Changes in Parents’ Spanking and Reading as Mechanisms for Head Start Impacts on Children

Abstract

This study examined whether Head Start, the nation’s main two-generation program for low-income families, benefits children in part through positive changes in parents’ use of spanking and reading to children. Data were drawn from the 3-year-old cohort of the national evaluation of the Head Start program known as the Head Start Impact Study (N = 2,063). Results indicated that Head Start had small indirect effects on children’s spelling ability at age 4 and their aggression at age 4 through an increase in parents’ reading to their children. Taken together, the results suggest that parents plays a role in sustaining positive benefits of the Head Start program for children’s behavior and literacy skills, one that could be enhanced with a greater emphasis on parent involvement and education.

For the past 50 years, the federal Head Start program has provided enriched early childhood education to over 31 million children from low-income families, including nearly 1 million in the 2012–2013 school year alone (Office of Head Start, 2014). Head Start was created in 1965 as an initiative in President Johnson’s War on Poverty with the goal of closing the achievement gap between children from low- and high-income families. Although the provision of half- and full-day preschool for 3- and 4-year-old children is its central feature, Head Start was designed as a two-generation program, meaning that it targeted both children and their parents. Building on ideas that co-founder Bronfenbrenner later developed into his ecological systems theory (1979), Head Start was developed under the theory that improving children’s experiences in the two major contexts of their lives, their homes and schools, would be more beneficial than intervening in either one alone (Zigler & Styfco, 2000).

Head Start has thus always had dual goals of directly intervening in the lives of children through the provision of high quality preschool and of indirectly affecting them by changing the behaviors of their parents. Indeed, Head Start regulations call for parents to be included in all aspects of the program and require that programs provide opportunities to enhance parents’ “parenting skills, knowledge, and understanding of the educational and developmental needs and activities of their children” (Office of Head Start, 2009, §1304.40.e.3). Less than half of Head Start centers meet this goal through the provision of direct instruction in the form of parenting classes (48.7%: Office of Head Start, 2014), although some Head Start centers may be co-located in community centers that offer parenting classes.

How, then, might Head Start improve parenting? There are likely to be two main mechanisms. One mechanism is parent engagement in their children’s learning. The Office for Head Start emphasizes that centers should engage parents as equal partners in their children’s learning, in large part by having parents and teachers share with each other what they have observed about a child’s learning and developmental progress (Office of Head Start, 2011). Such family-school connections help parents learn the strategies used by teachers to promote children’s learning and to guide their appropriate behavior, strategies that parents can then try at home. A second mechanism is through parent involvement in the Head Start centers, which is emphasized in current Head Start regulations (Office of Head Start, 2009). Parent involvement in centers can take several forms, including volunteering in classrooms, attending parent-teacher conferences, participating in parent committees, and attending workshops (blinded, 2015). Since its inception, Head Start’s philosophy was that parents would best learn positive parenting skills by observing teachers interacting with children (Allen, Sethi, Smith, & Astuto, 2007). Teachers demonstrate for parents how to be actively involved with their children’s learning; they can also model positive and child-centered means of guiding and controlling children’s behavior.

Despite Head Start’s existence for five decades as a two-generation program, and despite the evidence that Head Start does positively impact parenting (Gebler & Isen, 2013; Puma et al., 2010; Zhai, Waldfogel, & Brooks-Gunn, 2013), few efforts have been made to examine whether part of Head Start’s impact on children is mediated through such positive impact on parents. The current study did so by focusing on two discrete parenting behaviors, namely spanking and reading to children. We summarize below what is known about Head Start’s impacts on these parenting practices and why such impacts might translate into benefits for children.

Head Start and Parents’ Use of Spanking

Spanking remains a common parenting practice across the U.S. and among the target population for Head Start, namely poor and near-poor parents (Berlin, et al., 2009; Bradley, Corwyn, Burchinal, McAdoo, & Coll, 2001). Spanking persists despite a large body of correlational evidence (Gershoff, 2002) and a smaller but consistent body of longitudinal evidence (Coley, Kull, & Carrano, 2014; Ferguson, 2013; Gershoff, Lansford, Sexton, Davis-Kean, & Sameroff, 2012) finding that spanking is linked with increases in children’s problem behavior over time, over and above children’s initial levels of problem behavior. This evidence has led several professional organizations, including the American Academy of Pediatrics (1998), to recommend that parents not spank their children both because it is ineffective and because it puts children at risk for behavioral and emotional problems and physical injury.

The federal Head Start program does not permit spanking in any of its centers. Current Head Start policy requires that all staff and volunteers “will use positive methods of child guidance and will not engage in corporal punishment” (Office for Head Start, 2009, §1304.52.i.1.iv). Parents who volunteer in Head Start, or even those who observe the classroom at drop-off and pick-up, will witness teachers using methods other than spanking to manage children’s behavior. Seeing the successful classroom management style of teachers who use positive discipline or attending parenting classes at Head Start centers may lead parents to implement what they have learned at Head Start when disciplining their children at home.

Several correlational studies have linked Head Start participation with reductions in parents’ use of spanking. For example, data from the Family and Child Experiences Survey (FACES) 2006 Cohort, a longitudinal study of families in Head Start, revealed that parents significantly decreased their use of spanking over the course of the Head Start year (Aikens, et al., 2010). Using data from the Fragile Families and Child Wellbeing study, Zhai and colleagues (2013) documented similar patterns; spanking was less frequent among families in Head Start. In a separate national sample, Magnuson and Waldfogel (2005) found that parents whose children were in Head Start were less likely to spank than parents whose children were not in Head Start.

The link between participation in Head Start and reduced spanking found in these correlational studies has been confirmed through experimental evaluations of Head Start and its companion program Early Head Start. A national evaluation of the Early Head Start program, which differs from Head Start in being targeted to children under age 3 and in taking the form of home visits, centers, or a combination of the two, found that participating parents spanked their children less than control group parents when children were age 2 and age 3 (Vogel, Brooks-Gunn, Martin, & Klute, 2013). Evidence of the impact of Head Start on spanking also comes from the recent experimental evaluation known as the Head Start Impact Study (HSIS), a longitudinal study of nearly 5,000 3- and 4-year-old children who were randomly selected to receive Head Start treatment or to be in a control group. At the end of the Head Start year, parents of 3-year-olds had significantly reduced their use of spanking, an effect that persisted until the children were in kindergarten (Puma et al., 2012). It is important to note, however, that the official HSIS report did not examine whether these changes in parenting were associated with subsequent improvements in children’s behavior. We take on that task in the current study.

Head Start and Parents’ Reading to Children

Families’ home reading to their preschool children has long been considered a necessary condition for children’s emergent literacy (Bowman, Donovan, & Burns, 2001), although the evidence on this score is mixed (Phillips, Norris, & Anderson, 2008). There are many reasons to think that parents’ reading to children would benefit them, including that it exposes children to new words and novel syntax and thereby enhances their literacy skills, and that it constitutes a positive time that parents and children spend together which may enhance both reading skills (Sonnenschein & Munsterman, 2002) and the quality of the parent-child relationship.

Children from disadvantaged families are at a distinct disadvantage when it comes to language and literacy development. By the time they reach 2 years of age, children whose families have low socioeconomic status (SES) are 6 months behind their higher SES peers in language development (Fernald, Marchman, & Weisleder, 2013). A key reason for these socio-economic disparities appears to be the differences in how often parents engage in literacy-support activities. Low SES parents are less likely to talk with their children and to use a wide vocabulary (Hart & Risley, 2003). Poor parents are also substantially less likely to read to their children than are non-poor parents, regardless of race or ethnic group (Bradley et al., 2001).

As a program designed to close the achievement gap between poor and non-poor children, Head Start emphasizes language and literacy development through organized curricula and the availability of books and reading materials in the preschool environment, and in doing so, may increase the frequency with which parents read to their children at home. Indeed, a coordinated approach that promotes reading in the preschool and home environments may be necessary to boost children’s language and literacy skills; in an evaluation of an intensive literacy intervention implemented in Head Start classrooms, impacts on children’s language abilities were only realized if parents were actively involved in an at-home reading component of the intervention (Whitehurst, et al., 1994).

There is some evidence that Head Start may in fact increase the likelihood that parents read with their children. Children enrolled in Head Start have been found to have access to more cognitively stimulating materials at home, such as books and puzzles, than children not in Head Start (Zhai et al., 2013). In FACES 2006, Latino and non-English speaking parents increased the frequency of reading to their children over the course of the school year, although no change was seen in the sample as a whole (Aikens et al., 2010). In the experimental HSIS, parents in the Head Start condition also reported reading both longer and more often with their children compared with parents in the control condition (Gelber & Isen, 2013). These findings hint that an important way that Head Start may promote children’s language development is by increasing parents’ reading to their children. Although this possibility has been confirmed with the Early Head Start program (Raikes et al., 2014), whether increases in parents’ reading to their children mediate the impact of Head Start on children’s development has yet to be empirically tested.

The Current Study

Our aim was to determine whether the federal Head Start program has indirect effects on children’s behavior and literacy skills that are mediated through its direct effects on parenting behaviors. We analyzed data from the HSIS, a nationally representative experimental evaluation of Head Start mandated by Congress in its 1998 reauthorization of the program. Three- and 4-year-old children who had never attended Head Start were randomly assigned to a Head Start treatment group or to a control group and then were followed until the children reached third grade. For the present study, we focused solely on children in the 3-year-old cohort in order to link Head Start-induced changes in parenting with changes in children’s behavior and literacy skills before the start of formal schooling. We examined impacts on three aspects of child behavior (aggression, inattention, and social skills) and three aspects of children’s literacy skills (receptive vocabulary, letter-word identification, and spelling skills) using three waves of data.

We hypothesized that Head Start would have indirect effects on children’s behavior through a reduction in parents’ use of spanking, and that Head Start would have indirect effects on children’s literacy skills through an increase in parents’ reading to children. We also speculated, however, that there could be cross-domain associations. For example, a reduction in spanking could lead to an improvement in children’s achievement, consistent with recent findings that spanking is associated with lower achievement (Ferguson, 2013). Similarly, increases in the amount of time parents spend reading to their children may lead to a decrease in children’s problem behaviors, perhaps by improving the quality of the parent-child relationship. Given that such cross-domain effects of changes in parenting have rarely been examined, we could only speculatively hypothesize that there may be some cross-domain associations of Head Start-induced changes in parenting with children’s outcomes.

Method

Head Start Impact Study

Data for this project were derived from the HSIS, a nationally representative randomized control trial commissioned by the Department of Health and Human Services to examine the longitudinal impact of one year of Head Start enrollment on children’s early academic, socio-emotional, and behavioral outcomes. Two cohorts of children, one that entered the study at age 4 and a second that entered the study at age 3, were included in the HSIS. Data collection began in the fall of 2002. Four-year-olds were assessed across 5 waves of data (fall and spring of Head Start year, spring of kindergarten, spring of 1^st grade, and spring of 3^rd grade); 3-year-olds were assessed at an additional wave at the spring of their 4-year-old preschool year, for a total of six waves. Data were collected using parent interviews, direct child assessments, center-director and teacher interviews, and observations of classrooms. As noted above, we only included children from the 3-year-old cohort. Data for our analyses were drawn from the first three assessments, which we will refer to by waves (Wave 1 = fall of Head Start 3-year-old year; Wave 2 = spring of Head Start 3-year-old year; Wave 3 = spring of 4-year-old year).

Sampling Design

Sample selection for the HSIS was completed in multiple stages. Head Start programs which had centers with waiting lists were clustered with at least 7 other programs according to location. These program clusters were then stratified based on racial/ethnic composition, geographic location, urbanicity, and availability of state-level programs aimed at assisting low-income families. One program cluster was selected from each strata using proportional probability adjustments and then three programs from each cluster were randomly sampled. Next, three centers from each program were sampled, resulting in the selection of 383 centers. Lastly, children from each selected center were randomly assigned to the treatment or control group. Children and families assigned to the treatment condition were eligible for Head Start services, whereas those in the control group were not. Random assignment at the child level was performed separately for the 3- and 4-year-old cohorts, with an intentional over sampling of 3-year-olds. Approximately 2/3 of children were assigned to the Head Start treatment condition.

Sample Characteristics

In total, the HSIS sample was comprised of 4,442 children, 2,449 of whom were in the 3-year-old cohort. Our sample consisted of 2,063 children and their families (62% treatment) from the 3-year-old cohort who had a valid longitudinal child-level weight (LONGCHTRWT3). Descriptive statistics on all demographic characteristics are available in Table 1. Approximately half of the children were female. Seven in ten children were from minority households and 16% were children of immigrants. Roughly half of all children were from single-parent households or had unemployed mothers, and over a third of children had mothers with less than a high school education. Children in the control group received different types of child care, with 38% enrolled in a formal early education program, 22% in less formal child care arrangements (e.g., relative care), and 40% who were cared for by their parents. Balance in demographic characteristics across the treatment and control conditions was largely achieved (see Table 1). Not surprisingly, children in the treatment condition received more hours of formal child care than their control group peers, because Wave 1 occurred after the treatment year had already begun. Mothers of children enrolled in Head Start were also slightly older.

Table 1.

Weighted Child and Family Descriptive Data by Treatment Status

	Mean (SD) or proportion				Tests for significant differences
	Head Start treatment group		Control group
Child characteristics
Age in weeks (Wave 2)	214.48	18.23	214.60	15.95
Female	0.51		0.52
Race/ethnicity
Black	0.35		0.34
Hispanic	0.34		0.33
White/Other	0.31		0.33
Dual language learner	0.22		0.24
Special needs	0.13		0.11
Household characteristics (Wave 1)
Caregiver age	29.59	8.77	28.56	5.80	*
Teenage mother	0.13		0.16
Bio-parents live together	0.50		0.52
Parent marital status
Married	0.45		0.48
Divorced	0.13		0.13
Single	0.42		0.39
Maternal education
Less than high school	0.33		0.34
High school/GED	0.36		0.33
Beyond high school	0.32		0.32
Maternal employment
Unemployed	0.50		0.47
Part-time	0.17		0.19
Full-time	0.34		0.33
Home language (English)	0.75		0.71
Immigrant mother	0.16		0.15
Number of children	2.62	1.48	2.65	1.10
Urban area	0.80		0.80
Formal childcare hours	25.08	13.60	19.03	15.35	***

Note.

^***p < .001.

^*p < .05.

Measures

Weighted descriptives and reliability values for all constructs can be found in Table 2. Reliability values were obtained from the HSIS Technical Report (Puma et al., 2010).

Table 2.

Weighted Descriptive Statistics of the Focal Parent and Child Variables

	Mean (SD)				Tests for significant differences	Alpha
	Head Start treatment group		Control group
Parenting practices
Use of spanking (possible range: 0–3)
Wave 1	0.86	1.32	1.06	1.17	*	-
Wave 2	0.70	1.07	0.85	0.90	*	-
Reading to child (possible range: 1–4)
Wave 1	2.91	1.04	2.81	0.85		-
Wave 2	2.96	1.04	2.73	0.80	***	-
Children’s behavior
Aggression (possible range: 0–8)
Wave 1	3.12	1.89	3.02	1.54		.87
Wave 2	2.96	1.91	3.03	1.53		.61
Wave 3	2.61	1.96	2.76	1.52		.65
Inattention (possible range: 0–6)
Wave 1	1.83	1.70	1.91	1.32		.84
Wave 2	1.67	1.65	1.99	1.39	**	.62
Wave 3	1.61	1.65	1.78	1.30		.58
Social skills (possible range: 0–14)
Wave 1	12.23	1.99	12.05	1.58		.85
Wave 2	12.43	1.92	12.34	1.48		.62
Wave 3	12.64	1.65	12.49	1.53		.61
Children’s literacy skills
Receptive vocabulary (possible range: 128–408)
Wave 1	230.05	40.96	229.81	33.26		.61
Wave 2	257.67	37.31	250.85	32.24	**	.62
Wave 3	300.84	40.30	297.88	35.65		.70
Letter-word identification (possible range: 264–408)
Wave 1	294.54	25.79	293.43	18.49		.82
Wave 2	306.89	28.50	300.08	20.12	***	.87
Wave 3	332.95	31.10	329.86	25.10		.91
Spelling (possible range: 277–420)
Wave 1	334.43	24.93	334.23	23.09		.70
Wave 2	346.65	24.48	344.20	19.86		.73
Wave 3	376.85	29.01	376.91	22.35		.81

Note.

^***p < .001.

^**p <.01.

^*p < .05.

Parents’ Use of Spanking

At Waves 1 and 2, parents were asked whether they had spanked their children in the previous week, and, if so, approximately how many times. Responses ranged from 0 to 21 times in the past week. This item was skewed at both waves (means = .98 and .88, SDs = 1.52 and 1.50, skew = 4.51 and 3.07, Waves 1 and 2, respectively), with roughly 97% of parents reporting 3 or fewer spankings per week. Accordingly, we top-coded the variable at 3, which resulted in a scale from 0 to 2. This transformation greatly reduced the skew at both waves (means = .91 and .77, SDs = 1.23 and 1.00, skew = .97 and .84).

Parents’ Reading to Children

At Waves 1 and 2, parents reported on the number of times they had read to their children during the previous week. Responses ranged from 1 (“not at all”) to 4 (“every day”).

Children’s Behavior

At each wave, parents reported on their children’s behavior (1 = not true, 2 = sometimes true, 3 = very true) using items from the Child Behavior Checklist (CBCL; Achenbach, 1991). Parents reported on their children’s aggression (e.g., has temper tantrums; hits and fights with other children) and inattention (e.g., is very restless and fidgets a lot; can’t concentrate or pay attention for long). In addition to these CBCL scales, parents reported on children’s social skills and positive approaches toward learning using a measure developed for the Head Start FACES (Zill et al., 2006) that drew from the Personal Maturity scale (Entwistle, Alexander, Cadigan, & Pallis, 1987) and the Child Behavior Checklist for Preschool-Aged Children (Achenbach, Edelbrock, & Howell, 1987). Sample items were: makes friends easily, likes to try new things. For brevity, we refer to this outcome as social skills. Teacher reports of child behavior could not be used because teachers were not surveyed at Wave 1 and only 64% of children had teacher-reported data at Wave 2.

Children’s Literacy Skills

Children’s English receptive vocabulary skills were directly assessed at Waves 1, 2, and 3 using the Peabody Picture Vocabulary Test (PPVT; Dunn & Dunn, 1997). The PPVT is a widely used measure that evaluates children’s knowledge of words by asking children to match pictures to orally presented words. Children’s letter-word identification and spelling skills were assessed with subsets of the Woodcock Johnson (WJ-III; Woodcock, McGrew, & Mather, 2001). The letter-word identification scale asks children to identify letters and words in a book, whereas the spelling skills scale asks children to copy and write letters.

Covariates

To reduce the possibility of spurious associations, all focal analyses included a comprehensive set of family and child covariates, which, unless otherwise noted, were drawn from Wave 1 parent interviews. Six child covariates were included in all models: age upon spring assessment, gender, race/ethnicity, dual-language learner status, and disability status. Household covariates included in the models were: mothers’ age, education, employment status, immigrant status, marital status, whether children were born to teenage mothers, whether both biological parents lived with the child at birth, the number of children under 18 in the household, household language, and the hours per week children were enrolled in a formal child care arrangement. To control for variation in the length between assessments, all models also incorporated the number of weeks elapsed from September 1^st, 2002, until the Wave 1 parent and child assessments, the number of weeks between the Wave 1 and Wave 2 assessments, and the number of weeks between Wave 2 and Wave 3 assessments. As a sensitivity check, we ran a final set of models that included the type of child care the children experienced in their 4-year-old year (57% Head Start; 33% center-based care; 7% parental care; 3% informal care).

Analytic Strategy

Missing Data, Weights, and Clustering

To ensure that all cases were retained, missing data were addressed through full-information maximum likelihood estimation (FIML), a preferred method over listwise/pairwise deletion (Schafer & Graham, 2002). All models included longitudinal sampling weights, which corrected for non-random attrition across time and ensured that our sample was generalizable to the population Head Start children, and clustering by Head Start program to adjust standard errors as a result of the clustered sampling frame.

At Wave 1, children who were identified as dual-language learners (23%) were only assessed using the PPVT and the Letter-Word subscale of the Woodcock Johnson; they were not administered the Spelling subscale in fall 2002 and, thus, were missing this variable in the analyses. Their scores were estimated in the models using FIML. All participants, regardless of language status, received the full battery of assessment tests in Waves 2 and 3. All models included an indicator for language status to account for this non-random pattern of missing. We also ran two checks of the models for the spelling outcome, once dropping the dual-language learners all together and a second time adding fall PPVT as a control predicting Wave 1 and Wave 2 spelling; the results were the same as the models reported below.

Treatment Variable

Treatment was analyzed as an intent-to-treat (ITT) model, with all children retained in their original randomly assigned groups. As has been described in the HSIS reports (Puma et al., 2012), 14.9% of the 3-year-olds assigned to the Head Start group were no-shows (i.e., they did not attend a Head Start center), whereas 17.3% of 3-year-olds assigned to the control group were cross-overs (i.e., they attended a non-study Head Start center). Despite this noncompliance, we used ITT models in order to preserve random assignment and to be comparable with the official HSIS reports (e.g., Puma et al., 2012).

Cross-Lagged Models

To address our primary research questions, we employed a series of cross-lagged path models within a structural equation modeling (SEM) framework using Mplus version 7 (Muthén & Muthén, 1998–2013). Some key benefits of cross-lagged models are that they allow us to: (a) control for parents’ and children’s initial behaviors before the Head Start treatment; (b) model the stability of both parent and child behavior through autoregressive paths so that downstream outcomes represent change variables; (c) include both parent- and child-effects on each other’s behavior over time; and (d) estimate indirect effects of Head Start through mediational pathways. Separate models were run for each child outcome (aggression, inattention, social skills, receptive vocabulary, letter-word identification, and spelling) with both aspects of parenting behavior (spanking, reading) included simultaneously in each. Constructs measured at the same wave were co-varied in all models. Mediation was tested through the INDIRECT command.

Our hypothesized model for this study using these three waves of data is presented in Figure 1 and includes several components, including autoregressive effects and cross-lagged paths, designed to increase our confidence that we have ruled out alternative explanations. The autoregressive effects (a paths) account for any small mean differences in parent or child behaviors at entry to Head Start that existed despite randomization. Thus, paths a1 and a4 represent changes in parenting behavior over the course of the Head Start year, whereas path a2 represents change in child behavior or achievement over the course of that same year and path a3 represents change in child behavior of achievement from the Wave 2 to Wave 3.

Figure 1.

Hypothesized model for Head Start’s direct and parenting-mediated impacts on children’s behavior and literacy skills.

Note. a paths = autoregressive paths; b paths = cross-lagged effects; c paths = direct treatment effects; c paths = hypothesized paths of indirect treatment impacts.

The cross-lagged effects (b paths) account for the mutual influences parents and children have on each other over the course of the preschool year. Given the two-generation nature of Head Start, we felt it was important to account for the transactional nature of the parent-child relationship, as changes in one partner may be conditional on the initial behavior of the other (Sameroff, 2009). Our model thus includes parent-effect cross-lagged paths from parenting in fall of year 1 to child behavior and achievement in spring of year 1 (paths b1 and b4) as well as child-effect cross-lagged paths from child behavior and achievement in fall of year 1 to parenting in spring of year 1 (paths b2 and b3). We also wanted to be certain to isolate the parent-mediated associations of Head Start from its direct effects on children. Thus, we included direct paths from Head Start treatment to child behavior and achievement at the spring of year 1 (path c1) and the spring of year 2 (path c2). The inclusion of these paths meant that any parent-mediated effects of Head Start was over and above the direct impact the program had on children.

The final paths in our hypothesized model represented indirect pathways that are comprised of two sets of direct effects. The path from Head Start treatment to spanking at spring of year 1 (path d1) and the path from spanking at spring of year 1 to child behavior and achievement in spring of year 2 (path d2) created the indirect pathway for a spanking-mediated effect of Head Start. The same was true for the paths from Head Start treatment to parents’ reading to their child at spring year 1 (path d3) and from reading to the child at spring year 1 to the child’s behavior or achievement at spring of year 2 (path d4). Separate models were run for each of the six child outcomes in order to obtain unique estimates of the indirect effects.

Results

Impacts of Head Start on parenting behavior

We first examined direct impacts of Head Start on change in spanking and reading to children from the beginning (Wave 1) to the end (Wave 2) of the experimental year by including initial levels of the respective parenting behavior as a predictor. Head Start Participation was associated with a significant reduction in how often parents spanked their children (β = −.10, p < .05; see Table 3), over and above the strong stability in spanking between Waves 1 and 2 (β = .50, p < .001). Similarly, parents in the Head Start treatment group reported a significantly larger increase in how often they read to their children than did control group parents (β = .18, p < .01), even though parents were fairly stable in how often they read to their children during the Head Start year (β = .41, p < .001).

Table 3.

Unstandardized and Standardized Path Coefficients for Head Start Impacts on Parenting Practices and Child Outcomes.

	B (SE)	β	R2	Fit indices
Parenting practices
Treatment → spanking W2	−0.10 (0.05) *	−.10	.30	CFI = .970, RMSEA = .041, SRMR = .009, x2 (df = 7) = 30.79, p < .001
Treatment → reading to child W2	0.17 (0.06) **	.18	.22
Children’s behavior
Treatment → aggression W2	−0.11 (0.08)	−.06	.28	CFI = .992, RMSEA = .023, SRMR = .003, x2 (df = 5) = 10.25, p < .001
Treatment → aggression W3	−0.19 (0.09) *	−.11	.20
Treatment → inattention W2	−0.27 (0.08) ***	−.18	.27	CFI = .998, RMSEA = .010, SRMR = .002, x2 (df = 5) = 6.12, p < .001
Treatment → inattention W3	−0.19 (0.09) *	−.13	.18
Treatment → social skills W2	0.06 (0.10)	.03	.16	CFI = .990, RMSEA = .019, SRMR = .004, x2 (df = 5) = 8.56, p < .001
Treatment → social skills W3	0.10 (0.10)	.06	.13
Children’s literacy skills
Treatment → receptive vocab. W2	6.18 (1.74) ***	.17	.51	CFI = .998, RMSEA = .019, SRMR = .003, x2 (df = 5) = 8.86, p < .001
Treatment → receptive vocab. W3	1.13 (1.74)	.03	.51
Treatment → letter word W2	5.12 (1.41) ***	.21	.33	CFI = 1.00, RMSEA = .000, SRMR = .002, x2 (df = 5) = 3.02, p < .001
Treatment → letter word W3	1.63 (1.35)	.06	.31
Treatment → spelling W2	1.69 (1.22)	.08	.27	CFI = .999, RMSEA = .010, SRMR = .003, x2 (df = 5) = 6.08, p < .001
Treatment → spelling W3	−1.60 (1.39)	−.06	.26

Note. All models included autoregressive paths between wave 1 and wave 2 for the parenting behavior and between both waves 1 and 2 and between waves 2 and 3 for the child outcomes. W = wave of assessment. vocab. = vocabulary.

^***p < .001.

^**p < .01.

^*p < .05.

Impacts of Head Start on child outcomes

We next examined direct impacts of Head Start treatment on children’s outcomes at the end of the experimental year (Wave 2) and a year later (Wave 3). After controlling for a full set of family and child covariates, including children’s skills and behavior in the fall of the first year, we found that although there were no impacts of Head Start during the initial year on children’s aggression, Head Start children were rated as having reduced aggressive behavior at the end of their 4-year-old year (β = −.11; p < .05). Children in the Head Start group also demonstrated reduced inattention at the end of the Head Start year (β = −.18, p < .001), with favorable impacts persisting through the subsequent year (β = −.13, p < .05). No significant effects emerged for children’s social skills. Turning to the impact of Head Start on children’s literacy skills, there were positive impacts of Head Start on children’s receptive vocabulary (β = .18, p < .001), and letter-word identification skills (β = .21, p < .001), over the course of the Head Start year. Neither of these impacts persisted into the spring of the school year after Head Start. There also was no impact on children’s spelling skills.

Head Start was thus found to have direct impacts on change into the second year for two (aggression and inattention) of the six child outcomes examined. Despite the lack of significant main effects across all outcomes, we proceeded to test for indirect effects because it was possible that a small direct effect of Head Start on parenting may still contribute to a significant indirect effect on children, and because there is growing agreement that significant direct impacts are not a prerequisite for testing for indirect effects through mediators (MacKinnon, Krull, Lockwood, 2000; Shrout & Bolger, 2002), in part because detection of both direct and indirect effects can be difficult in models that include autoregressive paths (Shrout & Bolger, 2002), as ours do.

Parenting behaviors as mediators of Head Start effects on children’s behaviors

We next tested models with indirect effects through both spanking and reading to children simultaneously. Because each model includes the autoregressive paths for the child outcome across the three waves of data, the outcomes in this table should be interpreted as changes in the focal constructs. Table 4 presents the results from these six models, which include the two parenting-to-child outcome paths and the two indirect Head Start-to-parenting-to-child outcome paths. Improvement in social skills from Wave 1 to Wave 3 was not predicted by parents’ spanking or reading or by an indirect effect of Head Start mediated through parenting. There was similarly no evidence of an indirect effect of Head Start on change in inattention over this same period as mediated through parenting, although there was a direct link between parents’ reading to children at Wave 2 and reduced inattention at Wave 3 (β = −.06, p < .05).

Table 4.

Unstandardized and Standardized Direct and Indirect Path Coefficients for Parenting Behaviors as Mediators of Head Start Impacts on Children from Separate Models by Outcome.

	Aggression (R2 = .26)		Inattention (R2 = .26)		Social skills (R2 = .16)
	B (SE)	β	B (SE)	β	B (SE)	β
Spanking as the mediator
Spanking W2 → behavior W3	0.14 (0.04)**	.08	0.05 (0.04)	.03	−0.08 (0.05)	−.05
Treatment → spanking W2 → behavior W3	−0.02 (0.01)	−.01	−0.00 (0.01)	−.00	0.01 (0.01)	.00
Reading to child as the mediator
Reading to child W2 → behavior W3	−0.18 (0.05)***	−.10	−0.09 (0.04) *	−.06	0.05 (0.06)	.03
Treatment → reading to child W2 → behavior W3	−0.03 (0.01)*	−.02	−0.01 (0.01)	−.01	0.01 (0.01)	.00

	Receptive vocabulary (R2 = .60)		Letter word identification (R2 = .41)		Spelling skills (R2 = .36)
	B (SE)	β	B (SE)	β	B (SE)	β
Spanking as the mediator
Spanking W2 → literacy skills W3	−1.70 (0.76)*	−.04	0.50 (0.65)	.02	−0.37 (0.58)	−.01
Treatment → spanking W2 → literacy skills W3	0.15 (0.11)	.00	−0.04 (0.06)	−.00	0.03 (0.05)	.00
Reading to child as the mediator
Reading to child W2 →literacy skills W3	1.12 (0.93)	.03	0.53 (0.76)	.02	1.72 (0.75)*	.06
Treatment → reading to child W2 → literacy skills W3	0.18 (0.16)	.00	0.09 (0.13)	.00	0.29 (0.13)*	.01

Note. All models included autoregressive paths between wave 1 and wave 2 for the parenting behavior and between waves 1 and 2 and between waves 2 and 3 for the child outcomes. W = wave of assessment. All models fit the data well: CFI >.94, RMSEA < .05, SRMR < .05.

^***p < .001.

^**p < .01.

^*p < .05.

A stronger and more consistent pattern of results was found for change in children’s aggression (Table 4). Parents’ spanking at Wave 2 predicted increases in aggression from Wave 2 to Wave 3 (β = .08, p < .01), while parents’ reading predicted decreases in aggression over this same period (β = −.10, p < .001). However, only the indirect pathway from Head Start to reading to decreased aggression was significant (β_indirect = −.02, p < .05); the indirect effect of Head Start on decreased aggression through a reduction in spanking was not significant (β_indirect = −.01, p < .10). Figure 2 displays the coefficients for all of the focal variables in this model.

Figure 2.

Observed direct and indirect impact of Head Start on children’s externalizing aggression.

Note. Although not illustrated, within time measures were co-varied. Hypothesized paths of indirect treatment impact are bolded. All values represent standardized coefficients. Magnitudes of indirect effects are presented in Table 4. ***p < .001. **p < .01. *p < .05.

Parenting behaviors as mediators of Head Start effects on children’s literacy skills

We then tested the full models for the three literacy outcomes. Although an increase in spanking from Wave 1 to Wave 2 was linked with a decrease in receptive vocabulary from Wave 2 to Wave 3 (β = −.04, p < .05), there was no link between reading to children and vocabulary, and no evidence of an indirect Head Start effect for either parenting variable (see Table 4). Children’s improvement in letter-word identification skills from Wave 2 to Wave 3 was also not predicted by either parenting variable, and thus no indirect effects were found. An indirect effect of Head Start was found, however, for children’s spelling skills. The direct effect of an increase in parents’ reading to their children from Wave 1 to Wave 2 on improvement in children’s spelling skills between Wave 2 and Wave 3 (β = .06, p < .05), was in part the result of Head Start; Head Start was found to have a significant indirect effect on improvement in children’s spelling skills through an increase in how often parents read to their children (β_indirect = .01, p < .05).

Discussion

This study examined whether Head Start promotes children’s positive behavior and literacy skills in part by reducing parents’ use of spanking and by increasing the frequency with which they read to their children. The use of experimental data gave us confidence that any changes we observed in parenting over the course of the preschool year were attributable to Head Start and not to preexisting differences between the families. Cross-lagged models captured the complexity of pathways among the variables of interest and quantified the indirect effect of Head Start on children through parenting. The results from these analyses make clear that Head Start does improve parenting, but only in a few cases is that impact strong enough to lead to improvements in children’s behavior and literacy skills.

Our results confirmed the direct impacts of Head Start on parenting and child behavior that were reported in the final report of HSIS results from the Administration for Children and Families (ACF; Puma et al., 2012), even though our models were more conservative in that they included cross-lagged and autoregressive paths and thus focused on change, not level, in each key construct. Parents in Head Start reported a greater reduction in spanking and greater increases in reading to their children than did control group parents. Three-year-old children in Head Start were reported to experience reductions in attention problems and increases in vocabulary and letter-word recognition. There were two discrepancies between our results and the ACF report, namely that we did not replicate the impact on social skills at age 4 and that we did find that Head Start children in the 3-year-old cohort were rated as exhibiting lower levels of aggression at Wave 3 compared with children in the control group. The differences between our results and those of the ACF report are likely because we controlled for initial levels of child behavior in all models, which meant that our models accounted for initial differences between the treatment and control groups and focused on change rather than level of these outcomes.

Our analyses went beyond the ACF findings, however, by testing whether the observed impacts on parenting may be mechanisms for indirect effects of Head Start on child outcomes. We found this to be the case for two out of six examined pathways. There was an indirect effect of Head Start on a reduction in children’s aggressive behavior and on an increase in their spelling abilities through parents’ reading to their children. It is important to note that these indirect effects were over and above the direct effects of Head Start on change in children’s behaviors and literacy skills over time. We can conclude that Head Start promotes parents’ reading to their children, and that this change in turn is associated with small but significant changes in children’s aggression and spelling skills. Thus, while Head Start is often thought of as an enriched early childhood education program, it is clear that part of the impact depends on how well it lives up to its original design as a two-generation program (Zigler & Styfco, 2000).

Head Start was also associated with reductions in parents’ use of spanking, and this reduction in spanking in turn predicted a reduction in children’s aggression over time in our models. However, the indirect effect of Head Start on a reduction in aggression mediated through reduced spanking was not significant. Thus, while parents who reduced their spanking for any reason did see a reduction in child aggression over time, Head Start’s impact on spanking was not strong enough to fully account for this change. This study found that while Head Start does affect some positive change in parenting, that change is not strong enough to create significant improvements in children’s behavior and literacy skills over time.

In coming to these conclusions, we addressed a basic research question as well: are changes in two key components of parenting behavior, namely the extent to which parents spank and the frequency with which they read to their children linked to changes in children’s later behavior? There has been considerable debate about the causal link between spanking and child outcomes, with some arguing for a parent-to-child link (American Academy of Pediatrics, 1998; Gershoff, 2002) and others arguing for a third-variable or child-effect explanation (Larzelere, Kuhn, & Johnson, 2004). By regressing change on change, our results support the conclusion that spanking predicts children’s problem behaviors. Encouragingly, they also make clear that spanking is an important target for intervention, with reductions in spanking predicting reductions in children’s problem behavior.

Similarly, there is a running debate in the literature regarding the role that reading to children plays in their emergent literacy, with those arguing that it is crucial and perhaps even necessary (Bowman et al., 2001) being countered by those who more cautiously observe that the evidence is not as clear cut (Phillips et al., 2008). Our results provide evidence that an intervention that increases the frequency with which parents’ read to children is in turn associated with gains in children’s spelling skills. It is not entirely clear why the benefit of reading to children was seen for children’s spelling skills but not their receptive vocabulary or their letter-word identification skills. In looking more closely at the Spelling scale of the Woodcock-Johnson battery, the items tap into children’s abilities to trace and copy letters. It may be that 3-year-old children who read with their parents gain familiarity with the shapes of individual letters and can copy them, but cannot yet recognize and read letters or full words (as assessed by the Letter-Word Identification scale) or understand the meanings of the words (per the Receptive Vocabulary scale). Thus, reading to these young children is introducing them to letters and their function, a building block for later literacy skills.

Limitations

The experimental nature of the HSIS made it an ideal data set with which to determine whether Head Start has causal impacts on parenting behaviors. However, because the parenting mediators were not themselves experimentally manipulated, we cannot conclude with certainty that the treatment-induced changes in parenting caused the observed changes in child behavior and achievement (Shrout & Bolger, 2002). Indeed, our mediators can never be truly randomized, as it is impossible to randomly assign children to parents or to randomly assign some parents to spank or not, or to read or not. Parenting behavior is determined by a variety of factors, including parents’ age and education level to name but a few; and these factors may also predict the child outcomes of interest (Duncan, Magnuson, & Ludwig, 2004). Thus, any attempt to show that parenting predicts child outcomes must rule out such potential “third variable” explanations. Although our inclusion of key covariates and our focus on change make significant steps in this direction, it is still possible that some other unmeasured sources of bias are unaccounted for.

The treatment direct and indirect impacts on the child outcomes observed in this study were small in magnitude, βs = |.01 to .21|, even when they were statistically significant. These small effects are the result of our decision to focus on change in parenting mediators and child outcomes rather than just on the level of these variables at subsequent waves. Given the strong stability in parenting and child behaviors over time noted above, there is very little change for the Head Start treatment to predict. Indirect effects are often small and difficult to detect; indeed, our parenting indirect effects are of the same magnitude as found in the Early Head Start study (Jones Harden et al., 2012). The fact that Head Start did predict changes in both parenting mediators and in several of the child outcomes even when little change had occurred points to the importance of Head Start as a potential intervention.

Our study had a few limitations related to measurement. Because the HSIS did not have teachers rate child behavior at Wave 1 and there were no observer ratings of child behavior, we were forced to use the parent ratings of children’s behavior in our analyses. Because parents also reported on their parenting behaviors, the analyses involving parent ratings of child behavior suffer from shared measurement error and, therefore, the observed associations may be inflated. The analyses involving children’s literacy skills used direct assessments and so are free from shared measurement error. In addition, the internal reliability of some measures, particularly of children’s behavior, deteriorated over time (see Table 2), from alphas of .84 to .87 at Wave 1 to alphas of .58 to .65 at Wave 3. It may be that these measures are developmentally appropriate for 3 year olds but less so for 4 and 5 year olds. The lower reliability of the child behavior measures at later waves means that the observed associations among Head Start, the mediators, and the behavior outcomes in this study are likely to be conservative.

Although our models provide some evidence that part of Head Start’s impact on children appears to be indirect through its impact on parenting behavior, we were unable to identify which aspects of Head Start promoted parenting behavior change. Head Start’s emphasis on parent involvement is likely one key process for such change, and indeed we have found in non-experimental data that levels of parent involvement in Head Start centers predict positive changes in their parenting behavior which in turn predict improvements in child behavior (blinded, 2015). Such a model is difficult to do with the HSIS because over 60% of parents in the control group had not enrolled their children in formal early care and education programs; thus, these parents did not have data on school involvement. Additionally, our conclusions about Head Start impacts on parenting are restricted to families who enroll when children are 3 and cannot generalize to children who enroll at age 4.

Implications for Policy and Practice

The three significant indirect effects for Head Start in this study all had very small effect sizes, with standardized coefficients of |.01–.02|. These small effects would seem to have little practical significance, except that the overall impacts of Head Start on children’s behavior and achievement were also small, with standardized coefficients of |.21| or less. Thus, parents’ reading to their child accounted for roughly 18% of the overall impact of Head Start on children’s aggression and on their spelling skills. Similarly small effect sizes have been found in a meta-analysis of the links between preschool classroom quality and children’s language development and problem behaviors (Keys et al., 2013). The fact that indirect effects through parenting are of a similar magnitude to direct associations with the classroom environment indicates that changes in parenting can play an important role in both sustaining the effects of Head Start on children. Yet the small effects also make clear that Head Start is failing to affect large changes in parenting behavior that can benefit children. This may be in part because the Head Start program is vague in how centers should try to impact parenting. Although Head Start regulations require that programs “provide opportunities for parents to enhance their parenting skills” (Office for Head Start, 2009, §1304.40.e.3), the regulations do not specify what form those opportunities should take. One method may be direct instruction in parenting, which has been found to be effective across a range of parenting behaviors (Kaminski, Valle, Filene, & Boyle, 2008; Sandler, Schoenfelder, Wolchik, & MacKinnon, 2011). Parents can effectively be taught to spank their children less often (e.g., Scholer, Hudnut-Beumler, & Dietrich, 2010) and to read to them more often (e.g., Whitehurst et al., 1994). Adding parenting classes or doing more to foster parent involvement in classrooms may be relatively inexpensive ways to live up to Head Start’s initial two-generation mission.

Conclusion

The modest evidence for parent-mediated effects of Head Start on children provided in the current study suggest that affecting permanent changes in parenting behavior appears to be one key way that Head Start can benefit children and one that should receive more emphasis in order to increase program impact. Although Head Start provides many enriched experiences for children from low-income families, the program is only available for a year or two, whereas children are with their parents until they reach adulthood. Thus, any Head Start-induced changes in parenting that are sustained for a year or more could be a means by which Head Start can have long term impacts. While a year of Head Start cannot be expected to “inoculate a child against continuing disadvantage” (Zigler & Berman, 1983, p. 894), if it can convince parents to be less harsh with their children and more engaged in their learning, Head Start may help parents provide an essential buffer to economic disadvantage.

Figure 3.

Observed direct and indirect impact of Head Start on children’s spelling skills.

Note Although not illustrated, within time measures were co-varied. Hypothesized paths of indirect treatment impact are bolded. All values represent standardized coefficients. Magnitudes of indirect effects are presented in Table 4. *** p < .001. ** p <.01.* p <.05.