The Selection of Children from Low-Income Families into Preschool

Abstract

Because children from low-income families benefit from preschool but are less likely than other children to enroll, identifying factors that promote their enrollment can support research and policy aiming to reduce socioeconomic disparities in education. In this study, we tested an accommodations model with data on 6,250 children in the Early Childhood Longitudinal Study-Birth Cohort. In general, parental necessity (e.g., maternal employment) and human capital considerations (e.g., maternal education) most consistently predicted preschool enrollment among children from low-income families. Supply side factors (e.g., local child care options) and more necessity and human capital factors (e.g., having fewer children, desiring preparation for school) selected such children into preschool over parental care or other care arrangements, and several necessity factors (e.g., being less concerned about costs) selected them into non-Head Start preschools over Head Start programs. Systemic connections and child elicitation did not consistently predict preschool enrollment in this population.

Preschool is a powerful tool for reducing the intergenerational transmission of socioeconomic inequality (Currie, Garces, & Thomas, 2002; Duncan & Magnuson, 2013; Ludwig & Phillips, 2007). An emerging consensus across the broad developmental sciences is that boosting the preschool enrollment of children from low-income families is a cost-effective and a developmentally appropriate way to reduce the socioeconomic gap in educational attainment that is so crucial to the stratification of society (Heckman, 2013; NICHD Early Child Care Research Network [ECCRN], 2005). The associated policy support for universal preschool has heightened the need to understand why some low-income parents enroll their children in preschool and others do not (Haskins & Barnett, 2010; Waldfogel, 2006).

This study tackles this “why?” question by exploring the processes by which children from low-income families are selected into preschool programs, adapting a theoretical framework from the general case of parents’ selection of child care arrangements (accommodations) to the specific case of low-income parents’ selection of preschool (Coley, Votruba-Drzal, Collins, & Miller, 2014; Meyers & Jordan, 2006). Drawing on the Early Childhood Longitudinal Study-Birth Cohort (ECLS-B), we examine five potential mechanisms of selection cutting across the supply side (e.g., neighborhood settings), the demand side (e.g., parental needs, preferences, and circumstances), and the potential role of children themselves. This developmentally-oriented approach reverses the conventional direction of research; rather than preschool effects on children, selection of children into preschool is considered. At the same time, the lens for examining selection processes is widened to better recognize how children help to create their own ecologies through the responses they evoke from adults (Sameroff, 2009).

Such research can expand theoretical knowledge about differential preschool exposure as a mechanism of socioeconomic stratification in the U.S. In the process, it can identify segments of the low-income population most in need of attention and/or best positioned to inform efforts to reduce socioeconomic disparities in early education. Our use of ECLS-B is well-suited to this goal because it provides the opportunity and power to make numerous comparisons within a large low-income sample that is internally diverse. One weakness of these data is that they include quality assessments for only a subset of preschools, but this weakness is mitigated by recent studies (including in this journal; see Gordon et al., 2013) calling into serious question the validity of many extant strategies for measuring preschool quality. Thus, this large-scale picture of preschool selection among low-income families provides a foundation that can be built on by more intensive multi-method community-based approaches in the future.

Preschool Education and Children from Low-Income Families

One of the most powerful child-focused policy trends in recent history has been the expansion of preschool opportunities for low-income children, a trend supported by converging theoretical and empirical activity across disciplines. First, conceptual models from psychology and sociology have emphasized that socioeconomic disparities in school readiness are fundamental to the transmission of socioeconomic status and inequalities across generations. The argument is that children from low-income families enter the K-12 educational system with less developed academic skills that are then acted upon by formal and informal processes of schooling (e.g., ability grouping, teacher expectations) to create larger end-of-school disparities that undermine their ultimate socioeconomic attainment (Entwisle, Alexander, Entwisle, & Olsen, 2014; Pianta, Cox, & Snow, 2007). Second, informed by such perspectives, social and behavioral scientists have effectively established the impact of preschool enrollment on a host of short- and long-term outcomes, including school readiness. Notably, this impact is heightened for children from low-income families (Duncan & Magnuson, 2013; Gormley, Gayer, Phillps, & Dawson, 2005; Peisner-Feinberg et al., 2001; Waldfogel, 2006; Winsler et al., 2008). Third, economists have demonstrated that early childhood education programs can bring greater long-term returns to investment than other kinds of intervention targeting socioeconomic disparities in educational and occupational attainment (Heckman, 2013; Karloy, Kilburn, & Cannon, 2006).

With such supporting evidence, federal, state, and local governments have built on the model of Head Start and other pioneering early childhood education programs by expanding extant preschool programs and creating new ones with the goal of increasing the proportion of the low-income population enrolled in early education (Gormley et al., 2005; Love et al., 2005; Puma et al., 2005; Zigler, Gilliam, & Jones, 2006). Such efforts have achieved some success, but enrollment remains far from universal. Of all 3- and 4-year olds from eligible low-income families in the U.S., less than half are enrolled in Head Start, and over one-third are not enrolled in any form of preschool (Duncan & Magnuson, 2013; Haskins & Barnett, 2010). Unlike children from middle class families, for whom preschool has become more or less a normative stage of the educational career, children from low-income families often enter the K-12 system without prior educational experience (Clarke-Stewart & Allhusen, 2005).

The argument of this study is that the policy goal of increasing the enrollment of children from low-income families in preschool can be promoted by theoretically grounded research about which children are enrolled and which are not. In addition to shedding light on mechanisms for expanding and equalizing preschool access, this approach can identify qualifiers to what is already known about the significance of preschool. For example, uncovering the mechanisms that select children into preschool could support efforts to increase the prevalence or reach of that factor in the low-income population. It might also suggest that our empirical documentation of the effects of preschool on children has been underestimated or even mistaken to the extent that this selection factor has not been sufficiently taken into account analytically.

Our theoretical starting point is the accommodations framework, a conceptual model for understanding how parents use child care. In this model, originally formulated by Meyers and Jordon (2006), the situating of children in various care settings is conceptualized as reflecting the ways in which their parents accommodate to opportunities and constraints that they face as a result of their cultural and social contexts as well as their own capacities and circumstances; in other words, a person × context interplay. The model has four general mechanisms of selection into care settings. The first two are family needs (e.g., maternal employment) and family resources (e.g., income), which determine what parents need for care as well as the information that they have about care options and their access to various types of care. The second two deal with the larger cultural context—cultural norms and preferences (e.g., racial heritage) that affect both availability and accessibility of care and contextual factors that affect the potential supply of care options in a community. This framework is important because it moves away from viewing child care arrangements as an explicit choice and manifestation of preferences and instead recognizes that they arise out of specific developmental and ecological situations.

The tenets of the accommodations framework have been supported and expanded by empirical research, including a recent study by Coley and colleagues (2014) showing that the salience of mechanisms shift as children grow older and the normativeness and availability of non-parental care increases. Specifically, necessity and supply considerations become relatively less important and education-related concerns become more important as the focus shifts from care itself to preparation for school. In this way, the developmental period just prior to school entry is a unique time in the accommodation process with critical implications for the future. Although this framework includes socioeconomic status as a specific form of the resources mechanism, studying how the overall framework works within socioeconomic strata could be useful. For example, low-income parents face more constraints on preschool options than affluent parents because of inadequate funds, but they also have more preschool opportunities than many who are not technically poor but still economically distressed due to access to free government-funded preschool programs geared to the low-income population (Capizzano & Adams, 2004; Duncan & Magnuson, 2013). Thus, the accommodations framework is especially salient to the preschool experiences of children from low-income families.

A Model of Preschool Selection

This study continues the work by Coley and colleagues (2014) by expanding the accommodations framework while focusing on preschool in the low-income population. We do so by specifying five mechanisms of selection, beginning with the contextual and systemic one, moving through the parent-focused ones, and ending with the child-focused mechanism.

First, supply-side factors map onto the contextual factors in the original framework but also include parental perceptions of options in the community, not just the concrete physical and compositional aspects of the community context. On one hand, what matters are the characteristics of a community that increase the overall demand for preschool services, which then lead to more and better options in the early childhood program market, which—all else equal—facilitates the flow of children into preschool. Neighborhoods with more socioeconomically advantaged adult populations and higher numbers of children would fit such a community profile (Coley et al., 2014; Gordon & Chase-Lansdale, 2001; Dupere et al., 2013). On the other hand, parents’ perceptions of their neighborhoods—regardless of whether these perceptions are accurate—affect how they interact with their neighborhoods (Sharkey & Sampson, 2010). If parents do not think that they have good preschool options, they may not seek out any preschool and, instead, rely on informal arrangements. When parents live in places in which there is a high demand for quality preschool and when they think that there are good options available to them, they likely delve more deeply into the preschool market.

Second, systemic connections make up a new mechanism that taps into parents’ institutional engagement, a phenomenon hinted at more narrowly in the original accommodations framework. For example, Coley and colleagues (2014) used parental enrollment in health and human services programs, like Temporary Aid to Needy Families (TANF) as an indicator of families’ lack of economic resources. Another way to view such enrollment is to consider how it connects parents to a broad system of programs, activities, and organizations that they must learn to navigate and that introduces them to people (e.g., caseworkers) who may be able to help them locate other kinds of services for families in need. Such aspects of program enrollment may be especially relevant to parents who have the least knowledge about available supports and opportunities (Edin & Lein, 1997; Kalil & Ryan, 2010; Perreira et al., 2012). Thus, enrollment in government programs may be a sign of economic disadvantage, but it could also differentiate among disadvantaged families—low-income parents who have access to information that others may not. This differentiation could be relevant to understanding why some low-income parents are able to locate and access preschool options, as their connections to institutional networks may inform their actions. Mothers who are enrolled in various government programs and related services would then be more likely to have children in preschool than other low-income mothers.

Third, parental necessity maps onto the original needs mechanism but goes beyond family circumstances to include what parents think is right for their children. At issue is how much preschool is viewed as a form of child care (versus an educational investment) and, therefore, subjected to similar demand forces. Preschool is an explicitly education-focused fee-based setting that allows parents to work and manage other time constraints. That preschool could fill these needs while also providing educational enrichment would seem to make it a viable alternative to other care arrangements for children near school-entry. Moreover, given that public programs are free or subsidized, preschool could be a preferred alternative for low-income parents (Brooks-Gunn, Han, & Waldfogel, 2010; Clarke-Stewart & Allhusen, 2005; Gordon, Kaestner, & Korenman, 2008). The factors that might make low-income mothers need care for a child, therefore, would likely increase the odds of the child being in preschool. Among the most important are maternal employment, marital status, and number of children. Mothers who work for pay, are not partnered with the child’s other parent, have other children, and require the most convenient kinds of care would likely be in greater need of preschool doubling as non-parental care (Crosnoe et al., 2014; Coley et al., 2014; Morrissey, 2008; NICHD ECCRN, 2005).

Fourth, human capital considerations is a new mechanism that pulls educational factors from the resources mechanism for closer examination as a multi-faceted driving force encompassing both resources and preferences. What matters here is the extent to which preschool is viewed as a way to invest in children’s futures. Such views could indicate an awareness of preschool enrollment as a potential means of giving children a competitive edge in the K-12 system. The factors that might help low-income parents be more strategic, therefore, would likely increase the odds of their children enrolling in preschool (Augustine, Crosnoe, & Cavanagh, 2009; Grogan, 2012). One factor is maternal education. Regardless of how education affects need for preschool as a form of child care, it shapes how mothers—especially mothers from more disadvantaged segments of the population—manage and invest (in terms of time, money, effort) in their children’s educations. Another factor is mothers’ educational expectations for children. Especially in low-income communities, such expectations can encourage mothers to be more agentic in creating opportunities for youth (Furstenberg et al., 1999; Lareau, 2003; Magnuson, 2007; Morrissey, 2008). Enrollment in preschool could follow a similar pattern.

Fifth, child elicitation explicitly focuses on the role of children in parents’ decision-making that is subsumed within needs and preferences in the accommodations framework. The idea is that children help to create their ecologies (Bell, 1969; Belsky, 1984). For older youth, elicitation can be active, making demands of parents or making decisions themselves. For young children, it is more passive, with children’s traits and behaviors (including those with genetic underpinnings) drawing responses (Sameroff, 2009; Stattin & Kerr, 2000). When looking at preschool enrollment, such elicitation could take two seemingly orthogonal forms with the same underlying motivation of doing what is best for the child. Compensatory elicitation, for example, would occur when children’s skills and behaviors are low or problematic enough to make parents think that they need help—the child seems delayed or is difficult to manage. Enrichment elicitation, on the other hand, would occur when children’s skills and behaviors are advanced enough to motivate parents to take extra steps to support them—the child seems unusually advanced and mature (Clarke-Stewart & Allhusen, 2005; Pianta & Walsh, 1996; Tucker-Drob & Harden, 2012). Neither form is specific to any socioeconomic group. Yet, given how many economic, cultural, and social supports select children from middle class homes into preschool, such child effects may be less important overall for them. At the same time, child effects might be more likely to make a difference in the low-income population, where preschool enrollment is more variable and contingent on other circumstances (Augustine & Crosnoe, 2011).

Based on this model, our first objective is to compare indicators from five mechanisms of preschool selection in low-income families. Results can provide evidence of finer-grained mechanisms in socioeconomic disparities in education and suggest possible confounds in documented preschool effects on children. At the same time, by shedding light on heterogeneity in the low-income population, they can demonstrate the ways that low-income parents actively try to improve children’s future prospects despite the many constraints they face, serving as a counterpoint to often negative depictions of low-income parents in research on children.

Differences by Type of Preschool

Up until now, we have treated preschool as a single category of early educational experience, but it comes in many forms, including private and public. The former provide educational services for fees, whereas the latter are free or have subsidized rates and can be affiliated with public schools. The two offer different incentive structures for low-income families that can affect supply and demand (Haskins & Barnett, 2010; Zigler et al., 2006). Unfortunately, just as parents often have trouble differentiating between private and public beyond Head Start, differentiating between the two in ECLS-B is difficult with the exception of identifying children in Head Start (Gordon et al., 2014). In exploring preschool selection in the low-income population, we expect differences between Head Start and other preschools.

Head Start is a well-publicized program with great name recognition and demand, especially in low-income communities. It also has the stated goal of launching children onto the path to higher education, and it has a holistic two-generation focus that includes services and programs for parents and emphasizes healthy development as well as learning (Ludwig & Phillps, 2007; Puma et al., 2010; Zigler et al., 2006). Other preschool programs available to low-income families are unlikely to have such name recognition or public awareness of goals and practices. Hence, enrollment in them requires parents to do more legwork and research and draw more on social networks for information. Options are less clear-cut and familiar (Augustine et al., 2009; Clarke-Stewart & Allhusen, 2005). Moreover, most non-Head Start programs are not free and can be expensive. Perhaps reflecting these differences, past research suggests that the children who enter Head Start often differ from children who attend other types of programs, demonstrating more behavioral problems and having more disadvantaged parents (Lee, Zhai, Brooks-Gunn, Han, & Walfogel, 2014; Zhai, Brooks-Gunn, & Waldfogel, 2011).

The balance between the benefits and costs of Head Start suggests that the necessity mechanism might be stronger for this type of preschool among low-income families—parents use it because it meets child care needs in a potentially beneficial way that does not add to economic stress. The less clear-cut options and attributes of non-Head Start preschool increase uncertainty and require stronger motivations to overcome. The needs of children and larger contextual settings are likely an important motivational source in these circumstances, suggesting that child elicitation and the contextual mechanisms might be stronger for non-Head Start preschool—parents find programs for children who need them, especially when programs are more available and they have help finding them. Given the emphasis on future educational prospects in Head Start as well as the possibility that some non-Head Start programs may be higher-quality overall (if low-income parents can find them), the human capital mechanism is less likely to differ across the two types of preschool than the other focal mechanisms.

A second objective of this study, therefore, is to examine the degree to which preschool type moderates the five mechanisms of selection into preschool among children from low-income families. We look at the basic distinction between Head Start and non-Head Start programs, speculating that necessity will matter more in the former, child effects, supply-side factors, and systemic connections more in the latter, and human capital equally across the two.

Methods

The National Center for Education Statistics (NCES) designed ECLS-B to follow a representative sample of children born in the U.S. in 2001. It used a clustered design with birth certificates from the National Center for Health Statistics. Children were excluded if they were born to mothers under 15 years of age or if they died or were adopted before 9 months of age. An initial sample of 14,000 births yielded a final cohort of 10,700 children when they were 9 months of age. The cohort was assessed in the home when the children were 9 months, 2 years, and 4 years (preschool wave of data collection) and then upon entry into kindergarten. NCES requires that all reported sample and cell sizes be rounded to the nearest 50, which we do here.

Beginning with a preschool-wave sample of 7,800, our analytical sample included 6,250 children in this wave who met inclusion criteria. They were singleton births not yet enrolled in kindergarten, excluding 1,400 and 150 cases respectively. From here, our sample included 3,000 children from low-income families (income at or below 185% of the federal poverty line for family size). For comparison, we had a sample of 3,250 children from families above 185% of this line. To address non-response bias and ensure that our sample was representative of the large population of children, we used appropriate sampling weights. We employed missing data estimation to avoid losing additional cases through item- or instrument-level missingness.

Measures

The dependent variables came from the preschool wave when children were 4. Except in a few cases otherwise noted, predictors also came from this same wave. See Table 1 for descriptive statistics for all study variables—for the full sample and by income status.

Table 1.

Descriptive Statistics for Study Variables, Full Sample and by Family Income Status

	Full sample (n = 6,250)		Low-income (n = 3,000)		Not low-income (n = 3,250)		Income group difference
	M	SD	M	SD	M	SD	Sig.
Low-income family	.47
Enrollment (age four)
Enrolled in preschool	.58		.38		.76		***
Enrolled in Head Start	.12		.25		--
Other non-parental care	.11		.13		.10		***
Parental care only	.19		.24		.14		***
Covariates
Child gender (male)	.51		.52		.50
Child age (months)	52.92	4.15	53.03	4.20	52.82	4.11	*
Mother age at birth (years)	27.41	6.35	25.08	6.14	29.56	5.76	***
Mother born outside U.S.	.25		.24		.25
Mother race/ethnicity
White, Non-Latino/a	.47		.34		.56		***
Black, Non-Latino/a	.15		.24		.08		***
Latino/a	.17		.25		.11		***
Asian American	.14		.08		.20		***
Native American	.04		.06		.02		***
Multiracial/ethnic	.03		.04		.03		*
Prior child care (age 2)
Center-based care	.16		.13		.19		***
Parent care	.50		.56		.44		***
Relative care	.20		.21		.18		**
Non-relative care	.14		.09		.18		***
Supply side
Average income in zip code	56.51	22.31	47.29	13.79	65.07	25.11	***
% zip code with B.A. or higher	.24	.16	.18	.11	.30	.17	***
% zip code employed	.65	.08	.62	.08	.66	.08	***
% zip code < 18 years old	.39	.10	.40	.10	.39	.10	*
Perceived good care options	.79		.75		.82		***
Systemic connections
Receipt of SNAP	---		.62		---		---
Receipt of TANF	---		.31		---		---
Receipt of Medicaid	---		.82		---		---
Receipt of CHIP	---		.20		---		---
Parent had job training	---		.28		---		---
Help with housing	---		.25		---		---
Parental necessity
Maternal employment
Employed full-time	.42		.35		.48		***
Employed part-time	.18		.17		.19
Looking for work	.06		.10		.02		***
Not in labor force	.34		.37		.31		***
Mother not in household	.01		.01		.005		**
Household structure
Father in household	0.79		.66		0.91		***
Number of children under 18	2.40	1.14	2.67	1.28	2.15	.93	***
Preschool/care preferences
Care when child sick	2.29	.87	2.54	.76	2.06	.91	***
Close to home	2.56	.63	2.63	.60	2.49	.65	***
Reasonable cost	2.64	.57	2.77	.51	2.52	.59	***
Flexible hours	2.38	.75	2.53	.68	2.23	.79	***
Human capital considerations
Maternal education
Less than high school	0.14		0.26		0.03		***
High school graduate	0.26		0.38		0.15		***
Some college	0.30		0.28		0.32		***
Bachelor's degree	0.30		0.08		0.50		***
Education expectations for child
High school diploma or less	0.10		0.17		0.04		***
Attend some college	0.14		0.19		0.10		***
Complete a B.A.	0.42		0.32		0.50		***
Earn M.A. or higher	0.34		0.31		0.36		***
Preschool/care preferences
Prepare child for kindergarten.	2.81	.46	2.83	.44	2.79	.47	**
Small number of children	2.62	.57	2.57	.61	2.67	.53	***
Child elicitation
Mental skills
1st quartile (low)	0.26		.34		.18		***
Middle quartiles	0.50		.51		.49		*
4th quartile (high)	0.24		.15		.33		***
Motor skills
1st quartile (low)	0.27		.29		.25		***
Middle quartiles	0.49		.48		.50
4th quartile (high)	0.24		.23		.25		*
Child negativity toward parent
Low negativity	0.76		0.71		0.80		***
Moderate negativity	0.16		0.19		0.14		***
High negativity	0.08		0.11		0.06		***
Child persistence
Consistently lacks	0.06		0.08		0.04		***
Typically not	0.17		0.20		0.13		***
Half the time	0.26		0.27		0.24		**
Typically is	0.41		0.36		0.46		***
Consistently is	0.10		0.09		0.12		***

^*p <.05;

^**p <.01;

^***p <.01

Note: SNAP = Supplemental Nutrition Assistance Program; TANF = Temporary Assistance for Needy Families; CHIP =Child Health Insurance Program. Ns rounded to the nearest 50 per NCES requirements for ECL-B data.

Family income status

All analyses were stratified by household poverty status, assessed at the preschool wave. Children were categorized by whether or not they lived in a household with an income less than 185% of the U.S. Census federal poverty threshold for a family of their size. Information about household income was collected as part of the preschool CAPI instrument survey with a categorical series of detailed-range income questions. NCES created measures for household poverty by comparing the midpoint of the parent-reported category for 2005–2006 income with the 2005 weighted poverty thresholds from the U.S. Census Bureau. Within the sample of families identified as low-income, the income range was from less than $5,000 to less than $50,000 (for the largest families), with a mean between $20,000 and $25,000.

Preschool enrollment

Parents reported on all childcare arrangements that their children attended on a regular basis at the time of the preschool parent interview. They were asked about four categories separately, which we used to create four mutually exclusive groups. The focal category was parent-reported enrollment in a preschool (labeled preschool or pre-kindergarten). It was compared to parent-reported enrollment in: a) Head Start; b) other forms of child care (relative, non-relative); and c) parent care. Note that we did not examine comparisons with Head Start enrollment for the non-low income sample. Importantly, although we created mutually exclusive groups, parents could report on multiple arrangements, as the question asked them to identify all childcare settings that their children attended on a regular basis. Consistent with prior research (Bumgarner & Brooks-Gunn, 2014; Lee et al., 2014), our coding prioritized preschool, so that children were included in this category if they attended preschool even if they were also in Head Start (only 50 were in both) and/or had other child care arrangements. In other words, if children were currently attending any preschool, they were assigned to the preschool category. This categorization scheme meant that about 500 children who were in preschool and relative care were put in the preschool category rather than the other care category. If children were currently attending Head Start (but not another kind of preschool), they were assigned to the Head Start category. If they were currently receiving care from relatives or non-relatives, in or outside of their home, and were not in preschool or Head Start, they were assigned to the other care category. If they were receiving no non-parental care, they were placed in the parent care only category. Of note is that, technically, many of the children in our analytical sample would not be eligible for Head Start given our use of the 185% income threshold to measure poverty rather than the poverty line itself. Yet, as many as 4 of 10 children enrolled in Head Start come from families with incomes above the poverty line, most likely because of income volatility over time (West et al., 2010). In the ECLS-B, 40% of children in Head Start came from families with incomes between 100–185% of the poverty line. We believe that studying Head Start in this more inclusive sample is informative and note that controlling for whether a family had incomes higher than the line did not affect the results reported below.

Supple side indicators

We included four Census variables based on the zip code where families lived to examine how community characteristics influence childcare choices. They were average household income, the percentage of adult residents with at least a Bachelor’s Degree, the percentage of adults employed, and the percentage of households with a member under the age of 18. Parents also reported on whether or not there were good choices for childcare where they live. Parents could respond yes, no, or don’t know; we chose no as the reference category and do not report results for don’t know responses due to their low frequency.

Indicators of household systemic connections

We included indicators for receipt of four types of federally-provided benefits: 1) Food Stamps (now known as the Supplemental Nutritional Assistance Program), 2) Temporary Aid for Needy Families (TANF), 3) Medicaid, and 4) Children’s Health Insurance (CHIP). Parents also reported whether they had received job training, and whether they had received help with housing. Parents were asked about receipt of these benefits and services at 9 months, 2 years, and the preschool wave. We combined information from these time points to create variables that indicate whether the family had received the benefit/service since the focal child was born. Because of the strong link between poverty and many such factors, we did not examine systemic connections in the comparison models of children who were not from low-income families.

Indicators of maternal necessity

Indicators of maternal employment were developed for mothers at the preschool survey who were (1) not employed outside the home, (2) employed part-time (< 35 hours per week), (3) employed full time outside the home (≥ 35 hours per week), or (4) actively seeking employment. Measures of household structure at the preschool wave included variables for whether there was a residential father in the household (biological or not) and the number of children under 18 years of age living in the household. At the preschool wave, mothers also reported on desired characteristics of their child’s care setting, some of which tapped into necessity. They were asked how important it was that their child care arrangements would take care of their child when they are sick, were close to their home, were of reasonable cost, and had flexible hours. Response options were not too, somewhat, and very important.

Indicators of human capital considerations

Maternal educational attainment was reported by the mother or primary caregiver as part of the preschool parent survey. The categories ranged from 1 (less than a high school degree) to 4 (completion of a bachelor's degree or a higher professional or academic degree). Parents were first asked about their educational expectations for the child at the preschool assessment. Response categories ranged from 1 (to receive less than a high school diploma) to 6 (to finish a Ph.D., M.D., or other advanced degree). The lowest and highest categories were collapsed to ensure adequate sample size. Thus, the first category included parents who expected their child to complete a high school degree or less and the fourth category included parents who expected their child to complete any type of graduate degree. Two parent-reported items on desired child care characteristics (preschool wave) tapped human capital considerations. They assessed the importance parents placed on the childcare setting preparing the child for kindergarten and the importance of having a small class size.

Indicators of child elicitation

To ease interpretation, allow for consideration of nonlinearities that would be otherwise masked, and be consistent with the measurement of other sets of variables (e.g., necessity, human capital), child elicitation factors were measured in a categorical scheme, although some were more naturally quasi-continuous. Given that doing so might raise concerns about lost variance, results for categorical variables were compared to those for continuous variables. Although some marginal differences were observed, the general pattern was the same, and so the categorical approach was used in the final analyses.

NCES used a short form research edition of the Bayley Scales of Infant Development: Second Edition to assess child mental and motor skills when children were between 23 and 25 months of age. Item response theory (IRT) calibration and scoring allowed the development of age-standardized T-scores. We used procedures for complex sample survey data to calculate cut offs for quartiles using the T-scores. Estimated models included indicators for the first (lowest) and fourth (highest) quartiles in comparison to a reference group composed of children having scores between the 25th and 75th percentiles of the respective distributions for mental and motor skills. Child negativity toward parent was assessed with the two bags task, in which parents played with children using a book from one bag and then play materials from a second bag. Videos of these 10-minute interactions were then coded for various themes. For negativity, trained personnel coded the degree to which the child showed anger, hostility, or dislike toward the parent. Scale scores ranged from 1 to 7, with scores at the high end indicating that the child was repeatedly and overtly angry with the parent. Ordered categories were developed to contrast low negativity (score = 1) with moderate (score =2) and high (score ≥ 3) levels of negativity. The measure of child persistence in tasks was based on interviewer ratings of children’s persistence behavior during the Bayley administration. Ordinal categories developed for these analyses were: (1) consistently lacks persistence, (2) typically not persistent—one or two instances of persistence, (3) lacks persistence half the time, (4) typically persistent—lacks persistence in one or two instances, or (5) consistently persistent (reference group for analyses).

Child/household characteristics

Child gender and age at the preschool wave were covariates. At the 9 month interview, mothers designated their race/ethnicity with U.S. Census categories (Office of Management and Budget, 1997), from which we developed five mutually exclusive subgroups: (1) non-Hispanic Whites; (2) non-Hispanic Blacks; (3) Hispanics; (4) Asians, Pacific Islanders, and Native Hawaiians; and (5) mothers who reported being multiethnic and/or multiracial. Mother’s age at the time of the child's birth was collected by the hospital as entered on the child's birth certificate. A measure of mother's nativity was based on maternal reports of her birth country collected at the age 2 data collection. Information regarding the mother's country of birth was also available from the child’s birth certificate and was used to fill in missing values. A binary variable was created for mothers born in versus outside the U.S.

Prior child care arrangements

Child care reports from parents when children were 2 gauged continuity in non-parental care over time. Categories for this covariate included (1) no non-parental care, (2) relative care, (3) non-relative care, and (4) center-based care.

Plan of Analyses

A sequential modeling strategy examined the degree to which the five sets of focal selection factors (supply, systemic connections, necessity, human capital, child elicitation) predicted child enrollment in preschool at age four. All statistical analyses were stratified by family income to examine the extent to which potential determinants of preschool enrollment differed between children from low-income families and children from higher-income families.

For children from low-income families, logistic models identified significant predictors of non-Head Start preschool enrollment at age 4 compared to Head Start, other care, and sole parental care. For children from non-low-income families, binary logistic models were used to examine potential determinants of non-Head Start preschool enrollment versus non-parental care and other forms of care, given the income restrictions on Head Start. Each set of predictors was estimated in a separate model, with only those variables and the background covariates included. Then, a model including all potential predictors was estimated, to assess which predictors remained significant when all other pathways were included in the model.

Models were estimated with Mplus procedures for complex survey data, which corrected for the clustering of children within primary sampling units (a design effect that can deflate standard errors) and allowed for weighting to ensure sample representativeness and adjust for nonrandom attrition across waves (Muthén & Muthén, 2010). Full information maximum-likelihood estimation (FIML) addressed missing data. Simulation studies have shown that parameters estimated by FIML are unbiased and more efficient than those estimated by ad hoc methods, such as listwise deletion. FIML employs iterative algorithms and uses all available data for each observation to directly estimate parameter values, which maximizes the probability of producing the observed data (Enders, 2001; Graham, 2009).

Results

Referring to Table 1, the descriptive statistics indicate that, as expected, children from low-income families were less likely to be enrolled in a non-Head Start preschool (38%) in the year before school entry than children from higher-income families (76%). An additional 25% of children from low-income families were enrolled in Head Start, meaning that the majority (63%) were attending some form of private or public preschool. This 13-point income-related difference in formal preschool enrollment was statistically significant. Low-income families also differed from higher-income families on many factors expected to influence preschool enrollment. As a few examples, low-income mothers were younger, less educated, more likely to be non-Latina Blacks or Latinas, less likely to be working or married, more likely to work in socioeconomically disadvantaged neighborhoods, and they generally (but not uniformly) had lower educational expectations for their children. Children from low-income families also scored lower on cognitive (but not motor) assessments and were generally rated as more negative towards parents and less task-persistent.

Preschool Enrollment among Children from Low-Income Families

Table 2 includes the results of the regression models predicting preschool enrollment in the year before kindergarten entry versus being in sole parent care, Head Start, or another out-of-home care. These results are for children from low-income families only and were culled from a series of models comparing non-Head Start preschool enrollment to the other categories, so that coefficients in each column refers to the likelihood of enrollment in non-Head Start preschool relative to the other category listed at the top of the column. We have included odds ratios along with the unstandardized coefficients. The difference between the odds ratio and 1 is multiplied by 100 to give the percent change in non-Head Start preschool enrollment (henceforth referred to as preschool enrollment) associated with a one-unit change in the predictor. Because predictors were dichotomized or categorized into dummy variables, interpretation of a one-unit change in a predictor is straightforward, as is the comparison of effect sizes between two predictors.

Table 2.

Predictors of Preschool Enrollment among Children from Low Income Families

	Likelihood of being in non-Head Start Preschool versus…
	Parent care only				Head Start				Other types of care
	B	SE	Sig.	OR	B	SE	Sig.	OR	B	SE	Sig.	OR
Covariates
Child gender (male)	.25	.16		1.28	−.03	.13		.97	−.13	.20		.88
Child age (months)	.06	.02	**	1.07	.04	.02	*	1.04	.10	.02	***	1.10
Mom age at birth (yrs.)	.01	.02		1.01	.02	.01		1.02	.04	.02	*	1.04
Mom born outside U.S.	.05	.28		1.05	−.62	.24	*	.54	−.39	.30		.68
Maternal race/ethnicity
Non-Latino/a Black	.22	.27		1.24	−.46	.22	*	.63	.05	.26		1.05
Latino/a	.01	.31		1.01	−.50	.23	*	.61	−.10	.25		.91
Asian American	.45	.40		1.56	.47	.40		1.59	.43	.36		1.54
Native American	.52	.45		1.68	−.48	.45		.62	−.44	.45		.65
Multiracial/ethnic	.56	.59		1.75	−.58	.34		.56	.94	.61		2.55
Prior child care (age 2)
Center-based care	.64	.32	*	1.89	.17	.20		1.18	.81	.33	*	2.26
Relative care	.52	.26	*	1.67	−.20	.18		.82	−.75	.19	***	.47
Non-relative care	−.11	.27		.89	.02	.26		1.02	−.45	.26		.64
Child care supply side
Average income in zip code	.00	.01		1.00	.00	.01		1.00	−.01	.01		1.00
% zip code with B.A. or higher	−.37	1.63		.69	.10	1.23		1.11	.30	1.51		1.35
% zip code employed	−2.82	1.28	*	.06	.54	.94		1.72	−2.49	1.24	*	.08
% zip code < 18 years old	−.65	1.24		.52	−.24	.93		.79	.27	1.11		1.31
Perceived care options as good	.56	.21	**	1.76	.14	.16		1.15	.60	.20	**	1.83
Systemic Connections
Receipt of SNAP	−.08	.21		.93	−.33	.17		.72	−.25	.25		.78
Receipt of TANF	−.23	.23		.80	−.28	.21		.76	−.50	.28		.61
Receipt of Medicaid	−.45	.25		.64	−.46	.23	*	.63	.10	.29		1.10
Receipt of CHIP	−.26	.19		.77	−.01	.14		1.00	.12	.24		1.13
Parent had job training	.26	.20		1.30	.14	.15		1.15	−.22	.24		.80
Help with housing	.30	.26		1.35	−.23	.16		.80	.46	.23	*	1.58
Parental necessity
Maternal employment
Employed full-time	1.41	.22	***	4.09	.27	.19		1.30	−1.16	.27	***	.31
Employed part-time	.95	.24	***	2.58	.62	.17	***	1.86	−1.03	.28	***	.36
Looking for work	.17	.23		1.18	−.02	.22		.98	−.27	.37		.76
Household structure
Father in household	.05	.26		1.05	−.06	.14		.94	−.05	.23		.95
Number of children under 18	−.16	.08	*	.86	−.02	.06		.98	−.15	.07	*	.87
Preschool/care preferences
Care when child sick	−.19	.13		.83	−.24	.09	**	.79	−.24	.13		.79
Close to home	.17	.13		1.19	.24	.09	*	1.27	.33	.16	*	1.39
Reasonable cost	−.09	.14		.91	−.38	.14	**	.69	−.13	.18		.88
Flexible hours	−.20	.12		.82	.06	.10		1.06	−.88	.14	***	.42
Human Capital Considerations
Maternal education
Less than high school	−.77	.35	*	.46	−.99	.41	*	.37	−1.07	.44	*	.34
High school graduate	−.48	.31		.62	−.94	.38	*	.39	−.58	.45		.56
Some college	−.26	.34		.77	−.76	.40		.47	−.29	.45		.75
Parent education expectations
Attend some college	.05	.29		1.05	−.21	.24		.81	−.29	.45		.72
Complete a B.A.	.23	.28		1.26	.27	.22		1.31	.33	.29		1.39
Earn M.A. or higher	−.27	.25		.77	.28	.22		1.32	−.07	.30		.93
Preschool/care preferences
Prepare child for kindergarten	.57	.20	**	1.77	−.17	.18		.85	.71	.22	**	2.02
Small number of children	−.22	.15		.80	.11	.11		1.12	−.17	.18		.85
Child Elicitation
Mental skills
Low (1st quartile)	−.05	.21		1.56	.09	.15		1.09	.26	.20		1.30
High (4th quartile)	.10	.26		1.11	.09	.22		1.10	.46	.26		1.59
Motor skills
Low (1st quartile)	.44	.21	*	1.56	.26	.15		1.29	.16	.20		1.18
High (4th quartile)	.19	.22		1.21	−.12	.19		.89	−.01	.21		.99
Child negativity towards parent
Moderate negativity	.70	.28	*	2.02	.14	.22		1.15	.30	.35		1.35
High negativity	.19	.29		1.21	.04	.27		1.04	−.17	.29		.85
Child persistence
Consistently lacks	.05	.45		1.02	−.26	.38		.77	.32	.49		1.38
Typically not	.57	.43		1.77	−.20	.32		.82	.43	.37		1.54
Half the time	.14	.34		1.15	−.35	.27		.70	.59	.37		1.81
Typically is	.10	.33		1.11	−.42	.26		.66	.30	.32		1.35

^*p <.05;

^**p <.01;

^***p <.01

Note: The results presented come from three binary logistic models, rotating the reference category to achieve all comparisons with preschool enrollment as a dependent variable. The Perception of preschool/care options also included a “don’t know” category (not shown). SNAP = Supplemental Nutrition Assistance Program; TANF = Temporary Assistance for Needy Families; CHIP = Child Health Insurance Program

Models were estimated in stages, due to concerns about multicollinearity and overcontrolling. Each model included the covariates, and then we iteratively added and subtracted the indicators for the five mechanisms. The final model presented in Table 2 included all predictors. We focus on the final model—examining coefficients that remained significant net of all other covariates and indicators. We discuss these results in terms of the difference between preschool and the other educational setting (Head Start) and between preschool and the two non-educational settings (parental care, other care). Results from models including each focal factor as the sole predictor and then iteratively adding blocks of factors for each mechanism are available in Appendix Tables 1 and 2.

Beginning with the covariates, the most consistent difference between children from low-income families enrolled in preschool and those in non-educational settings was that the latter children were slightly older. Among all children enrolled in educational settings, White children and those with native-born parents were more likely to be in preschool than in Head Start. For example, the children of mothers born outside of the U.S. were 46% less likely to be in a preschool than in Head Start, net of all of the other factors in the model.

Turning to the two contextual mechanisms, the supply-side factors and systematic connections had only a few observable associations with enrollment. When low-income parents perceived good child care options in their communities (a supply-side factor), they were more likely to have a child enrolled in preschool (over 75% more likely than both parental care and other care). Such enrollment was lower (by 92–94%) when children lived in neighborhoods with higher employment rates. Little evidence suggested that connections to systemic connections promoted preschool enrollment, beyond an inverse association of Medicaid receipt and enrollment in preschool (vs. Head Start) and a positive association of housing help with enrollment in preschool (vs. other care). Several other systemic connections predicted Head Start over preschool enrollment when this block of factors was examined on its own, but these associations fell to non-significance in the final model.

As for parental necessity, current maternal employment was consistently associated with preschool enrollment in this population. Not surprisingly, part-time and full-time employment were both associated with preschool enrollment when compared to parent care. Part-time employment was also associated with increased odds of preschool enrollment when compared to Head Start. Conversely, part- and full-time employment were both associated with decreased odds of preschool, when compared to other non-parental care. In terms of household structure, children from low-income families were less likely to be in preschool than in non-educational settings when their parents had more children. Turning to the aspects of early care and education that low-income parents prioritized, children were 39% more likely to be in preschool (vs. other care) when their parents needed care close to home and 58% less likely when they needed flexible hours. Children were less likely to be in preschool (and more likely to be in Head Start) when their parents needed care for sick children and reasonably priced care and more likely to be in preschool (and less likely to be in Head Start) when their parents needed care close to home.

Human capital considerations also seemed to factor into the preschool enrollment of children from low-income families. Across all three comparisons, mothers with the lowest levels of education (less than a high school degree) were less likely to have their child enrolled in nonHead Start preschool. Parent educational expectations were not associated with enrollment patterns among low-income families in the final model. For the comparison between preschool and the non-educational settings, parents wanting “children to be prepared for kindergarten” was a significant differentiating factor. When parents prioritized this aspect of child care, their children were more likely to be in preschool than parental care (77% higher odds) or other non-parental care (102% higher odds).

The final mechanism was child elicitation. There were only two significant elicitation indicators. First, children who showed low motor skills at age 2 were more likely to be enrolled in preschool (56%) at age 4 than to be solely in parent care. Children who were rated as moderately negative towards their parents were also more likely (102%) to be in preschool than in parental care. In the prior elicitation model but not in the final model, the observed effect of low motor skills also differentiated preschool enrollment from Head Start.

Comparisons with Children from Non-Low-Income Families

The above results yielded insights into how children from low-income families were selected into preschool. What those results did not show was whether these patterns were unique to such families or instead generalizable across the socioeconomic spectrum. To explore this issue, we re-estimated all models (the models for each mechanism and the final model) in the ECLS-B subsample of children from families whose incomes exceeded 185% of the federal poverty line for their household sizes. These results are presented in Table 3. Note that we did not estimate the comparison between non-Head start preschool and Head Start because children from non-low-income families are not eligible for Head Start. We used post-hoc coefficient comparison tests to formally examine group differences and focus on factors that were significantly different across income groups (Clogg, Petkova, & Haritou, 1995; Paternoster et al., 1998). As mentioned previously, we also dropped the systemic connections mechanism.

Table 3.

Predictors of Preschool Enrollment among Children from Non-Low-Income Families

	Likelihood of being in non-Head Start Preschool versus…
	Parent care only				Other types of care
	B	SE	Sig.	OR	B	SE	Sig.	OR
Covariates
Child gender (male)	−.24	.15		.79	−.11	.16		.89
Child age (months)	.06	.02	**	1.06	.10	.02	***	1.11
Mom age at birth (yrs.)	.05	.02	**	1.05	.04	.02	*	1.04
Mom born outside U.S.	.40	.30		1.49	.59	.28	*	1.81
Maternal race/ethnicity
Non-Latino/a Black	.17	.39		1.18	−.03	.27		.97
Latino/a	−.03	.28		.97	−.31	.23		.73
Asian American	.12	.39		1.12	−.15	.34		.86
Native American	.49	.41		1.62	−.22	.57		.80
Multiracial/ethnic	−.64	.48		.53	−.29	.78		.75
Prior child care (age 2)
Center-based care	.38	.29		1.47	.82	.31	**	2.27
Relative care	.34	.26		1.41	−.84	.26	***	.43
Non-relative care	.37	.32		1.45	−.58	.23	*	.56
Child care supply side
Average income in zip code	.02	.01		1.02	.01	.01		1.01
% zip code with B.A. or higher	.20	1.11		1.22	.44	1.26		1.56
% zip code employed	−.75	1.51		.47	−.64	1.67		1.53
% zip code < 18 years old	−.69	1.45		.50	.08	1.15		1.09
Perceived care options as good	.32	.25		1.37	.28	.21		1.32
Parental necessity
Maternal employment
Employed full-time	1.45	.26	***	4.24	−1.40	.40	***	.25
Employed part-time	.75	.25	**	2.13	−.98	.38	**	.38
Looking for work	−1.25	.49	**	.29	−2.01	.56	***	.13
Household structure
Father in household	−.63	.33		.54	.13	.28		1.14
Number of children under 18	−.12	.10		.88	−.18	.10		.83
Preschool/care preferences
Care when child sick	−.37	.11	***	.69	−.19	.12		.83
Close to home	.01	.19		1.01	.17	.15		1.19
Reasonable cost	.15	.19		1.16	.14	.18		1.15
Flexible hours	−.29	.15	*	.75	−.65	.13	***	.52
Human Capital Considerations
Maternal education
Less than high school	−.96	.41	*	.38	−.72	.49		.49
High school graduate	−.65	.27	*	.52	−.54	.30		.59
Some college	−.51	.26	*	.60	−.10	.20		.90
Parent education expectations
Attend some college	.49	.42		1.63	.10	.44		1.10
Complete a B.A.	.63	.38		1.89	.16	.39		1.17
Earn M.A. or higher	.39	.41		1.48	.01	.40		1.01
Preschool/care preferences
Prepare child for kindergarten	.69	.18	***	2.00	1.15	.16	***	3.17
Small number of children	.14	.20		1.15	−.26	.19		.77
Child Elicitation
Mental skills
Low (1st quartile)	.37	.31		1.44	.00	.26		1.00
High (4th quartile)	−.18	.22		.83	−.04	.22		.96
Motor skills
Low (1st quartile)	.18	.24		1.20	.18	.22		1.19
High (4th quartile)	−.08	.21		.92	.35	.24		1.41
Child negativity towards parent
Moderate negativity	−.08	.21		.92	−.55	.27	*	.58
High negativity	−.10	.36		.90	−.48	.36		.62
Child persistence
Consistently lacks	.31	.58		1.36	.53	.50		1.69
Typically not	−.19	.40		.83	.16	.32		1.18
Half the time	−.10	.36		.90	−.06	.27		.94
Typically is	.16	.29		1.18	−.05	.28		.95

^*p <.05;

^**p <.01;

^***p <.01

Note: The results presented come from two binary logistic models, rotating the reference category to achieve both comparisons with preschool enrollment as a dependent variable. SNAP = Supplemental Nutrition Assistance Program; TANF = Temporary Assistance for Needy Families; CHIP = Child Health Insurance Program

Beginning with the contextual mechanisms, perceptions of good local care options and local employment rates were not associated with greater odds of preschool enrollment in this subsample as they were in the low-income subsample. Turning to the other mechanisms, a need for flexible hours reduced the likelihood of preschool enrollment relative to children being in a non-educational setting, a pattern not found in the low-income subsample. The full gradient of maternal employment differentiated children from non-low-income families on their odds of preschool enrollment, compared to only certain employment statuses in the low-income subsample. For example, mothers who reported currently looking for employment were less likely to have their children enrolled in preschool compared to both parental care and other care, a significantly different pattern from the low-income sample. Although having more children did significantly reduce the odds of enrollment in the low-income subsample only, these apparent income differences were not significant. Human capital considerations mattered somewhat differently in the subsample of families that were not low-income. Higher maternal education predicted preschool enrollment, just as for the low-income families, but this advantage was observed across the education gradient (not demarcated at high school graduation) relative to parental care and was not observed in relation to other non-parental care. Like in the low-income subsample, children were more likely to be in preschool (over parental or other care) when their parents prioritized school readiness when selecting care options. Unlike in the low-income sample, the significant association between moderate negativity (a marker of child elicitation) and preschool enrollment was relative to other care rather than parental care and was negative rather than positive. This difference was marginally significant at p < .10.

Discussion

The goal of this study was to examine the potential mechanisms of selection of children into preschool programs. Building on theory (Meyers & Jordan, 2006) and past research (Coley et al., 2014), this study took the new step of examining such mechanisms in the low-income population. The literature shows that preschool programs are advantageous for later achievement for children from low-income families (Duncan & Magnuson, 2013). What is less clear, however, is how and why low-income parents enroll their children in a preschool program.

In general, children from low-income families were more likely to be enrolled in preschool when their parents perceived their neighborhoods as offering good care options for them (supply side), did not have many restrictions (e.g., time, convenience) on what they needed from care options (necessity), and were interested in supporting school readiness (human capital). Children who had moderate problems with parents or had low motor skills were also more likely to be enrolled in preschool, the sole evidence for child elicitation. Preschool enrollment was in part about availability (real or perceived), in part about the need for flexibility (the special circumstances that parents prioritized in care accounted for the unexpectedly weak role of maternal employment in enrollment), and in part about a valuing of educational opportunity (directly aligned with the general mission of preschools). No one mechanism really stood out as a driving force, and the systemic connections and child elicitation mechanisms offered little explanatory power. In line with the accommodations framework, therefore, the “choice” of preschool for low-income children is a reflection of many small factors converging to create supply and demand rather than any single factor or set of factors.

Based on evidence that preschool has educational benefits for children, our initial concern was simply whether children from low-income families were enrolled in preschool or not. This distinction, however, begs the question: compared to what? We concentrated on whether children were enrolled in preschool compared to care settings that were not formally educational (parent care, other types of care such as relative care) and then compared to a public preschool option (Head Start). The results of all possible comparisons revealed that this basic distinction was well-motivated, as the predictors of preschool enrollment over non-educational settings were more similar to each other than they were to the predictors of preschool enrollment over Head Start. An aspect of supply (perceptions of good local options), necessity (having fewer children at home), and human capital considerations (an emphasis on care that would prepare children for school) predicted preschool enrollment over non-educational settings but not over Head Start. Yet, this general pattern did not always hold. For example, mothers who reported any employment were not the sole caregivers of their children, but the children of low-income employed mothers were more likely to be in non-educational, non-parental care than in preschool or Head Start. Overall, the narrower differences in selection between non-Head Start preschool and Head Start were, contrary to expectations, focused more on necessity and human capital considerations. What selected children from low-income families into preschool over Head Start was primarily about whether their parents had weaker employment constraints on the care that they could use (e.g., if they worked part-time), fewer specific child care needs (e.g., they were less concerned with cost), and slightly more educational attainment (e.g., they had graduated from high school). Thus, families who enrolled their children in Head Start appeared to be qualitatively different from other families also identified as low-income. Because the selection criterion for Head Start is largely based on insufficient income (as well as other criteria), these results likely represent the experiences of the most disadvantaged families or those in the most difficult situations. Also important to note is that low-income and Head Start-eligible families are often grouped in the same categories when examining issues of income and child development. This finding would suggest that the two groups differ in their motivations toward preschool selection and should be examined separately when considering the issue of preschool enrollment.

In many ways, these selection patterns were not unique to the low-income population. Many factors operated in similar ways in the sample of families that were not low-income. The biggest distinction was more about degree. Fuller gradients of maternal employment and educational attainment in preschool enrollment appeared among non-low-income families. For example, whether mothers had college experience or not selected children in such families into preschool over parent care, and whether mothers were unemployed but looking for work (rather than unemployed and not looking) selected such children out of preschool compared to parent or other care. Whether such income-related differences in selection into preschool might moderate positive outcomes of preschool enrollment is an important question of future research.

This research suggests a few other issues that need more consideration in the future. First, it reveals the diversity among low-income families. Often treated as a monolithic population, relative gradients of income deprivation among those low in income may produce diverse behaviors. The reasons or factors that contribute to selecting a preschool program depend on the amount of money available to be selective as well as other factors that enable or block selective parents from acting on what they want. For example, for parents with the fewest monetary and associated resources, Head Start may have been the only real option for preschool, whereas parents who were still poor but had other resources on which to draw might have been able to find out about and secure opportunities in other types of programs, including those connected to elementary schools or subsidized in other ways (Fuller, 2007). Second, preschool can be differentiated in many more ways than a dichotomy between Head Start and not Head Start. We could not distinguish among public/private, adult/child centered, academic/play, or other kinds of preschool programs, nor could we include factors related to the quality of such settings. These additional categories may have given us a better idea of the types of choices that parents were making in selecting programs for their children during the critical early childhood period, but what we have done here provides a good foundation on which to build. Third, the generalizability of this study through the use of national data is a strength, but it provided us little purchase on causal inference. Much more needs to be done to determine if selection mechanisms were causal in nature (Duncan & Magnuson, 2013).

With these future directions and associated limitations in mind, we argue that this research elucidated the potential ways in which children from low-income families end up in various preschool programs and thereby qualified current knowledge about observed preschool effects on child outcomes. As such, we have added a unique view on how selection effects may be important for achievement and behavior outcomes of children as they enter K-12 schooling. To the extent that these findings are replicated in other studies, we would be able to design programs that would increase enrollment in preschool. Marketing preschools as effective variants on child care and as potential compensatory strategies for children in need could be effective for the low-income population, as would increasing awareness more generally.