Predictors of Stability and Change in Private Safety Nets of Unmarried Mothers

Abstract

Although the importance of social supports for single mothers in times of crisis is widely recognized, little is known about the stability of such “private safety nets” over time, as children age and maternal and household characteristics change. This study uses multilevel models and 4 waves of data from the Fragile Families and Child Wellbeing Study to describe trajectories of social support perceptions for 3,065 unmarried mothers. Results suggest that, following a birth, most unmarried mothers perceive the availability of support, but these support perceptions disintegrated somewhat in subsequent years. Mothers who appeared to have the greatest need for support—those without stable employment or a stable partner—experienced more rapid deterioration of their perceived safety nets than more advantaged mothers. Future research should examine network composition and conditions for support provision among the most vulnerable single mothers and consider how safety net stability influences maternal and child health and well-being.

The 1996 passage of the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA), described by Haskins (2001: 264) as the nation’s renegotiation of its social contract with the poor, formalized a new public philosophy of individual and family self-sufficiency. PRWORA replaced the federal entitlement program known popularly as “welfare” or more formally, Aid to Families with Dependent Children (AFDC), with Temporary Assistance to Needy Families (TANF). As its name implies, TANF benefits are time-limited and, in keeping with the theme of personal responsibility, contingent on recipients’ finding and keeping employment. The putative goal of TANF was to empower poor families by providing training, childcare assistance, and other resources necessary for them to find employment and attain self-sufficiency.

Evaluations of the program by researchers and the press suggest that TANF has fallen far short of this goal. For example, a multi-year, ethnographic investigation of Oregon welfare agencies and their clients observed that bureaucratic valorization of employment leads recipients to accept any paid position in order to retain their benefits—regardless of pay, transportation costs, or compatibility with child care schedules and other family needs (Morgen, Acker & Weigt, 2009). Urban Institute researchers analyzed data following welfare leavers and found that about one-fifth of families who exited TANF in the early 2000s left without employment, a working spouse, or any form of cash assistance (Acs & Loprest, 2007). TANF caseloads have declined since welfare reform, although poverty has risen. Nationally, while 68% of poor families received cash assistance in 1996, only 27% of poor families did so in 2009 (Pavetti, 2011).

Federal estimates indicate that 18% of families with children lived at or below the poverty level in 2010 and that more than half of these families were headed by unmarried women (U.S. Census Bureau, 2011). How have these low-income families survived the unraveling of the public safety net in the TANF era? Ethnographic accounts describing the precarious existence of low-income, single-mother families in the United States have long pointed to their reliance on instrumental and material support from family and friends (Edin & Lein, 1997; Stack, 1974). In recent years, a growing body of research has quantified the importance of these “private safety nets” or perceived access to sources of instrumental, material, and financial support in case of an emergency (Harknett, 2006). This work generally confirms the value of private safety nets for low-income mothers’ survival particularly in the era of welfare reform and a limited public safety net (i.e., TANF, food stamps, Medicaid). Poor and near-poor mothers who perceive that they have support should they need it fare better than their peers, reporting higher rates of employment, less financial hardship, and less reliance on public assistance (Harknett, 2006; Henly, Danziger & Offer, 2005). Several studies suggest that the children of low-income mothers with strong safety nets also benefit, showing greater socio-emotional adjustment and greater cognitive development than the children of mothers with more fragile safety nets (Jackson, Brooks-Gunn, Huang, & Glassman, 2000; Ryan, Kalil, Leininger, 2009).

In light of accumulating evidence of the beneficial effects of support perceptions, some recent research has examined support correlates. These studies point to a negative association between need and support perceptions: those most likely to need support, including single mothers, and immigrant and low-income parents, are least likely to perceive its availability (e.g., Harknett & Hartnett, 2011; Turney & Kao, 2009). In this paper, we extend this recent work by considering how individuals’ perceptions of support availability change over time, an issue of considerable importance given the time-limited nature of public assistance. Focusing on single mothers, we address two specific, related questions. First, we ask how mothers’ perceptions of support from friends and families change over time, following a child’s birth when perceptions are likely at their highest levels. Here, we consider the persistence of support perceptions: which mothers perceive a safety net consistently in the years subsequent to their child’s birth, and which mothers’ perceptions change? Second, we address variation in the nature and rate of change in support perceptions, with the aim of identifying which correlates of initial support perceptions gain or lose salience over time. For example, do mothers who start out with the fewest resources also experience the most rapid drop-off in support perceptions? Or are the support perceptions of the neediest single mothers more resilient than those who have more personal resources to fall back on? By identifying trajectories of perceived support and the variables that shape these trajectories, we can identify vulnerable families and consider the best role for public safety net programs.

Literature Review

Single mothers are a vulnerable population with much to gain from a personal safety net. Using a nationally-representative sample of urban single mothers, Teitler and colleagues (2004) exposed single mothers’ financial susceptibility during their children’s first year and revealed just how much they turn to others for help. In terms of public safety net programs, 83% of single mothers used the Women, Infants, Children (WIC) nutritional program, 70% used Medicaid, 48% used Food Stamps, and 33% used TANF. Most single mothers also utilized personal safety nets as 86% relied on their children’s fathers and 64% relied on family and friends. Public and private safety nets are not enough in most instances, however. A full 63% of single mothers experienced a material or medical hardship during their children’s first year, the most common of which was borrowing money to pay bills (30%) (Teitler et al., 2004).

Although actual support receipt demonstrates single mothers’ high level of need and suggests the likelihood that they will call on others for assistance, the current analysis examines support perceptions. As Harknett (2006) observed, support perceptions capture access to a potential safety net; this potential, although intangible and perhaps unrealizable, is unequivocally desirable (Harknett & Hartnett, 2011). Realized support, although more easily quantified, obscures the distinction between the individual’s need and her network’s resources. Moreover, realized support typically is time-delimited, while support perceptions may fluctuate over time as situational contexts change and as network relationships evolve (Swartz, 2009). Despite the intangibility of perceived support, a substantial literature attests to its beneficial effects for individual well-being: those who have others to call on in times of need fare better mentally, emotionally, physically, and financially (Harknett, 2006; Henly et al., 2009; House, Umberson, & Landis, 1988; Sarason, Sarason, & Pierce, 1990). Indeed, Wethington and Kessler (1986) found that in a national survey of adults, perceived support is more closely tied to psychological health and wellbeing than received support. Specifically, among respondents who had experienced a recent stressful event, the degree to which respondents felt they had someone they could count on for help was more closely associated with later distress than respondents’ receipt of actual supports.

Prior studies indicate that most single mothers believe they have access to at least some assistance should they need it. In their longitudinal analysis of Michigan TANF recipients, for example, Henly and her colleagues (2005) reported an average support score of 0.86 (on a scale of 0 to 1) at baseline and 0.83 two years later. Similarly, Ciabattari (2007) found that a nationally representative sample of single mothers of one-year olds averaged 3.2 of a possible four points on a scale of perceived material supports. Harknett (2006) observed that less than one-fifth of the low-income mothers in her three-county sample said they lacked one or more forms of support.

These numbers suggest both that most single mothers have access to a private safety net and that perceptions of support are not universal. The research literature provides insight into why support perceptions vary. Not surprisingly, mothers’ support perceptions appear to be linked to their integration into a network of family members and friends. Household composition is one indicator of social integration. The presence of other adult household members—such as a partner, parent, or other relative—may provide financial, instrumental, or emotional resources, enhancing a mother’s sense of support (Harknett, 2006; Nichols, Elman, & Feltey, 2006; Park, 2005). In a recent study, Harknett and Hartnett (2011) found that marriage and cohabitation were associated with higher rates of perceived support. Participation in a religious community also is associated with support, both perceived and realized. Ellison and George (1994), for example, found that individuals who attended religious services frequently have larger social networks, more contact with network members, and receive more social support. In terms of tangible assistance, each incremental increase in religious attendance increased the odds of receiving gifts or presents from friends and relatives by 31%.

Although the ethnographic literature describes the importance of support networks for the survival of disadvantaged families, the quantitative evidence suggests that the most vulnerable are the least likely to believe that they have access to a private support network. Multiple studies have reported negative associations between single mothers’ support perceptions and various indicators of economic status and human capital, including public assistance receipt and unemployment (Harknett, 2006; Harknett & Knab, 2007; Harknett & Hartnett, 2011; Henly et al., 2005). A possible explanation for this pattern is the strong norm of reciprocity tied to participation in a support network: drawing on support from family members and friends obligates one to provide future support (Pearlin, 1985). Women who feel unable to reciprocate may self-select out of support networks in order to avoid the obligations that network membership may entail. A second explanation for the negative association between socioeconomic indicators and access to private support networks is rooted in the homophilous nature of social networks. Because social networks tend to be homogeneous along multiple dimensions, including social and economic characteristics (McPherson, Smith-Lovin, & Miller, 2001), disadvantaged mothers’ family and friends may themselves lack the capacity to provide material support to other network members.

Single mothers’ perceptions of social support also are correlated with their demographic characteristics and the characteristics of their children. Support levels tend to be higher for younger mothers and for new mothers than for mothers who are older or who have other children (Belsky & Rovine, 1984; Turney & Kao, 2009). This finding suggests the importance of network members’ perceptions of a mother’s need in determining the level of support they are willing to provide. At the same time, however, mothers who are in poor health or disabled or who have children with health issues perceive lower levels of support (Harknett & Hartnett, 2011). Perhaps, then, network members’ assessments of mothers’ needs are offset by other factors that are not readily observable in quantitative research. Alternatively, mothers who have health issues themselves or are caring for ill or disabled children may find it difficult to maintain the relationships necessary to build strong support networks.

Two individual characteristics that may serve as broad markers of other traits are mothers’ race/ethnicity and nativity status; both are correlated with support perceptions, although the reasons for these correlations remain unclear. Ethnographic studies generally highlight the strong kin networks available to mothers of color and to immigrant parents (e.g., Keefe, 1996; Mirande, 1985; Stack, 1974), but the findings of quantitative studies are mixed. Some studies suggest that Whites have stronger kin support networks (e.g., Brewster & Padavic, 2002; Hogan, Eggebeen, & Clogg, 1993) and others suggest that minority families do (e.g., Gerstel, 2000; Hogan, Hao, & Parrish, 1990). In their recent work on support perceptions among the parents of kindergartners, Turney and Kao (2009) observed that parents’ support perceptions vary by support type and by race/ethnicity and nativity status. Overall, they found that White native-born parents perceived greater support; yet, minority and immigrant parents more often reported having someone available to watch a young child or to provide emotional support.

Change versus Stability in Support Perceptions

Although the birth of a child is in most cases a celebrated event, it invariably occasions change in family life, disrupting sleep patterns, daily routines, and family members’ normal roles (Antonucci & Mikus, 1988). Along with these typically short-run changes are other, frequently longer-term effects of an additional family member, including increased monetary expenses, job disruption and its associated loss of income, and even changes in the nature of established relationships (Cowan & Cowan, 2000). Because they may lack the emotional and financial support of a co-resident parent, single mothers are particularly vulnerable to the short-run and longer-term stressors associated with an additional family member. Their higher rates of exposure to chronic strains and more frequent “life events” (e.g., job loss, residence change) leave single mothers with higher rates of distress and adversely impact their parenting abilities (e.g., Jackson et al., 2000; McLanahan & Booth, 1989).

The importance of social support in helping new parents cope and the dynamic nature of this support has been well-documented in studies of married couples. Around the time of a birth, family members and close friends generally step up to help the new parents, providing emotional, instrumental, and material support. In their longitudinal analyses of new parents, Belsky and Rovine (1984) observed increased contact with family members in the immediate post-partum period, as did Bost and her colleagues (2002). Indeed, married parents enjoy a “support bubble” that begins to expand shortly before the birth of their first child and continues to grow over the child’s first three months. This bubble subsequently deflates, as friends and family members perceive that the new parents have adjusted to their new responsibilities (Belsky & Rovine, 1984; Bost et al., 2002; Gameiro et al., 2010).

We expect that single mothers, too, experience heightened support around the time of the birth. As the excitement following the birth begins to fade, however, and the mother recovers from the physical demands of pregnancy and parturition and adjusts to the new demands on her time and energy, support from family and friends may fall off as it does among married couples (Bost et al., 2002; Belsky & Rovine, 1984; Gameiro et al., 2010). It is also possible that single mothers’ safety net trajectories do not follow those of their married counterparts. The norms of support provision may differ when mothers are single. A father’s absence could mean that family members and friends extend a safety net to single-mother families for a longer period.

Not all single mothers perceive the support of a private safety net at the time of a child’s birth; in light of the negative association between need and support, the experiences of mothers who lack a safety net are of particular interest. It may be that their safety net perceptions remain consistently low as time passes because their networks lack the capacity to provide support or because of the obligation of reciprocity entailed by participation in a support network, as noted above. Circumstances change, however—human capital increases, relationships evolve—raising the possibility that women who lack a safety net at their child’s birth may gain social support in subsequent years.

Whereas our descriptive analyses consider the patterns of change and stability in single mothers’ perceptions of support, our multivariate analyses address the possibility of change in the correlates of support perceptions. The factors that shape women’s support perceptions may increase or decrease in salience, as their children grow and their personal circumstances change. The influence of economic vulnerability may wane over time, for example, if mothers’ social networks expand as their children enter preschool; alternatively, economic vulnerability may become increasingly determinative of support perceptions if the stress of long-term disadvantage corrodes social ties.

Research Hypotheses

Prior research with this data set and others leads us to anticipate that, at each observation point, a substantial majority of single mothers will report access to instrumental or material support in the event of an emergency. We also anticipate that the average level of support perceptions will deteriorate somewhat over time, as expressions of support from family members and friends diminish in the weeks and months following a birth. We do not expect to see a uniform disintegration, however; instead we anticipate that although safety net loss will be more common, at least a small share of single mothers will gain access to the support of a private safety net over the five year observation window. Such gains are most likely in response to changes in social relationships (e.g., marriage), although they may also be tied to change in economic circumstances (e.g., graduating from college). Finally, we expect that the effects of the covariates on support perceptions will be moderated by time, such that the rate of safety net deterioration will be more rapid among the most economically vulnerable.

Method

Data

This study uses longitudinal data from the Fragile Families and Child Well-being Study (FFCWS), which follows a cohort of 4,898 children born between 1998 and 2000 to predominately unmarried, low-income parents. These “fragile families” were selected from a stratified random sample of 20 cities drawn from all U.S. cities with 200,000 or more residents (see Reichman, Teitler, Garfinkel, & McLanahan, 2001 for details regarding the FFCWS study and sampling procedures). Mothers (and fathers when available) were interviewed initially within 72 hours of giving birth, in either English or Spanish. Interview topics included parents’ relationships with each other and with the child, living arrangements, work and financial status, parental and child health, and their general social and psychological well-being. Follow-up telephone questionnaires were conducted one, three, five, and nine years later, with less frequent in-home assessments. Mothers’ response rates were 86% for the baseline survey and 90%, 88%, and 87% for the follow-ups (see http://www.fragilefamilies.princeton.edu/study_design.asp). The analyses reported here use the baseline data and data from the one-, three-, and five-year follow-ups.

Our sample is restricted to the 3,709 mothers who were not married at the baby’s birth and who reported that the focal child lived with them all or most of the time at each subsequent interview (n = 3,665). Our descriptive and multivariate analyses excluded mothers missing data on the time-constant variables in the analysis (n = 469; 13% of eligible sample) and left-censored cases with missing data at Baseline (n = 48; 1%). We also excluded the small number of women who identified as other than Hispanic, non-Hispanic Black, or non-Hispanic White (n = 83; 2%), leaving us with a final sample size of 3,065.

Measures

Dependent variables

Our analyses address mothers’ perceptions of their ability to access financial and instrumental support as needed. Admittedly, perceptions may be inaccurate and more reflective of personality characteristics (e.g., optimism) than reality (Dunkel-Schetter & Bennett, 1990; Sarason, Pierce, & Sarason, 1990). Even so, perceptions remain important for maternal (Harknett, 2006) and child well-being (Ryan et al., 2009) and they generally have a stronger relationship to child and family well-being than other support measures (Sarason, Sarason, & Pierce, 1990).

Our measure of perceived support is based on mothers’ responses to a series of three questions about whether “if you needed help during the next year, could you count on someone to… Loan you $200? Provide you with a place to live? Help you with emergency child care?” Mothers who answered yes to all three questions were coded as 1, to indicate a full safety net; mothers who answered no to one or more of the questions were coded 0. Because the baseline survey’s wording of the child care assistance item differed (“help with babysitting or child care”) from that in the follow-up surveys, we conducted additional analyses to determine whether the wording change appeared to make a difference to the findings. The results, available on request, suggested that the wording change had no detectable impact.

In addition to the results for this global measure of a personal safety net, we also analyzed separate models for each of the three components of this safety net: loans, child care, and shelter. The relationships between the predictors and each component were consistent with respect to direction and strength, although the relationships tended to be slightly stronger for access to financial support (available on request). Given the strong similarity among models and our interest in evaluating whether single mothers have allies to call upon to meet various basic necessities, we present the models using the dichotomous measure of a full versus not full net.

Time-constant covariates

Mother’s age at the focal child’s birth is measured in years and mean-centered. We have a six-category measure of race-ethnic identity and nativity status, based on questions asked at the baseline interview about racial self-identification, Hispanic origin, and place of birth. This information allows us to distinguish six groups of mothers: non-Hispanic Black immigrant, non-Hispanic Black native born, Hispanic immigrant, Hispanic native born, non-Hispanic White immigrant, and non-Hispanic White native born. In the analyses, effects are measured relative to the reference category, non-Hispanic, White native born. In addition, we have a single indicator of the mother’s socioeconomic background, her mother’s (focal child’s grandmother’s) educational attainment, coded 1 for grandmothers who earned at least a high school diploma or its equivalent and 0 otherwise.

Time-varying and cumulative covariates

Maternal and household characteristics can change in the five years following a child’s birth and the longitudinal nature of the FFCWS allowed us to construct measures that capture such changes from one interview to the next. Many of the time-varying covariates are categorical, and we measure them cumulatively over successive interviews rather than measuring them as dummy variables at each time point, irrespective of previous values. This approach allows us to capture each predictor’s full range of change over time. The range of possible values increases at each time point, beginning with a zero-one dichotomy at baseline and expanding to a five-point range at the Year 5 follow-up, where a 0 indicates the continued absence of a predictor and 4 indicates its continued presence over the full observation period.

At each interview, mothers completed a household roster specifying the age, gender, and relationship of every person living with them. Using this data, we constructed two interview-specific dichotomies representing household composition: one indicating whether the mother reported living with a romantic partner and a second indicating whether she reported living with at least one of her child’s grandparents (her own parents or her child’s father’s parents). Summing the interview-specific dichotomies over time yielded two cumulative indicators of household composition: number of interviews in which the mother lived with a partner and number of interviews living with at least one of the child’s grandparents.

Our household membership indicators also include a cumulative measure of the number of biological children born to the mother following the birth of the index child. This measure was built from interview-specific dichotomous variables coded 1 if the mother had given birth since the last interview. In Years 1 and 3, mothers were asked this question directly; in Year 5, the question was not asked, so we constructed the variable using the ages of biological children in the household and interview dates. By Year 5, this variable ranges from 0 (no additional children) to 4 (a new baby at each follow-up).

Our indicators of mother’s economic status capture the effects of employment, poverty, and public assistance receipt. Following the FFCWS, we define employment as “regular” employment for pay lasting two weeks or more. Because prospective mothers often leave the labor market prior to a birth, we coded baseline employment status as 1 for mothers who reported employment within the year prior to the birth and 0 otherwise. At each subsequent wave, mothers were defined as currently employed if they reported working for pay at a regular job within the two weeks prior to the interview. We relied on the poverty indicator constructed by FFCWS researchers from household income and size to determine poverty status at each wave. Mothers who fell at or below 100% of the poverty level were coded as poor. To measure public assistance receipt at baseline, mothers were asked if they had received income from public assistance, welfare, or food stamps in the past year. At later waves, mothers described their use of TANF and food stamps since the last interview. Mothers who indicated that they had used either were coded as receiving public assistance.

We have two measures of mother’s human capital: education and health status. We measure maternal education with a set of dummy variables distinguishing mothers with a high school diploma or general equivalency diploma (GED) from those who did not attain either and from those who attained more formal schooling. Mothers were asked to rate their health status on a five point scale. We dichotomized the variable such that mothers with fair or poor health were distinguished from those with good, very good, or excellent health.

To measure their integration in a religion-based community, mothers were asked how frequently they attended religious services. Mothers were coded as 1 if they attended at least once weekly or more or 0 if they attended less than once weekly. Religious participation is measured cumulatively, so that by Year 5, its values range from a low of zero, indicating less than weekly service attendance throughout the observation period, to a high of 4, indicating service attendance at least once weekly at all interviews.

Controls

Our models control for characteristics of the focal child that may influence the willingness or ability of family and friends to provide support. These characteristics are child is mother’s first-born (coded 1, 0 otherwise), child’s gender (females coded 1, males coded 0) and low birth weight (less than 2,500 grams coded 1, 0 otherwise).

Analytic Strategy

We start with a statistical portrait of unmarried mothers at the time of their child’s birth, presenting their percentage distribution across the covariates at the baseline interview, and for the time-varying covariates, the percentage that experienced one or more status changes subsequent to baseline. Descriptive statistics were weighted using Stata’s svy commands, making them representative of all single mothers of the 1999 birth cohort living in U.S. cities of at least 200,000 persons.

We then turn to estimating multivariate models of change in support perceptions, using what Singer and Willett (2003) refer to as the multilevel model for change. In practice, we estimated models that have structural (i.e., fixed) and stochastic (i.e., random) components, using xtmelogit in Stata, Version 10. This approach required us to convert the record for each sample member to a person-year format to accommodate the time-varying outcome and the time-fixed and time-varying covariates. Each mother with complete data at Baseline contributed one person-year record for each year she was observed, yielding a total of 10,650 person-years. The model estimates change trajectories using all available data and taking right-censoring (i.e., early exits from the sample) into account.

The multilevel model for change also is able to handle spacing differences in interview schedules, a characteristic of the FFCWS, which interviewed women at varying intervals. Rather than generating trajectories from hypothetical time points (e.g., the average time distance between interviews), the multilevel model for change allows time between interviews to vary across individuals (Singer & Willett, 2003). Here, we measure time in annualized century months, a demographic convention based on months elapsed since January of 1900.

The multilevel model for change comprises a system of equations estimated at two levels (Singer & Willett, 2003; Snijders & Bosker, 1999). The Level-1 model considers within-mother change (e.g., how do support perceptions change for single mothers over time?) and allows us to describe the nature of each mother’s deviation from her baseline status over subsequent interviews. The Level-2 model measures between-mother differences in change (e.g., how do changes in support perceptions vary across individuals?). The Level-2 model uses the heterogeneity in change across individuals to determine the shape of individual growth trajectories, conditional on each individual’s covariate values and the interrelationships among covariates. The two Level-2 equations allow us to address (a) how demographic, household, and socioeconomic characteristics influence single mothers’ perceived support at the time of the focal child’s birth, and (b) how these same predictors influence the rate at which mothers’ support perceptions change over the subsequent five years.

Results

Univariate Results

Table 1 presents a statistical portrait of the unmarried mothers supporting this analysis, weighted to represent the mothers of the 1999 birth cohort in U.S. cities of at least 200,000 persons. Looking first at the outcome variable, at the baseline interview, approximately 82% of the unmarried mothers reported having a complete safety net; that is, access to financial assistance, child care, and housing should they need it. Examining safety net components individually, similar percentages of mothers reported access to financial assistance (88%), to child care (90%) and to housing (91%). Although these figures may seem high in light of the levels of disadvantage suggested by the remainder of the table, this level is comparable to estimates in prior studies (Ciabattari, 2007; Harknett, 2006; Henly et al., 2005).

Table 1.

Baseline Distribution of Mothers across Categories of the Dependent Variable and Covariates: Fragile Families and Child Well-Being Study^a

Variable	Percent
Outcome
Complete safety net	81.8%
$200 loan	87.8
Child care	90.2
Housing	90.7
Time-fixed Covariates
Mother’s age at baseline:
Less than 20 years	24.6
20 – 29 years	59.7
30 years or older	15.7
Race and nativity status:
Black immigrant	0.9
Black native-born	53.5
Hispanic immigrant	14.9
Hispanic native-born	17.8
White immigrant	0.6
White native-born	12.3
Grandmother has HS diploma	57.3
Focal child characteristics
First-born	39.8
Low-birth weight	10.6
Female	46.5
Time-varying Covariates
Mother’s educational attainment:
Less than HS/GED	43.2
HS/GED only	32.8
More than HS/GED	24.0
Economic characteristics:
Employed within the past year	79.1
Receives public assistance	40.9
Lives at or below poverty level	40.3
Household characteristics:
Living with a partner	40.3
Living with a parent	32.9
Number of biological children in household
1	39.8
2	31.0
3 or more	29.1
Mother’s health is fair or poor	7.9
Mother attends religious services weekly or more	16.9

^aUnweighted N = 3,065. Percentages weighted to represent 1999 birth cohort in U.S. cities with population of 200,000 or more

The second block of variables in Table 1 presents the time-fixed covariates. One-fourth of the respondents supporting our analyses were 19 or younger at their child’s birth. The majority—about 60%—were between the ages of 20 and 29, the primary childbearing years in the United States, and 16% were 30 years or older. Members of this urban sample also were predominantly women of color: over half identified as non-Hispanic, Black, nearly one-third as Hispanic, and 13% as non-Hispanic White. Over 16% of these mothers—primarily Hispanic-identified—were born outside of the United States. The mothers of almost three-fifths of respondents had completed high school.

The remaining fixed covariates are controls for characteristics of the focal child. The focal child was the first birth for nearly 40% of the mothers in this analysis. Just under half were females. Approximately 11% of the focal children weighed less than 2,500 grams at birth—substantially higher than the national average for birth cohorts in the late 1990s (Martin et al., 2003)—but likely a reflection of the over-representation of Black and economically disadvantaged mothers in the FFCWS.

The level of economic disadvantage in this sample is clear from the time-varying covariates. At baseline, 43% had not completed high school or earned the graduate equivalency diploma; less than one-quarter reported any schooling or vocational training beyond high school. Although almost four-fifths reported employment in the year preceding the focal child’s birth, 40% reported a poverty-level or lower income and a similar share reported receiving public assistance. Most of these mothers were living with partners (40%) or parents (33%), and at the baseline interview, 60% had at least one child other than the focal child living with them. Few (8%) reported being in poor or only fair health, and only 17% reported regularly attending religious services.

A single snapshot of socioeconomic and household characteristics at the baseline interview is insufficient to capture the complexity of these mothers’ lives. Table 2 shows, for the outcome and the time-varying covariates, the percentage of mothers experiencing a status change and the direction of the change over the five-year period. Over 40% of mothers lost or gained at least one safety net component in this time span. Relatively few enhanced their educational credentials; fewer than 10% of those without high school diplomas or GEDs earned one, and only 14% of those with a high school degree at baseline acquired additional schooling. Over three-fifths of mothers entered or exited employment at least once, and over half of mothers entered or exited poverty (55%) or moved on or off public assistance (51%). A substantial share of these single mothers experienced changes in their household compositions. Almost three-fifths of mothers experienced the entrance or exit of a romantic partner, and 46% saw one or more parents move in or out. Almost one-half of mothers (49%) experienced a birth within five years of the focal child’s birth. Weekly religious attendance also varied; nearly one-half of mothers either started or stopped attending services weekly or more. Changes in health status were somewhat less common, with less than one-third entering or exiting fair or poor health.

Table 2.

Percentage of Single Mothers Reporting a Status Change,^* by Covariate: Fragile Families and Child Well-Being Study.

	Percent	Percent	Percent	Percent
Variable	Reporting Change	Lost	Inconsistent	Gained
Outcome
Complete safety net	40.2	27.2	59.5	13.3
Time-varying Covariates
Mother’s educational attainment:
Less than HS/GED	8.5	---	---	8.5
HS/GED only	13.6	---	---	13.6
More than HS/GED	0.0
Socioeconomic Characteristics:
Employment	60.4	26.3	63.3	10.4
Public assistance	50.7	18.1	46.7	35.3
At or below poverty level	55.1	15.6	53.7	30.7
Household characteristics:
Living with a partner	57.7	20.9	48.2	31.0
Living with a parent	45.5	38.5	49.0	12.5
#biological children in household	49.1	---	---	49.1
Mother’s fair or poor health status	27.9	11.4	52.6	36.0
Religious service attendance	47.7	7.2	47.9	44.9

^*“Percent Lost” includes those who answered “yes” at Baseline, answered “no” at a later time, and never answered “yes” again. “Percent Inconsistent” includes those who experienced more than one change. “Percent Gained” includes those who answered “no” at Baseline, answered “yes” at a later time, and never answered “no” again.

^---Not applicable

Turning to the nature of status changes, mothers who experienced a status change rarely gained and kept resources. Instead, they either experienced repeated changes over the five years or lost resources. Of those mothers whose support perceptions changed—about 40% of the sample—nearly 60% reported at least one additional change and only 13% gained and kept a net. Employment status, public assistance receipt, poverty status and living arrangements evidenced similar patterns: inconsistency and increased vulnerability were more common than positive change. Although mothers who experienced change tended to fare worse socioeconomically as time passed, some changes were more often positive. Of those whose partnership status changed, mothers more frequently gained than lost a partner (31% versus 21%). Change in religious service attendance was more likely to involve increased (45%) than decreased (7%) participation.

Mixed Effect Logistic Regression Estimates

Table 3 presents the results from the mixed effects logistic regression models of perceived social support. Model A is an “empty” model (Snijders & Bosker, 1999), estimated without predictors in order to ascertain whether there is sufficient variation in perceived support both between mothers and over time to merit further investigation. The between-person constant in the fixed effect model approximates the average log-odds of reporting a full set of social supports across mothers and time. Transformed, this constant indicates that the average probability of a full net during the five-year period is 0.843 (P = exp(2.231)/1 + exp(2.231)). Because the constant is statistically significant, we can reject the null hypothesis that this probability is unvarying across individuals and over time. The residual variance of the random effects intercept also is statistically significant, suggesting substantial and potentially explainable residual variation between mothers over time.

Table 3.

Mixed Effects Logistic Regression Models of Perceived Safety Net: Fragile Families and Child Wellbeing Study, Baseline—Year 5

	Model A	Model B	Model C	Model D
Fixed Effects
Between-mother constant
(initial status)	2.231***	1.989***	2.209***	2.861***
	(0.073)	(0.082)	(0.228)	(0.272)
Time-fixed covariates
Race and nativity
Black immigrant			−1.035**	−0.904*
			(0.350)	(0.436)
Black native-born			−0.351*	−0.301
			(0.148)	(0.182)
Hispanic immigrant			−0.633**	−0.769**
			(0.206)	(0.250)
Hispanic native-born			−0.109	−0.117
			(0.173)	(0.213)
White immigrant			−1.548*	−1.747*
			(0.727)	(0.880)
White native-born (reference)			---	---
Grandmother HS diploma			0.339**	0.316**
			(0.103)	(0.111)
Mother’s age			−0.038***	−0.058***
			(0.010)	(0.012)
Child is mother’s firstborn			0.346**	0.349**
			(0.126)	(0.134)
Child is female			0.201*	0.224*
			(0.093)	(0.100)
Child low birth-weight			−0.150	−0.139
			(0.146)	(0.156)
Time-varying Covariates
Mother’s educational attainment
Less than HS diploma/GED			−0.218***	−0.544***
			(0.059)	(0.094)
HS diploma/GED only			−0.065	−0.250**
			(0.055)	(0.091)
More than HS/GED (reference)			---	---
Economic characteristics
Employed			0.071	−0.022
			(0.058)	(0.084)
Lives at or below poverty level			−0.398***	−0.354***
			(0.063)	(0.063)
Receives public assistance			−0.192**	−0.339***
			(0.058)	(0.085)
Household characteristics:
Lives with partner			0.107	0.192**
			(0.058)	(0.058)
Lives with parent			0.181**	0.283***
			(0.069)	(0.069)
Additional biological children			−0.103*	−0.095*
			(0.041)	(0.043)
Reports fair/poor health			−0.451***	−0.655****
			(0.076)	(0.125)
Attends religious services weekly			0.140*	0.163**
			(0.057)	(0.057)
Within-mother constant (rate of change)		−0.041	0.225***	−0.232**
		(0.028)	(0.058)	(0.088)
Race and nativity
Black immigrant				−0.225
				(0.144)
Black native-born				−0.061
				(0.060)
Hispanic immigrant				0.090
				(0.085)
Hispanic native-born				−0.018
				(0.070)
White immigrant				−0.028
				(0.389)
White native-born (reference)
Mother’s educational attainment
Less than HS diploma/GED				0.106***
				(0.020)
HS diploma/GED only				0.065**
				(0.020)
More than HS/GED (reference)				---
Economic characteristics
Employed				0.058**
				(0.020)
Receives public assistance				0.067***
				(0.020)
Reports fair/poor health				0.066*
				(0.031)
Random Effects
Variance Components
Variance: initial status	4.330***	2.792***	1.830***	2.876***
	(0.292)	(0.376)	(0.301)	(0.397)
Variance: annual change		0.069**	0.117***	0.096***
		(0.033)	(0.035)	(0.032)
Covariance		0.358***	0.342***	0.146
		(0.065)	(0.058)	(0.074)
Model statistics
Rho	0.568	0.020	0.034	0.028
Goodness-of-fit
Deviance	10374.18	10344.88	9856.68	9735.27
AIC	10380.18	10354.88	9906.69	9807.27

Note: Model rho calculated with the variance of the level-one residual fixed at π²/3 (Snijders & Bosker, 1999: 224).

^*p < .05.

^**p < .01.

^***p < .001.

In the empty model, the residual intraclass correlation coefficient of the latent responses, rho, has two interpretations. It represents the average correlation between any pair of composite residuals (e.g., between Baseline and Year 1), and it quantifies the relative magnitude of the within- and between-person variance components, providing a measure of between-subject heterogeneity (Singer & Willett, 2003). The value of rho for Model A exceeds 0.5, suggesting both a high degree of residual autocorrelation and, more importantly, that somewhat more than half of the total variance in the probability of a full safety net is attributable to differences between mothers.

Model B adds to Model A the annualized century-month indicator to capture the rate of change in support perceptions, and a random time component. The result is an unconditional growth model (Singer & Willet, 2003) that allows individual mothers to differ in their initial safety net status and in the probability that their status will change. A likelihood ratio test comparing the deviance values for Models A and B yields a χ² of 29.30 (df=3, p < .0001), indicating that Model B provides a better fit to the data. Comparison of the Akaike IC statistics for Models A and B supports this conclusion.

Because Model B explicitly models the annual probability of change, the between-mother constant of Model A is now interpreted simply as the average initial status for all mothers. Here, as in Model A, it is statistically significant so we reject the null hypothesis that the log-odds of reporting a full safety net at the child’s birth are equal across individuals. The within-mother constant fails to attain statistical significance and thus provides insufficient grounds to reject the null hypothesis that the average probability of change does not vary. In other words, although mothers do vary in their initial odds of reporting a full set of social supports, the typical mother experiences no change in the probability of reporting a full safety net over the subsequent five years. As we show below, however, this result is misleading: the non-significant estimate of within-mother change masks important between-mother differences in change trajectories.

In the stochastic portion of Model B, statistically significant residuals for both initial status and annual change point to unexplained between-mother variability in the log-odds of both a full safety-net at the child’s birth and—more importantly—the probability of change over the subsequent five-year period. The residual covariance indicates a strong, positive correlation (0.816 = 0.358/ √2.792×0.069) between initial status and the probability of annual change. Mothers who reported a full net at their child’s birth more frequently experienced a status change (e.g., lost one or more components of their safety net) in subsequent years than did mothers who did not report a full net. In short, losing a full net was more likely than gaining a full net.

A final point of comparison between Models A and B is the residual intraclass correlation coefficient, the value of which may increase or decrease depending on the model covariates (Rabe-Hesketh & Skrondal, 2008). In Model B, rho is 0.021, indicating a substantial reduction in the relative magnitude of the within- and between-mother variance components. An intermediate model (results not shown) confirms that this reduction is due entirely to the inclusion of the random time effect. This change foreshadows our key finding: the strong dependence of mother-specific change trajectories on between-mother differences.

The results in Models A and B confirm that single mothers’ support perceptions do change over time, and suggest that this change reflects between-mother differences. Therefore, we expanded Model B to include the full set of demographic, household, and economic covariates; the results are shown in Model C. As expected, Model C provides a better fit to the data, indicated by statistically significant decrease in the deviance and markedly lower AIC statistics relative to Model B. The addition of the covariates changes the interpretation of the two model constants. The between-mother constant no longer represents initial safety net status for all mothers; rather, it represents initial status for a subset of mothers—those with a value of 0 on all other model predictors. The within-mother constant now represents the average monthly rate of change in perceived support controlling for all covariates. In Model B, this constant was negative and non-significant; in Model C, however, it is positive and significant, suggesting that the probability of perceiving full support increases over time. As we observe below, however, the positive sign on this coefficient is an artifact that will change in Model D, once within-mother change is taken into account.

The coefficient estimates suggest that mothers’ demographic, economic, and household characteristics shape their support perceptions. Consider first mothers’ fixed characteristics. Native-born Blacks and immigrants of all races were less likely than native-born Whites to report a full safety net. Mothers whose own mothers had completed high school had 40% (exp^b=1.40) higher odds of a full net, all else equal. Younger mothers were more likely than their older peers to perceive the support of a full net regardless of other characteristics; each one-year increase in age at the focal child’s birth decreased the odds of a full net by 4%. Compared to their higher-parity peers, first-time mothers had 41% higher odds of full support, and those whose focal child was a girl had 22% higher odds.

Time-varying characteristics, too, shape support perceptions. Mothers who failed to complete high school or attain an equivalency degree by the Year 5 interview had 20% lower odds of a full net than their better-educated peers. Each additional report of a below-poverty household income lowered the relative odds of a full net by 33%, and each additional report of public assistance receipt was associated with a 17% reduction in odds. Time lived with a romantic partner or a parent increased the odds of reporting a full safety net, while each additional child reduced those odds. Religious service attendance also was associated with perceived support; those who reported attending at least once weekly had 15% higher odds of reporting a full net than those who did not attend. Time spent in fair or poor health was associated with reduced odds of reporting a full net.

Comparing the variance components of Models B and C also yields insight into support trajectories. The covariates explained 34.5% (1 - (1.830/2.792)) of the variability in initial status. Still, the statistical significance of the initial status variance indicates that potentially-explainable residual variation remains. The variance component for annual change suggests that the variability in the rate of change increased with the addition of covariates. This is not the case; rather, the increase is an artifact of a difference in scaling of the dependent variable with the addition of covariates.

Model C’s statistically significant variance components reinforce the notion that additional covariates could improve model explanation and fit. The between-mother covariates introduced in Model C do not address within-mother change. Model D builds upon Model C to consider how predictors influence which mothers gain or lose complete safety nets over time by interacting covariates with months elapsed since the child’s birth. We tested time interactions for each covariate; the presented model includes only those interactions that provide the best fit to the data as indicated by the AIC and deviance statistics.

The addition of the time-interactions led to an 18% decrease in within-mother variance, indicating that the covariate-by-time interactions explain within-mother, not between-mother, variation. The increased initial-status variance component reflects additional residual variance introduced by the covariate-time interactions. The reduction in the residual intraclass correlation from 0.342 to 0.146 and its loss of statistical significance indicate that the addition of the interaction terms accounts for the relationship between initial status and the rate of change observed in earlier models. In other words, change over time in the effects of some covariates helps explain why safety net loss was more common than safety net gain.

Figure 1 illustrates the nature of these changes, using public assistance receipt (Panel A) and mother’s educational attainment (Panel B). Each panel shows the change over time in the predicted probability of perceiving a full safety net associated with the specified covariate, holding the remaining covariates constant at their average values. To simplify the display, the graphs depict the predicted probabilities for mothers who maintained their initial status over the full five years (i.e., those who received public assistance continuously following their child’s birth versus those who never received public assistance, and those who maintained the same level of education over the five-year period versus those who attained the next degree level.)

Figure 1.

Predicted Probability of Perceiving a Full Safety Net, by Specified Covariates: Fragile Families and Child Wellbeing Study, Baseline—Year 5

The pattern of time-dependence for public assistance receipt suggests that the negative impact of this form of disadvantage on mothers’ support perceptions cumulates over time. At baseline, mothers receiving public assistance were about ten points less likely to perceive a full safety net than mothers who were not receiving public assistance. Although the predicted probability of perceiving a full net decreased over time for both groups, it fell more sharply for those on public assistance, increasing the difference to over 20 points. The absence of public assistance receipt is more important for retaining an intact net than for having one initially or, stated differently, women who reported continuous public assistance suffered increasing disadvantage with each passing year. Results for mothers’ health and employment status suggest a similar story.

As shown in Panel B, however, greater initial disadvantage is sometimes associated with less deterioration in support perceptions. Mothers who did not complete high school or obtain a GED over their children’s first five years had increasingly higher probabilities of perceiving a full net compared to their better educated peers. In short, their nets appeared more resilient than those of mothers who started with more education. Results for mother’s age at focal child’s birth point to a similar protective effect of youth, as younger mothers reported more access initially and less deterioration over time than older mothers.

Discussion

The well-documented deterioration of the public safety net has increased the salience of what Harknett (2006) has termed “private safety nets” for low-income families, who are disproportionately headed by single mothers. Although most single mothers report access to emergency support from friends and family members, the stability of their perceptions over time—in the face of changes in their human capital, economic circumstances, marital status, and living arrangements—was unclear. In this paper, we set out to describe the relative stability of single mothers’ safety net perceptions over a five-year period following the birth of a child, using prospective data from five waves of the Fragile Families and Child Well-Being Study. We operationalized safety nets as perceived access to three basic necessities in case of an emergency: $200, a place to live, and child care (Harknett & Hartnett, 2011; Ryan et al., 2009). We had two primary aims: (1) to determine whether and how mothers’ support perceptions changed over an extended period of time, as the child aged through toddlerhood and, potentially, other aspects of their lives also changed, and (2) to identify whether the rate of change differed across groups defined by demographic and socioeconomic characteristics.

Like others before us (Harknett, 2006; Henly et al., 2005; Turney & Kao, 2009), we find that most single mothers report having the support of a private safety net. Although mothers with greater social and economic resources—more schooling, higher income, good health, and the presence of other adults in the household—were somewhat more likely to perceive this safety net, overall, 82% of mothers reported access to all three forms of support in the days following their child’s birth. Most often, those mothers who reported a full net at their child’s birth retained it over the child’s first five years; however, instability was not uncommon. Indeed, over 40% of mothers experienced at least one change in safety net status over this time period. Among these mothers, losing a full safety net was more common than gaining one: about 13% reported gaining a net, over one-quarter lost their nets, and nearly 60% reported repeated status changes.

Multivariate results revealed, first, that mother, child, and household characteristics all contributed to between-mother variation in safety net perceptions over time. Education, employment, living with a romantic partner, and religious service attendance were associated with greater access to the full set of supports; below-poverty household income, public assistance receipt, subsequent births, and poor health were associated with less access. This pattern suggests that mothers with greater resources were more likely than their peers to have access to a full set of supports. We observed two exceptions to this overall pattern, however: very young mothers and first-time mothers reported a full safety net at higher rates than older and more experienced mothers.

The analyses also point to differential rates of change in mothers’ support perceptions. Overall, mothers with great need (e.g., those without stable employment or a stable partner or stable health) were not only more likely than more advantaged mothers to report a dearth of support, but also they experienced a steeper decline in perceived support over time. Our analysis supplements earlier findings on informal safety nets: they are not equally available among single-mother families (Harknett, 2006; Harknett & Hartnett, 2011). Particularly troubling from a policy perspective is the marked deterioration of private support for mothers in poor health and for mothers receiving public assistance. At the baseline interview, the predicted probability of a full safety net among single mothers who reported public assistance receipt in the previous year was about 85% compared to about 93% for those not receiving assistance. Five years later, the predicted probability for mothers receiving assistance throughout the observation period was 56% compared to nearly 80% for mothers who received no support over the same period. Among single mothers who reported poor health at baseline, the predicted probability of perceiving a full safety net was low—just under 60%—compared to over 90% for single mothers in good health. At the fifth-year follow-up, the predicted probability of a full net for mothers who consistently reported good health was just under 80% but for those who consistently reported poor health, the predicted probability of perceiving a full set of supports was just over 20%.

Policy Implications

The implications of these findings are disturbing in an era of time-limited public welfare. Others have documented that mothers who face significant educational, employment, and health barriers are more likely to lack safety nets and depend upon welfare cash assistance (e.g., Harknett, 2006). Our analysis reveals that these barriers become more salient in shaping safety net availability as children age. Complimenting Harknett’s (2006) findings, the findings reported here suggest that mothers most likely to need informal support are also the most likely to have nowhere to turn.

Our analyses provide evidence that personal safety nets become less available as children enter toddlerhood and that social and economic covariates become more important in distinguishing between mothers who do and do not have private safety nets. These findings yield two central policy implications. First, private safety nets are not a reliable fallback for single mothers with young children, and the current structure of the means-tested benefit programs that comprise the public safety net, including TANF and Supplemental Nutrition Assistance Program (SNAP), discourage families from planning or budgeting for emergencies. Strict asset limits mean that single mothers who have managed to accumulate any savings are ineligible for assistance to help them weather job loss or illness. Higher asset limits for program eligibility would provide families the opportunity to become self-reliant in times of need, as would the expansion of asset development programs that focus on government-matched Individual Development Accounts (IDAs) designed to fund opportunities to exit poverty (e.g., home down payments, educational financing). These programs encourage financial literacy and self-reliance, and provide a source of funds in the event of a crisis (McKernan & Ratcliffe, 2008; McKernan & Sherraden, 2008).

A second policy implication of the unreliability of private sources of assistance is the need for improved access to the public safety net and greater program literacy. Our descriptive statistics revealed high rates of employment volatility and low rates of educational enhancement over a five-year period. These findings point to the failure of public work supports. Less than 7% of low-income working families receive available supports, including food stamps, children’s health insurance, earned income tax credits, and child care subsidies (Zedlewski et al., 2007). Less than 15% of eligible families were able to benefit from childcare subsidies alone (Boots, Macomber, & Danzinger, 2008). Increasing public safety net program access and program literacy are critical to engendering self-sufficiency among low-income, single mothers and to improving the lives of their children.

Limitations

A notable advantage of mixed-effects estimation is its ability to control for stable, but unobserved, factors (e.g., personality traits), that may be linked to safety net perceptions and its covariates. More charismatic mothers, for example, may be more likely to be employed and more likely to have and to maintain a safety net. Despite this advantage, however, potential bias from omitted variables and less stable characteristics must be acknowledged. The models presented here do not include depression or substance abuse because they did not predict support perceptions in initial analyses; both, however, may play important roles in social support networks. For example, depressed mothers may struggle to maintain relationships with partners, family, and other potential network members. Also, our analysis examined an admittedly crude safety net of access to money, childcare, and housing. Single mothers who have access to these minimal goods may lack basic necessities and, possibly, those who lacked access to safety nets may have what they need.

Future Directions and Conclusion

Our findings suggest that vulnerability is associated with a steeper decline in safety net perceptions over time, but our data do not reveal why the most vulnerable tend to lose their safety nets more rapidly. A fruitful direction for future research is a more nuanced description of social support networks and the norms of giving and receiving assistance. Are single mothers’ perceptions of potential supporters in times of need reasonably accurate? Does the accuracy of these perceptions vary by mothers’ characteristics? What happens to a mother once she has drawn on available support? Prior research highlights the importance of reciprocity (e.g., Stack, 1974). Do the demands of early parenting mean that mothers’ networks disintegrate because mothers do not have the capacity to reciprocate? In addition to contextualizing single mothers’ safety nets, future research should examine the consequences of safety net trajectories for families. How do safety net trajectories relate to mothers’ educational or romantic trajectories? What is the impact of a present, absent, or inconsistent private safety net on children's developmental outcomes?

Certainly, future welfare policy efforts would benefit from a deeper understanding of single mothers’ informal networks. Current welfare policy assumes that informal networks will pick up when families reach the limits of support or are otherwise sanctioned. Extant evidence indicates that private safety nets are beneficial for those families who have access to them (Harknett, 2006; Henly et al., 2005); yet, we find that the most disadvantaged mothers do not have consistent or reliable access to private sources of support for basic necessities. Indeed, our analyses indicate that the most disadvantaged mothers—and their children—have the least resilient private safety nets: the greater the need, the faster their deterioration. Although time-limits and work requirements may have led to the shrinking of welfare rolls, it is unlikely that this PRWORA has led to better lives for the most vulnerable mothers and their children. Instead, extreme vulnerability may preclude mothers from meeting PRWORA’s work requirements and from activating a personal safety net.