Abstract
We examine the influence of individual characteristics and targeted and universal social policy on single-mother poverty with a multilevel analysis across 18 affluent Western democracies. Although single mothers are disproportionately poor in all countries, there is even more cross-national variation in single-mother poverty than in poverty among the overall population. By far, the United States has the highest rate of poverty among single mothers among affluent democracies. The analyses show that single-mother poverty is a function of the household’s employment, education, and age composition, and the presence of other adults in the household. Beyond individual characteristics, social policy exerts substantial influence on single-mother poverty. We find that two measures of universal social policy significantly reduce single-mother poverty. However, one measure of targeted social policy does not have significant effects, and another measure is significantly negative only when controlling for universal social policy. Moreover, the effects of universal social policy are larger. Additional analyses show that universal social policy does not have counterproductive consequences in terms of family structure or employment, while the results are less clear for targeted social policy. Although debates often focus on altering the behavior or characteristics of single mothers, welfare universalism could be an even more effective anti-poverty strategy.
Introduction
The poverty of single mothers has long been a source of controversy and concern. Even the seventeenth-century poor laws of Colonial America distinguished between the “deserving” poor and “undeserving” husbandless mothers (Sidel 2006). Public debate has intensified as single-mother families and their presence among the poor have risen in recent decades (McLanahan and Percheski 2008; Wu 2008). Policymakers faced with the need to protect single mothers from economic insecurity, while curtailing their perceived dependence on welfare, have struggled with a so-called “new American dilemma” (Garfinkel and McLanahan 1986). Often stereotyped and blamed for their own disproportionate poverty (Sidel 2006), single mothers have been called “the most prominent lightning rod for political attacks,” provided with assistance only reluctantly and with stipulations that are intended to counter what is perceived as problematic behavior (Handler and Hasenfeld 2007:184).
Beyond public debates, single mothers have received considerable attention from demographers and poverty researchers. An interdisciplinary literature shows a strong association between single motherhood and children’s, women’s, and overall poverty (Ananat and Michaels 2008; Bianchi 1999; Rank 2005; Seccombe 2000; Thomas and Sawhill 2002). Single mothers are particularly vulnerable because of their typically lower wages, lack of spousal support, and the burdens of raising children (Seccombe 2000; Sorensen 1994). In turn, single motherhood is regarded as a key mechanism in the reproduction of poverty and inequality (Edin and Lein 1997; Lichter et al. 2006; Martin 2006; McLanahan and Percheski 2008; Musick and Mare 2004).
A recent literature has emerged on cross-national differences in single-mother poverty as well (e.g., Christopher 2002; Kilkey and Bradshaw 1999; Rainwater and Smeeding 2004; Sorensen 1994). For example, Huber et al. (2009) analyzed macro-level variation in single-mother poverty across affluent democracies over the past 30 years. Others have investigated individual-level single-mother poverty within countries at one point in time, by examining one country (e.g., Kamerman 1995; Rose 1995), a small set of countries (e.g., Christopher et al. 2002), or a larger set of affluent democracies (e.g., Kilkey and Bradshaw 1999). Cross-national research on child poverty has also highlighted the role of single motherhood (Chen and Corak 2008; Heuveline and Weinshenker 2008). By providing a comparative perspective, this literature has broadened the field and called attention to the role of national social policy in shaping the economic security of single-mother families.
Our study builds on these literatures by conducting a cross-national, multilevel analysis of single-mother poverty. Specifically, we assess whether targeted or universalist social policies can explain variation across affluent democracies. We concentrate on social policy because many recent studies demonstrate the centrality of the welfare state to cross-national variation in children’s, women’s, single-mother, and overall poverty (e.g., Brady 2009; Heuveline and Weinshenker 2008; Huber et al. 2009; Misra et al. 2007; Moller et al. 2003; Rainwater and Smeeding 2004). Despite the welfare state’s centrality in previous studies, we know little about the relative effects of targeted and universal social policy for poverty. Targeted social policies purportedly concentrate resources on the well-being of the most vulnerable, and yet many contend that universal welfare states more effectively promote social equality.
This study offers several unique contributions. First, previous studies in Demography on single motherhood and poverty have typically studied only the United States (e.g., Fitzgerald and Ribar 2004; Lichter et al. 2006; Martin 2006; Musick and Mare 2004; Wu 2008). The United States may be unusual in both the high level of single motherhood and the high likelihood of poverty among single mothers. To assess the generalizability of relationships between nonmarital fertility, union formation, social policies, and poverty, it is essential to broaden the scope of comparison and examine cross-national differences in single-mother poverty. Second, this is one of the first multilevel, cross-national analyses of single-mother poverty. Although macro-level studies have assessed the effects of cross-national differences in social policy, such studies cannot incorporate individual-level information on the characteristics of single mothers. In contrast, individual-level analyses within countries have incorporated precise data on single mothers, but can only compare national differences in social policy qualitatively. Hence, a multilevel analysis can more rigorously test the effects of social policy on single-mother poverty (Brady et al. 2009; Misra et al. 2007). Third, although there has been a literature on the causes of universal versus targeted social policy (e.g., Nelson 2007), there has been little empirical research comparing their effects. Given the salience of social policy for a variety of demographic outcomes, there is a clear need for scrutiny.
Social Policy and Single-Mother Poverty
The Case for Targeting
Many argue that the principal anti-poverty strategy should be to concentrate resources on those at greater risk of poverty, like single mothers (Barry 1990; Barth et al. 1974; Besley 1990; Le Grand 1982; Tullock 1997). Purportedly, targeting has three major advantages. First, targeted social policies are more efficient. In an environment of finite or austere budgets, targeting focuses scarce resources on those most in need (Blank 1997; Squire 1993). As Greenstein (1991:457) explained, “With the funds available for social program interventions likely to remain limited, too heavy an emphasis on costly universal approaches could result in too few resources being directed to those at the bottom of the economic ladder.” Instead of subsidizing middle-class families, targeted programs allow the market to supply resources to those able to work and save. Hence, targeted programs avoid the redundancy and reverse-redistributive effects of superfluous state support to the affluent (e.g., U.S. Social Security old-age pensions collected by the nonpoor) (Tullock 1997). This point grows out of the common criticism that the primary beneficiaries of social policies are typically the nonpoor and that only a small share of welfare spending actually goes to those at or near poverty (Goodin and Le Grand 1987). By avoiding expending precious resources on those not at risk of poverty (often called “leakage”), targeted programs free up additional resources that can be channeled into ensuring the basic security of low-income households (Collier and Dollar 2001; Squire 1993). Moreover, because the poor are more likely to spend assistance on consumption than savings, targeted programs are more likely to translate into basic needs, such as housing and food. In sum, targeted programs concentrate the transfer of scarce resources to those most in need and thus make the biggest difference in alleviating poverty (Blank 1997).
Second, anti-poverty policies without means testing may be counterproductive (Gilbert 2002). Whereas some welfare states provide a variety of programs as an entitlement of citizenship, targeted programs can provide incentives for work and marriage, and can induce the poor to leave poverty. There has long been concern that generous social policies have adverse labor-supply effects (Lindbeck 1998), encourage dependency and longer poverty spells (Bane and Ellwood 1994), and provide an incentive for single parenthood (Lichter et al. 1997; Moffitt 2000) or fertility (Schellekens 2009). Because means-tested programs often cut off support after a household reaches a certain income level, such policies may encourage families to leave welfare programs and the lower incomes that are associated with them. Means-tested programs can be designed to taper off as a household’s income rises (e.g., the U.S. Earned Income Tax Credit (EITC)) or coupled with work requirements to encourage families’ pathways out of poverty (Bane and Ellwood 1994; Blank 1997; Leisering and Leibfried 1999; Mead 1986). In contrast, comprehensive welfare generosity for all citizens could indirectly increase poverty by encouraging unemployment, labor force withdrawal, and single parenthood.
Third, targeted policies for the poor have the potential to avoid the biases inherent in a general welfare state. Regardless of whether welfare states generally reduce poverty, an extensive literature demonstrates that welfare states also reproduce social hierarchies. Most relevant for our study is the role welfare states play in reinforcing gender inequality. Many scholars point out that gender inequalities—especially the feminization of poverty—remain resilient in societies that are relatively economically equal (Brady and Kall 2008; Gornick 2004; Misra 2002). Indeed, several social democratic and Christian democratic welfare states that have accomplished broader economic equality have also maintained traditional breadwinner gender roles and lower female labor force participation (Sainsbury 1999). As a result, feminists often contend that purportedly generous comprehensive welfare states disproportionately advantage men, strengthen patriarchy, do not enhance women’s autonomy, and often leave women (especially single mothers) particularly vulnerable to poverty. In an influential critique of the welfare state literature, Orloff (1993) argued that generous welfare states often fail to allow women to form and maintain autonomous nonpoor households. Partly in response, Esping-Andersen (1999) acknowledged that single motherhood is a “new social risk” that most welfare states were not built, and may be ill equipped, to manage. Given these concerns, targeted social policies aimed at assisting single-mother households—not comprehensive generous welfare states—may be more likely to alleviate single-mother poverty (Kilkey and Bradshaw 1999).
The Case for Universalism
Many others argue that the better approach to reducing poverty is to build a generalist, comprehensive welfare state (i.e., “universalist”) (Skocpol 1991). For universalists, targeting is an inferior approach because it has the unanticipated consequence of actually delivering fewer economic resources to those in need. Universalist welfare states may not be designed to provide economic security for specific marginalized groups. However, proponents of universalism counter the criticisms presented in the previous section by stressing that many policies that are not designed to redistribute end up having a redistributional impact (Sefton 2006; Skocpol 1992). Korpi and Palme (1998) referred to this as the “paradox of redistribution”: the more benefits are targeted at the poor, the less they actually reduce poverty. Because universalist welfare states are larger and more generous for the overall population, universalism ends up being more effective at reducing inequality and poverty (Nelson 2004). Purportedly, this is because universalist welfare states crowd out more inegalitarian private alternatives to social policy and because even earnings-related social insurance usually has an element of redistribution built into it (Korpi and Palme 1998). Consistent with arguments for welfare universalism, recent studies show that generous welfare states tend to have less poverty for all groups, not just for two-parent families or male breadwinners (Brady 2009; Christopher 2002; Huber et al. 2009). In fact, welfare state generosity is one of the most influential factors explaining cross-national differences in poverty (Brady et al. 2009; DeFina and Thanawala 2003; Moller et al. 2003; Rainwater and Smeeding 2004). Thus, single-mother poverty could be lower in an environment of welfare universalism simply because the entire population has less risk of poverty.
One reason why welfare universalism tends to entail larger and more generous social policies is that universal programs garner greater political support than targeted programs (Skocpol 1991; Wilson 1996). Universal welfare programs are less stigmatizing to the poor and more generous in the long run, largely because universalism is more politically popular and thus gains better and more secure funding (Sefton 2006). Korpi and Palme (1998) argued that the “institutional structures” of universalist welfare states enhance the formation of coalitions for welfare generosity and unite the interests of poor and nonpoor citizens (also see Nelson 2007). This argument is consistent with the literature on social policy–feedback effects (Skocpol 1992) and the role of constituencies of beneficiaries and “ratchet effects” in the new politics of the welfare state (Huber and Stephens 2001). As Skocpol (1992:531) wrote, “Policies not only flow from prior institutions and politics; they also reshape institutions and politics, making some future developments more likely, and hindering the possibilities for others.” Universalist programs become defined as citizenship entitlements or rights, and thus subsequently become difficult to underfund or cut back. Hence, welfare universalism is path dependent because the politics of social policy reinforce already established programs that have constituencies of beneficiaries and normative expectations attached to them (Brooks and Manza 2007).
Although targeting might appear to be more efficient, there are several unanticipated ways in which universalism may ultimately devote a greater share of resources to actual assistance. Means-tested targeted programs require monitoring and screening of the poor, which is administratively expensive (Blank 1997; Lindert 2004; Sefton 2006). In addition, stringent screening often sets up barriers to receiving benefits, inappropriately disqualifies recipients, and discourages enrollment in related programs (Currie 2006). Rather than devoting resources to assist the upward mobility of the poor, this constrains welfare services staff to spend time and effort on surveillance and enforcement (Piven and Cloward 1993). Just as some argue that universalist programs trigger disincentives, others suggest that targeting is counterproductive. According to this literature, targeting discourages employment because benefits will be cut off once one rises above the means-tested line (Squire 1993). For example, Edin and Lein (1997) showed that in the United States, Aid to Families with Dependent Children (AFDC) forced mothers to make choices between low-wage employment with no health insurance and greater hardship and uncertainty, or welfare with a guaranteed income and Medicaid for one’s children. They explain that the lack of universal health insurance and publicly provided childcare actually may have encouraged welfare dependency. In his comprehensive study of social welfare since the eighteenth century, Lindert (2004) concluded that the rise of universalism actually reduced work disincentives because everyone shared basic rights to income, health care, and other public services. Further, careful empirical studies have often failed to find evidence that less-targeted and more-generous welfare benefits actually encourage single motherhood and welfare dependency (Blau et al. 2004; Carlson et al. 2004; Fitzgerald and Ribar 2004).
Finally, universalist welfare states are better able to address the heterogeneous risks that low-income families and vulnerable groups face. Universal welfare states reduce poverty precisely because of the comprehensive scope of their social insurance, transfers, and welfare services (Kamerman 1995). People become eligible for targeted programs only after they have fallen into poverty, while universal programs enhance the well-being of all. As a result, universal policies reduce everyone’s chances and costs of risks, such as illness, and are more likely to prevent descents into poverty (Krishna 2007). Thus, universal welfare states distinctively offer a complex of integrated and interdependent programs that protect the poor, along with all citizens, from a wide variety of insecurities and risks (Wilensky 2002; Zuberi 2006).
Research Questions
The preceding discussion leads us to three related research questions. First, are both targeted and universal benefits effective at reducing single-mother poverty? Second, is one more effective than the other? Third, do targeted or universal benefits have counterproductive consequences that worsen poverty for single-mother families?
Methods
Individual-Level Data
The Luxembourg Income Study (LIS) provides the micro-level data, and the individual is the unit of analysis. The LIS is a set of cross-nationally and historically harmonized and nationally representative individual-level data sets (LIS n.d.). We analyze a data set near the year 2000 for 18 affluent Western democracies. We first confine our sample to women aged 18–54. Then, we select only those in female-headed households in which the head of household is neither married nor cohabiting and children are present.1 This sample excludes children, those over 64 years old, and men living in these households. To be clear, however, we do control for other people in the household. In additional analyses, we estimate all models for lone-mother households (i.e., single-mother households containing no other adults). The conclusions are consistent, so we present those results in Tables 5 and 6 in the appendix. The analyses merge the 18 countries into one file containing 15,116 individuals. Table 1 contains descriptive statistics and sources.
The dependent variable is poverty. We follow the vast majority of cross-national poverty studies and use the relative headcount measure of poverty (Brady 2003; Brady et al. 2009; DeFina and Thanawala 2003; Moller et al. 2003; Rainwater and Smeeding 2004; Smeeding 2006). An individual is defined as poor (= 1; nonpoor = 0) if she resides in a household with less than 50% of the median household income. We calculate household income after taxes and transfers using the standardized LIS variable DPI.2 To adjust for household size, DPI is divided by the square root of household members. The poverty threshold is calculated for each country and includes all individuals. The sample is reduced to single mothers only after the poverty threshold is calculated. As Table 1 displays, 26.9% of the sample is poor. At the end of the results, we discuss alternative dependent variables.
The analyses incorporate several demographic variables. Because households have multiple members and pool resources and expenses, several individual characteristics are measured at the household level. We include binary variables for no one employed and multiple earners in the household (reference = one earner). Using the LIS-standardized measures of education, we include binary measures of head low education and head high education (reference = medium).3 Next, we control for the age of the head and a binary variable for child under 5.4 We also include the number of other adults and number of children under age 18 in the household.
Country-Level Data
We use several archival sources, although Huber et al. (2004) are often the proximate source. Table 2 shows the values of the country-level variables and details on the LIS and poverty rates. Except for economic growth, the country-level variables are measured in the same year as the LIS survey.
To control for the economic context within a country (Gundersen and Ziliak 2004), we include two variables. Economic growth is the three-year average (t, t – 1, t – 2) of the annual rate of change in gross domestic product (GDP), measured in real purchasing power parity (PPP) dollars. Unemployment is the percentage of the labor force without employment.
To assess social policy, we examine two measures each of targeted and universal social policy. The first measure of targeting is the single-mother entitlement. We collected and coded original data measuring the amount of targeted cash assistance that a single mother with one child (under 3 years old) would receive if she was not employed.5 We tabulate family assistance, child rearing, and other cash benefits in current local currency, and then divide by the median equivalized household income from the LIS.6 Thus, this is the percentage of the median equivalized income that a single, nonemployed mother with one child is statutorily entitled to receive from the state.7 This measures the means-tested cash assistance officially targeted for single mothers.
The second measure of targeting is the targeting ratio, which follows the literature on “targeting efficiency” (Creedy 1996; Kakwani and Subbarao 2007; Mahler and Jesuit 2006) and others’ estimations of benefits in the microdata of the LIS (Behrendt 2000; Mahler and Jesuit 2006; Smeeding 2006). This measure calculates the equivalized value of total government assistance received by single-mother and all households in each country in the LIS.8 We then estimate the ratio of benefits received by single-mother households over those received by all households. This assesses whether single mothers receive disproportionate benefits compared to the overall population. A ratio of 1 (i.e., unity) means that they receive the same amount. Unlike the first targeting measure, which focuses on the official, statutory targeted assistance, this gauges the total actual benefits received by a targeted group.9
For the first measure of universalism, we construct a welfare state index. This comprehensive index of welfare generosity is a standardized scale (mean = 0, standard deviation (SD) = 1 across the 18 countries) of social welfare expenditures, social security transfers, and government expenditures as a percentage of GDP, as well as public health spending as a percentage of total health spending (alpha = .87).10 This index incorporates several classic measures of welfare generosity and combines measures that, as others have shown, significantly influence cross-national variation in poverty (e.g., Brady 2009; Brady et al. 2009; Huber and Stephens 2001; Moller et al. 2003).
As a second measure of universalism, we calculate the universal replacement rate. Like the targeting ratio, this uses the actual government assistance received based on the LIS. We estimate the mean equivalized total government assistance (social insurance plus social assistance) received for all households in each country. We then convert this into a percentage of the median equivalized household income. Thus, this is the average percentage of median income that typical residents of a country receive from the state.
Multilevel Modeling Technique
Partly because we have only 18 countries, we only estimate random intercept models and treat the individual-level coefficients as fixed effects.
Table S1 in Online Resource 1 contains a correlation matrix. There is no evidence of collinearity among individual-level or country-level variables. Because there are only 18 countries, it is important to keep the models parsimonious at Level 2. As discussed in the Results section, when we include two social policy variables in one model, we omit the economic context variables. Nevertheless, even if we include the economic context variables with the two social policy variables, the results are robust (analyses available upon request).
Results
Descriptive Patterns
Table 2 displays the cross-national patterns in single-mother poverty and the country-level variables.11 Across the 18 countries, the average single-mother poverty rate is 24.2%. However, there is substantial cross-national variation. In fact, there is more variation in single-mother poverty than in overall poverty (coefficients of variation .5 vs. .4). Canada, the Netherlands, the United Kingdom, and the United States are more than 1 standard deviation (11.3) above the mean, whereas Denmark, Finland, Norway, and Sweden are more than 1 standard deviation below the mean. Denmark has a single-mother poverty rate of only 5.7%. By far, the United States has the highest poverty rate among single mothers, at 41.3%, which indicates how unusual the U.S. case is relative to other affluent democracies.
The cross-national mean in single-mother poverty is significantly higher than the mean in overall poverty of 9.6 (t = 6.5, p < .001), and single-mother poverty is higher than overall poverty in all 18 countries. The ratio of single-mother poverty to overall poverty is displayed in the third column of Table 2. On average, single mothers are nearly 2.7 times more likely to be poor than the typical person, which is consistent with core claims in the feminist literature on the welfare state. France, Germany, Luxembourg, and the Netherlands exhibit ratios substantially above the mean, although Denmark, Finland, Ireland, Italy, Norway, and Sweden are substantially below the mean. Only for Denmark is the ratio of single-mother poverty to overall poverty near 1. Despite having the highest single-mother poverty rate, the United States is slightly below the mean in this ratio. Thus, one reason for the high rate of single-mother poverty in the United States is simply the high overall poverty rate.
The last row of Table 2 displays the country-level bivariate correlations (N = 18). The single-mother poverty rate is moderately correlated with the overall poverty rate, the rate of single motherhood, and the ratio of single-mother to overall poverty (r > .53). Economic growth and unemployment are not correlated with the single-mother poverty rate.
Single-mother poverty is strongly negatively correlated with both measures of welfare universalism (r > .56). However, it is not correlated with either measure of targeted social policy, and both correlations are positively signed. The lack of correlation with the single-mother entitlement results partly because the single-mother entitlement is insufficient to lift a family out of poverty in every country except France.12 In all other countries, a single-mother household relying exclusively on government benefits will have an income below the poverty threshold. Although the welfare state index and universal replacement rate closely correspond to well-known patterns in welfare generosity among affluent democracies (Hicks 1999; Huber and Stephens 2001), both measures of targeted welfare are negatively correlated with the measures of welfare universalism.
To provide concreteness to these correlations, we graph the bivariate relationships between single-mother poverty and the universal replacement rate and the targeting ratio. These two are chosen because they prove to be most consequential in the analyses that follow. Figure 1 shows a clear negative relationship between the universal replacement rate and single-mother poverty. The United States stands out for its particularly high single-mother poverty and low universal replacement rate, but the cross-national correlation remains strong if we omit the United States (r = –.57). By contrast, Fig. 2 shows that the targeting ratio is simply not very associated with single-mother poverty. Indeed, the sign is positive, and the countries with the highest targeting ratios also have high levels of single-mother poverty (e.g., Australia and the United Kingdom).
It is also instructive to compare the universalism and targeting of select countries. Despite its slightly above-average welfare state index, Italy has a targeting ratio less than 1 and a single-mother entitlement of 0 because single mothers are not guaranteed any welfare benefits solely for being a single mother.13 Sweden has a below-average targeting ratio and a single-mother entitlement less than 6%. This is the case even though Sweden has the highest values for the welfare state index and universal replacement rate (and Italy is higher than average on both). By contrast, Australia and Ireland have the second- and third-highest single-mother entitlements, and Australia has the highest targeting ratio (and Ireland has the fourth highest). However, these two have the second- and third-lowest values in the welfare state index and are below average in the universal replacement rate. Although Italian and Swedish single mothers do not receive particularly generous targeted assistance, they do reside in societies where 24.1% and 29.8% of GDP is devoted to social welfare expenditures. Alternatively, social welfare expenditures only amount to 18.0% of GDP in Australia and 13.6% in Ireland. In Sweden, all parents, not just single mothers, have access to publicly subsidized childcare and paid parental leave as well as a universal tax-free child/family allowance for each child (Kamerman 1995; Whitehead et al. 2000). Because many Italian single mothers coreside within intergenerational families, generous public pension, healthcare, and disability programs provide an indirect but salient layer of support for their economic security (Rainwater and Smeeding 2004:128, 130). Australia and Ireland feature single-mother poverty rates above the cross-national mean, while Italy and Sweden are below the cross-national mean. Thus, the descriptive patterns suggest that universalism more effectively reduces single-mother poverty than targeting.
Multilevel Analyses
Table 3 displays the odds ratios and z scores for the models of single-mother poverty. Throughout, the individual-level predictors are significant and stable. Multiple earners in the household, a head of household with high education, age of the household head, and additional adults in the household reduce the likelihood of poverty for single mothers. The presence of multiple earners in the household reduces the odds of poverty by a factor of 3.2—the largest negative effect. If the household head has high education, the odds decline by a factor of 2.3. For each 10 years the head is older, the odds of poverty are reduced by a factor of 1.27. For each additional adult in the household, the odds decline by a factor of 1.14. Thus, single mothers are less likely to be poor if other adults and earners reside in the household and if the head of household is older and more educated.
Conversely, unemployment, a head of household with low education, having young children, and the number of children in the household increase poverty among single mothers. If no one is employed in the household, the odds of poverty increase by a factor of 7.3—the largest effect overall. If the head of household has low education, the odds grow by a factor of 1.7. The presence of a child under 5 and each additional child increase the odds of poverty by a factor of about 1.3.
Models 1–4 also include controls for the economic context of the country. Although economic growth is always insignificant, unemployment is significantly positive in Models 1 and 2 (odds = 1.1) but insignificant in Models 3 and 4.
In Model 1, the welfare state index is significantly negative. The odds of single-mother poverty are reduced by a factor of 1.9 for a one-unit increase in the welfare state index. The index is constructed with a mean of 0 and a standard deviation of 1, but because some countries contribute more cases to the sample, the actual standard deviation is slightly greater than 1. For a standard deviation increase, the odds of single-mother poverty should decline by a factor of 1.95. These effects are comparable to the poverty-reducing effects of a highly educated household head, and are larger than the effect of the head of household aging 10 years or having another adult in the household. In addition, the welfare state index’s effect more than offsets the effects of having a low-educated head of household, a child under 5, or additional children.
Model 2 demonstrates that the universal replacement rate also has a significant negative effect. For a standard deviation increase in the universal replacement rate, the odds of single-mother poverty are expected to decline by a factor of 1.9. Hence, the universal replacement rate’s effect is comparable to the welfare state index and larger than several demographic controls.
Model 3 shows that the targeting ratio is negatively signed but not close to statistical significance (z = –0.2). In Model 4, the single-mother entitlement rate is also negatively signed but not near significance (z = –0.3). These insignificant results parallel the weak bivariate correlations cited earlier. In these initial models, the targeted welfare measures are not as effective at reducing single-mother poverty as the measures of welfare universalism.
The next models omit the two economic controls to maintain parsimony at Level 2 with only 18 countries.14 Models 5 and 6 combine the welfare state index with the two measures of targeting. In both models, the welfare state index is significantly negative, although the effect is slightly larger in the model including the targeting ratio (Model 5). For a standard deviation increase in the welfare state index, the odds of single-mother poverty decline by a factor of 1.9–2.5. Although the single-mother entitlement variable remains insignificant in Model 6, the targeting ratio is now significantly negative in Model 5. For a standard deviation increase in the targeting ratio, the odds of single-mother poverty decline by a factor of 1.5. Thus, targeting appears to significantly reduce single-mother poverty when controlling for welfare universalism, although the effect of targeting is smaller than the effect of universalism.
Models 7 and 8 include the universal replacement rate along with the two measures of targeting. Like the welfare state index, the universal replacement rate is significantly negative in both models, and the effect is slightly larger controlling for the targeting ratio (Model 8). For a standard deviation increase in the universal replacement rate, the odds of single-mother poverty are expected to decline by a factor of 1.7–2.5. Also like Models 5–6, single-mother entitlement remains insignificant, but the targeting ratio becomes significantly negative. For a standard deviation increase in the targeting ratio, the odds of single-mother poverty decline by a factor of 1.7. In Model 7, the effect of the universal replacement rate is larger than the effect of targeting. However, targeting reduces single-mother poverty when controlling for the more consequential universal replacement rate.
To further illustrate the influence of targeting and universal welfare, it is helpful to consider the difference between the United States and other affluent democracies (cf. Table 2). If the United States increased its welfare state index to the cross-national mean while holding all other variables constant at their means, the odds of single-mother poverty would decline by a factor of 3.8 (i.e., Model 5). If the United States increased its welfare state index to Sweden’s level (the most generous welfare state), the odds of poverty for single mothers would decline by a factor of 13.3. If the United States increased its universal replacement rate to the cross-national mean or to Sweden’s level, the odds of single-mother poverty would be expected to decline by factors of 4.5 and 17.9, respectively (based on Model 7). Hence, the poverty reduction resulting from the United States moving to an average level of either measure of welfare universalism is larger than the effect of any other variable except for that of no one employed in the household. The poverty reduction resulting from the United States moving to Sweden’s level of welfare universalism would be larger than the effect of any variable.
Conversely, if the United States substantially increased its targeting ratio, single-mother poverty would decline more modestly. If it increased its targeting ratio to the cross-national mean, the odds of single-mother poverty would be expected to decline by a factor of 1.3 (i.e., Model 7). Even if the United States increased its targeting ratio to Australia’s level (the highest ratio), the odds of single-mother poverty would decline by a factor of 5.1. Further, Table 3 provides no evidence that increases in the single-mother entitlement would reduce single-mother poverty.
Supplementary Analyses
As discussed earlier, a longstanding concern has been that generous social policy encourages single motherhood or poverty-worsening qualities of single motherhood, such as nonemployment or additional children. To determine whether social policy has counterproductive effects, Table 4 assesses the effects of the universal replacement rate or the targeting ratio on (a) whether an 18- to 54-year-old woman resides in a single-mother household (sampling all or only those with children in the household); and (b) whether a single-mother household contains multiple earners, (c) no employed people, or (d) more children. Because these characteristics influence single-mother poverty, if social policy has counterproductive consequences, the two measures should have significant positive effects in Models 1, 2, 4, and 5, and negative effects in Model 3.15
The first two models assess whether social policy influences the odds that an adult woman resides in a single-mother household. In both models, the universal replacement rate is negatively signed but not significant. The targeting ratio, however, is positively signed and nearly significant. Thus, universal social policy does not encourage single motherhood. Because the targeting ratio is also positive and significant for lone motherhood in Table 6 in the appendix, there is some concern that targeting encourages single motherhood. However, one should be cautious about claiming a causal effect of targeted social policy on single motherhood. We stress that the important conclusion is that welfare universalism does not counterproductively increase the presence of single motherhood.16
The last three models assess whether social policy discourages having multiple earners, or encourages nonemployment or having more children. In all three models, there is no evidence that the universal replacement rate has counterproductive effects. This is also the case for the welfare state index (not shown) and in models of lone mothers (see Table 7). However, there is some concern with the counterproductivity of targeting (although there is not for the single-mother entitlement; not shown). The targeting ratio is positive and significant for nonemployment in Table 4, and for lone-mother nonemployment in Table 6. One should be cautious about interpreting this as causal because of potential endogeneity. Single mothers may receive greater welfare benefits because they are more likely to be unemployed, so the targeting ratio is probably positively influenced by unemployment differences between single mothers and others. Ultimately, there is no evidence that universal social policy measure has counterproductive employment or parenting consequences, and there is only suggestive evidence that targeting might indirectly worsen single-mother poverty.
Finally, we also considered alternative measures of the economic standing of single mothers. Table 7 in the appendix shows the effects of social policy on two alternative measures of poverty (40% and 60% of the median) and three measures of household income (a binary measure of whether income is greater than the median, and the ratios of income or logged income over the median). Although social policy does not affect whether a single-mother household has an income greater than the median, the other alternatives present results consistent with those cited earlier.
Conclusion
This study provides one of the first multilevel, cross-national analyses of single-mother poverty. Incorporating data on 18 affluent democracies, we assess the effects of individual characteristics and social policy. We show that single mothers are disproportionately more likely to be poor in all 18 countries. Our analyses reveal that both individual characteristics and social policy shape single-mother poverty. Single-mother households with multiple earners, well-educated and older heads of household, and multiple adults are less likely to be poor. Those with no one employed, low-educated and younger heads of household, and multiple children are more likely to be poor.
Our central conclusion is that generous, comprehensive, and universal welfare states substantially reduce the poverty of single mothers. The welfare state index and universal replacement rate are strongly negatively associated with single-mother poverty. If the United States increased its welfare state index to the cross-national mean or to Sweden’s level, the odds of single-mother poverty would decline by a factor of 3.8 or 13.3, respectively. If the United States increased its universal replacement rate to those levels, the odds of single-mother poverty would decline by a factor of 4.5 or 17.9, respectively. As noted earlier, these effect sizes are large in comparison with the individual-level variables. Although policy and demographic debates often focus on altering the behavior or characteristics of single mothers (e.g., encouraging education, employment, having fewer children, and marriage), welfare universalism could be an even more effective anti-poverty strategy.
In addition, universal social policy is much more effective than targeted social policy. Although the targeting ratio is significantly negative when controlling for welfare universalism, its effect is modest relative to the two measures of welfare universalism. Single-mother entitlement is never significant, and the targeting ratio is insignificant when universalism measures are omitted from the model. Moreover, there is some evidence that targeting may be counterproductive, whereas there is no evidence that universalism is counterproductive. Weighing these various results, the analyses support welfare universalism over targeting for reducing single-mother poverty.
Scholars have argued that welfare universalism is more effective because universalist social policies tend to be larger, in part because they garner greater political support. Universalism is also associated with less poverty for all groups, thus lower single-mother poverty may be a byproduct of that broader social equality. As Table 2 shows, there is a positive association between overall and single-mother poverty rates. There is also a positive association between single-mother poverty rates and the Gini index of income inequality (r = .64; details available upon request). Notably, the two measures of targeting are negatively associated with the two measures of universalism across the 18 countries (cf. Table 2). This is consistent with Korpi and Palme’s (1998) “paradox of redistribution”: countries with more-targeted social policy have less-generous overall welfare states. Further, welfare universalism may alleviate single-mother poverty because the complex of integrated and interdependent universal welfare programs better protects against heterogeneous risks than targeted means-tested assistance. Interestingly, we find that targeting may actually be effective in a context of welfare universalism, as suggested by the significantly negative targeting ratio in Models 6 and 8 of Table 3. Plausibly, when social policy is already generous for all citizens and the odds of poverty for all are low, it may be beneficial to direct even greater benefits to single mothers relative to average residents.
Future research can address the limitations of the present analysis. First, it would be valuable to explore these relationships with longitudinal data on individuals and/or countries. Although the LIS is cross sectional, it offers at least five data sets for these 18 countries, and one could investigate how single-mother poverty and social policy have changed over time. Second, with a few exceptions (e.g., Behrendt 2000), the literature on targeted and universal social policies has not been empirical. As a result, it would be productive to apply this research design to other vulnerable groups to assess the effectiveness of targeted and universal social policies. Third, qualitative and mixed methods could enhance our understanding of the mechanisms linking universal welfare generosity and lower single-mother poverty. It would be particularly valuable to incorporate comparative longitudinal in-depth interviews of single mothers in countries like Denmark and the United States. Finally, since the 1990s, U.S. social policy has perhaps become less targeted at single mothers, as evidenced by the decline of AFDC and the rise of food stamps as well as the EITC and other tax credits. Nevertheless, eligibility often varies depending on parental employment. Thus, there continues to be a need for research on how U.S. social policy provides incentives and disincentives for employment among parents.
We conclude by underlining one central implication of our study. As noted at the outset, an extensive demographic literature explores the relationship between single motherhood and poverty. Indeed, it would be reasonable to suggest that single motherhood may be the most well-studied correlate of poverty. Although a literature exists on cross-national differences in single motherhood, most American poverty debates about single motherhood have been based solely on studies of the United States. Our analyses demonstrate how unusual the U.S. case is, with the second-highest rates of single motherhood and the highest rate of single-mother poverty among the 18 affluent countries studied. This suggests that we should be much more cautious in generalizing from the U.S. case. Perhaps the focus on the United States by American poverty researchers has constrained our scope of vision about the nature and possible solutions to single-mother poverty. Using a cross-national comparison, our study demonstrates that the social policy context in which single mothers reside matters as much as or even more than their individual characteristics for poverty.
Acknowledgments
We thank Demography reviewers, guest editor Suzanne Bianchi, editor Stewart Tolnay, as well as Liz Ananat, Lane Destro, Andrew Fullerton, Bob Jackson, Stephanie Moller, Jennifer Moren Cross, Stephen Morgan, Emilia Niskanen, and David Reingold for assistance and suggestions.
Appendix
Notes
We code couples using the variable “married,” which includes married and nonmarried cohabiting couples (including same-sex couples). Unfortunately, the LIS does not provide sufficient information to identify the mother of the children. So, our sample includes other 18- to 54-year-old women residing in the household. We address this problem by controlling for other adults and multiple earners in the household and by estimating the models on lone mothers. Although Rainwater and Smeeding (2004:109–110) defined single-mother households simply as female-headed households with children present, we employ an even more stringent definition by including only those not married or cohabiting.
DPI includes disposable cash and noncash income after taxes and transfers (including food stamps; housing allowances; and tax credits, such as the EITC).
The categories are (a) less than secondary (low), (b) secondary or some tertiary (medium), and (c) completed tertiary or more (high). Unfortunately, the LIS does not provide sufficient detail to code vocational/technical secondary education.
In analyses available upon request, we variously add head of household’s age-squared, age of the respondent, and dummy variables for the respondent or head being under age 25. The results are consistent.
Slightly more than one-third of the sample has a child under 5, and the average single-mother household has 1.7 children (see Table 1). We define targeted benefits for a mother with a child under 3 because this maximizes the value of targeted benefits, giving this measure the best chance of being consequential (i.e., countries usually give greater benefits for young children). One could construct alternative single-mother entitlement rates for various numbers and ages of children; however, it is difficult to reduce these to one estimate per country.
For the United States, this is the mean benefit of Temporary Assistance for Needy Families (TANF) across states. One could include the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) program. However, we were unable to identify any cross-national source on means-tested in-kind benefits. We did not include food stamps and/or housing assistance because those are means-tested for all and not targeted for single mothers. Adding in-kind benefits in the United States would only raise the already above-average single-mother entitlement and thus would lead to an even less significant effect (cf. Tables 2 and 3).
We use the term “nonemployed” to make clear that we do not include unemployment benefits here. Although a nonemployed single mother might qualify for unemployment benefits, this is not a benefit targeted at single mothers. (It is targeted at the unemployed.) Moreover, many single mothers have not been previously employed long enough to qualify for unemployment insurance.
“Total government assistance” sums social insurance, social assistance transfers, alimony, and child support. To equivalize this measure, we divide by the square root of household members. We tested several derivations of this measure, and the results are robust (e.g., concentrating on social assistance targeted to low-income households and adding or subtracting social insurance, alimony/child support, child/family benefits, unemployment compensation, and maternity/family leave benefits). We present the comprehensive measure because there is often targeting implicit in what are statutorily considered universal programs.
Yet another alternative would measure the mean total government assistance received by single mothers (in each country standardized over the median) that is not received by the general population. This “absolute” measure of targeted benefits would be the difference between what the target group and general population receive. In analyses available upon request, this produces results nearly identical to the targeting ratio.
In analyses available upon request, we substitute each of these indicators as well some alternatives (e.g., family assistance as percentage of GDP). The results are consistent. Also, there is no evidence of significant interaction effects of our welfare state measures with welfare regimes or of regime main effects.
We also include the Ns for each country. Please note that a few countries have samples of fewer than 200 cases. For these (e.g., the Netherlands), the mean level of single-mother poverty should be read with caution.
The single-mother entitlement is calculated as a percent of median equivalized household income, and poverty is defined as less than 50% of median equivalized household income. Thus, France is the only country with a value higher than 50%.
For both Italy and Spain, the single-mother entitlement is 0. Both provide family assistance only as a supplement to employment earnings. For example, a single mother in Italy is eligible for family assistance if she is employed, the only wage earner in the family, and low-income. As explained in the Methods section, this measure assumes nonemployment, following the argument that this benefit is solely for being a mother with young children.
Even though unemployment was significant in Models 1 and 2, both economic context variables would be insignificant if included, and the other results would be consistent.
These models are intentionally parsimonious, including only a few individual-level controls. The results are not sensitive to the inclusion of other individual-level controls. Although we include both social policy measures in the same models, the results are robust if modeled separately. The first four are multilevel logit models, and the last is a multilevel Poisson model.
Although not shown, the welfare state index would be negatively signed and insignificant, and the single-mother entitlement would be positively signed and insignificant. As shown in Table 6, the universal replacement rate is negatively signed and insignificant for lone motherhood as well.
