2007;50(4):561–577.
  • Gratton B, Gutmann MP, Skop EH. Immigrants, their children, and theories of assimilation: Family structure in the United States, 1880 to 1970. History of the Family. 2007;12(3):203–222 . [PMC free article] [PubMed]
  • Gratton B, Gutmann MP. Emptying the nest: Older men in the United States, 1880–2000. Population and Development Review. 2010;36(2):331–356. [PubMed]
  • Greville Thomas NE. United States life tables and actuarial tables 1939–1941. Washington, DC: U.S. Government Printing Office; 1947.
  • Gutmann MP, Pullum-Piñón S, Pullum TW. Three eras of young adult home leaving in twentieth-century America. Journal of Social History. 2002;35:533–576.
  • Hacker J David. Decennial life tables for the white population of the United States, 1790–1900. Historical Methods. 2010;43(2):45–79. [PMC free article] [PubMed]
  • Hill Joseph A. United States life tables. Washington, DC: U.S. Government Printing Office; 1936.
  • Kertzer D. Toward a historical demography of aging. In: Kertzer D, Laslett P, editors. Aging in the past: Demography, society, and old age. Berkeley: University of California Press; 1995. pp. 363–385.
  • Kobrin F. The fall of household size and the rise of the primary individual in the U.S. Demography. 1976;13(1):127–138. [PubMed]
  • Kotlikoff LJ, Morris JN. Why don’t the elderly live with their children? A new look. In: Wise DA, editor. Issues in the economics of aging. Chicago: University of Chicago Press; 1990. pp. 149–172.
  • Kramarow E. The elderly who live alone in the United States: Historical perspectives on household change. Demography. 1995;32(3):335–352. [PubMed]
  • Levy M. Aspects of the analysis of family structure. In: Coale AJ, editor. Aspects of the analysis of family structure. Princeton: Princeton University Press; 1965. pp. 1–63.
  • McGarry K, Schoeni RS. Social Security, economic growth, and the rise in elderly widows’ independence in the twentieth century. Demography. 2000;37(2):221–236. [PubMed]
  • Michael R, Fuchs V, Scott S. Changes in the propensity to live alone, 1950–1976. Demography. 1980;17(1):39–58. [PubMed]
  • Newman K. The accordion family: Boomerang kids, anxious parents, and the private toll of global competition. Boston: Beacon Press; 2012.
  • Pampel FC. Changes in the propensity to live alone: Evidence from consecutive crosscultural surveys. Demography. 1983;20(4):433–447.
  • Parsons T. The social structure of the family. In: Anshen RN, editor. The family: Its function and destiny. Oxford: Harper; 1949. pp. 173–201.
  • Plant RJ. Mom: The transformation of motherhood in modern America. Chicago: University of Chicago Press; 2010.
  • Ruggles S. The transformation of American family structure. American Historical Review. 1994;99(1):103–128.
  • Ruggles S. Multigenerational families in nineteenth-century America. Continuity and Change. 2003;18:139–165.
  • Ruggles S. The decline of intergenerational coresidence in the United States, 1850–2000. American Sociological Review. 2007;72:964–989. [PMC free article] [PubMed]
  • Ruggles S, Brower S. Measurement of household and family composition in the United States, 1850–2000. Population and Development Review. 2003;29(1):73–101.
  • Ruggles S, Goeken R. Race and multigenerational family structure. In: South SJ, Tolnay SE, editors. The changing American family: Sociological and demographic perspectives. Boulder: Greenwood; 1992. pp. 15–42.
  • Ruggles S, Alexander JT, Genadek K, Goeken R, Schroeder MB, Sobek M. Integrated Public Use Microdata Series: Version 5.0 [Machine-readable database] Minneapolis: University of Minnesota; 2010.
  • Sarkisian N, Gerena M, Gerstel N. Extended family integration among Euro and Mexican Americans: Ethnicity, gender, and class. Journal of Marriage and Family. 2007;69(1):40–54.
  • Smith DS. Accounting for change in the families of the elderly in the United States, 1900-present. In: Van Tassel D, Stearns PN, editors. Old age in a bureaucratic society: The elderly, the experts, and the state in American history. New York: Greenwood; 1986. pp. 87–109.
  • Smits A, van Gaalen RI, Mulder CH. Parent-child coresidence: Who moves in with whom and for whose needs? Journal of Marriage and Family. 2010;72:1022–1033.
  • Stanger-Ross J, Collins C, Stern MJ. Falling far from the tree: Transitions to adulthood and the social history of twentieth-century America. Social Science History. 2005;29(4):625–648.
  • Stettersten RA., Jr A time to leave home and a time never to return? Age constraints on the living arrangements of young adults. Social Forces. 1998;76(4):1373–1400.
  • Treas J, Bengtson VL. The demography of mid- and late-life transitions. Annals of the American Academy of Political and Social Science. 1982;464:11–21.
  • United States Department of Health, Education, and Welfare. Life tables for 1949–51, 1959–61, 1969–71, 1979–81. http://www.cdc.gov/nchs/products/life_tables.htm.
  • Veevers JE, Gee EM, Wister AV. Homeleaving age norms: Conflict or consensus? International Journal of Aging and Human Development. 1996;43(4):277–295. [PubMed]
  • Ward R, Logan J, Spitze G. The influence of parent and child needs on coresidence in middle and later life. Journal of Marriage and Family. 1992;54(1):209–221.
  • Weiss J. To have and to hold: Marriage, the baby boom, and social change. Chicago: University of Chicago Press; 2000. [PubMed]
  • Wilmoth JM. Living arrangements among older immigrants in the United States. The Gerontologist. 2001;41:228–238. [PubMed]
  • Wolf DA, Soldo BJ. Household composition choices of older unmarried women. Demography. 1988;25:387–403. [PubMed]
PubReader format: click here to try


Related citations in PubMed

See reviews...See all...


  • PubMed
    PubMed citations for these articles

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...