ACF OPRE: Head Start Family and Child Experiences Survey (FACES)

CHAPTER V

Relationship of Program and Classroom Characteristics to Children’s Cognitive Gains and Social Development in Head Start

The Family and Child Experiences Survey that began in fall 2000 (FACES 2000) found that the quality of Head Start centers and classes was generally good as judged by observational instruments that are widely employed to assay the quality of early childhood learning environments. An earlier round of the study that began in fall 1997 (FACES 1997) had similar findings. Both longitudinal studies also found that children in Head Start made significant progress toward national averages in some areas of early academic knowledge and skill, notably vocabulary knowledge and early writing skills. But in other areas, notably letter recognition and early math skills, children held their own but did not draw closer to national averages during the Head Start year. Even in those areas where they made significant progress, they still entered kindergarten considerably behind their more advantaged peers.

RESEARCH QUESTIONS

The analysis reported in this chapter used multilevel regression models to address the following research questions:

Do Head Start programs and classes differ in the average achievement levels that children have attained when they leave the program? Do they differ in the cognitive gains children make during the program year? Do they differ in the extent of changes children show in their cooperative social behavior or conduct problems?
Do children in programs that employ one of the two integrated curricula that are widely used in Head Start—Creative Curriculum or High/Scope—show larger cognitive gains or behavioral improvements than children in programs that employ other curricula?
Do children in programs that have more ample resources, as indicated by paying higher salaries to their lead teachers, show larger cognitive gains or behavioral improvements than children in programs that have less ample resources?
Do children in Head Start classes that are of higher quality, as indicated by their receiving higher scores on the Language scale of the Early Childhood Environment Ratings Scale–Revised (ECERS-R) or the Caregiver Interaction Scale (CIS), show larger cognitive gains or behavioral improvements than children in classes that are of lower quality on these measures?
Do children in Head Start classes led by better prepared teachers show larger cognitive gains or behavioral improvements than children in classes led by teachers who are less well prepared? Indicators of teacher preparation that were examined included whether the teacher had a bachelor’s degree or associate’s degree, her years of teaching experience, her annual salary as a deviation from the program mean salary, and her score on a scale that measured positive attitudes and knowledge about early childhood educational practices.
Do children who participate in Head Start classes for a longer period each day—who attend “full-day” classes—show larger cognitive gains or behavioral improvements than children who participate for a shorter period of time—who attend “part-day” classes?
Do children in Head Start classes with lower child:staff ratios, and that provide more attention to the needs and preferences of individual children, as indicated by a higher score on the Assessment Profile Individualizing scale, show larger cognitive gains or behavioral improvements than children who are in classes with higher child:staff ratios or lower Individualizing scores?
Do children whose parents do more educational activities at home with their children show larger cognitive gains or behavioral improvements than children whose parents do fewer educational activities? We examined whether children whose parents read to them on a daily basis at home showed greater gains in Head Start than children whose parents read to them less often. Frequency of reading was reported by the parents themselves in parent interviews in the fall of 2001.

Why was this the case? Variations across local Head Start programs—in the achievement gains that children make in Head Start and in the levels of skill and knowledge with which they leave the program—offer potential explanations. Variations in cognitive and social-emotional development could be related to differences in classroom quality. Or they could be associated with differences in the type of curriculum or instructional approach that the programs or classroom teachers employ.

If this were indeed the case, then the performance of the national Head Start program might be improved by encouraging more local programs to improve their quality or make use of curricula or instructional approaches found to be associated with greater gains in children’s knowledge and skills or greater improvements in their classroom conduct and social-emotional well-being.

This chapter explores variations in child achievement and behavior across local Head Start programs and classes. It uses multilevel modeling to test hypotheses about early education program and classroom characteristics that many child development scholars believe to be associated with enhanced cognitive growth or emotional maturation in preschool children (Phillips, Mekos, Scarr, McCartney, & Abbott-Shim, 2000; Whitebook, Howes, & Phillips, 1989). The characteristics that are thought to make a difference for program effectiveness include the following:

using an integrated and comprehensive preschool curriculum;
having more ample program resources;
providing classrooms that are of higher quality as early learning environments;
employing a better prepared teaching staff;
providing preschool services for a longer period each day;
conducting educational activities in smaller groups with more personal adult attention to the needs and preferences of individual children; and
encouraging parents to engage in more educational activities with their children at home.

This chapter examines relationships between these characteristics and several measures of children’s cognitive development and classroom behavior, while controlling for the influence of other variables. Cognitive dependent variables consisted primarily of direct assessment measures of children’s letter recognition and pre-reading skills, vocabulary knowledge, early writing skills, and early math skills. Dependent variables in the social-emotional realm consisted of teacher and parent ratings of children’s approaches to learning tasks and cooperative and problem behavior. Control variables included measures of the socioeconomic and ethnic composition of the families and children participating in each program and classroom. Other control variables were characteristics of the child like age, sex, and disability status and measures of parents’ literacy skills.

FACES 2000 included a wider range of program and classroom characteristics that could be related to differences in achievement than did FACES 1997. FACES 2000 added interview questions and observational procedures that looked more systematically at areas like the curriculum used in each center and classroom, training and support for that curriculum, teacher salary levels, teacher knowledge and beliefs, and the use of child portfolios and other procedures aimed at individualizing instruction. In addition, the sample design was modified to yield a larger number of sample children in each sample classroom. This produced more stable estimates of class means and more variation in child characteristics in each classroom subsample. The modified design made it possible to carry out multilevel regression analysis at three levels: the program, classroom, and child level. In FACES 1997, multilevel regression analysis could be carried out at only the center and child levels.

The multilevel analysis of FACES 2000 data did indeed show that some of the program and classroom level characteristics listed above were significantly related to variations in the size of the gains children made in Head Start. In presenting the analysis findings, we enumerate the program and classroom characteristics that seemed to make a difference. We describe the nature and size of the relationships involved. We also list factors that were not significantly related to gains in achievement or behavior and discuss possible reasons why hypothesized relationships failed to materialize.

CONCEPTUAL FRAMEWORK

The conceptual framework that guided our analysis was a multilevel, multi-causal model of the influences that shape children’s cognitive and social-emotional development and the factors that help determine the nature of the experience children have in Head Start. (See Figure 5.1.) This view posits that children’s development in the early years is primarily a function of the experiences they have in their families. Children from low-income families, whose parents tend to have lower educational attainments than other parents, often do not experience the same extent or quality of intellectual stimulation at home as children from middle-class families (Phillips et al., 1998). Furthermore, their parents are less able to purchase high-quality supplementary or substitute care in the marketplace. In some cases, children from low-income families may also not receive as much emotional support from parents as they need for optimal development. A center-based early childhood learning environment such as Head Start may help provide experiences that would be beneficial for the development of all children, but especially for those from higher-risk family environments (NICHD Early Child Care Research Network, 2000). Furthermore, parents’ involvement in their children’s educational experiences may also be an important factor, and one that programs can foster.

Figure 5.1. Analytical Model of Multilevel Factors Predicting to Classroom Quality and Children’s Achievement and Gains in the Head Start Year
[D]

The nature of the learning environment that a given child experiences in Head Start depends on the training and experience of teachers in the program, and the resources available to them in terms of facilities, materials, and teaching assistants. Programs with more resources are likely to be better able to provide adequate facilities and materials and recruit and retain talented and well-prepared teachers (Whitebook, Howes, & Phillips, 1989).

But the character of the classroom environment that a given Head Start program is able to provide for a child is not just a question of program resources. It also depends on the educational philosophy to which the program adheres, and the kind of curriculum centers and teachers are encouraged to follow. Other things equal, children would be expected to do better in programs that employ well-thought-out curricula that are comprehensive and integrated in terms of educational activities and assessment methods. This is especially the case if the program is able to provide teachers with adequate training and support in the curriculum. At the same time, children’s progress in a given cognitive or social-emotional area depends on whether the program’s basic philosophy and curriculum of choice are supportive of efforts to bolster that area of child development.

ANALYSIS METHOD

The analysis method used to examine associations between Head Start program and class characteristics and children’s cognitive and social-emotional development was multilevel linear regression modeling, using the SAS PROC MIXED computer program (Singer, 1998; Bryk & Raudenbush, 1992). Multilevel modeling shows how the average achievement scores of a sample of classes, schools, or other educational units (in the present case, Head Start programs and classes) relate to a set of characteristics of those units, such as measures of program demographics and classroom quality. Simultaneously, this type of modeling can examine how the achievement scores of individual children in each program and class relate to a set of child-level characteristics, such as child demographics and home literacy activities. The method provides a numerical estimate of how sizable the program-to-program and class-to-class variation in average scores is, relative to the child-to-child variation in scores within classes.

The primary dependent variables were the gains each child made between the fall and the spring of the Head Start year in their cognitive assessment or behavior ratings scale scores. Models were also constructed of assessment and ratings scores attained by Head Start children in the FACES national sample in the fall of 2000 and the spring of 2001.21 Each analytic model had three levels. The first level involved variation in average assessment scores or average gain scores across the 43 programs in the FACES national sample, expressed as deviations of the program means from the overall mean score for the entire sample. The second level involved variations of class means from the overall program means. And the third level involved variation in individual children’s scores or gain scores around the class means.

There were three levels of independent variables used to model or predict the assessment scores or gain scores. At the program level, the independent variables consisted of measures that represented the curriculum employed by the program, average teacher salary levels in the program, and average demographic and socioeconomic characteristics of the children who attended each program and their families. At the classroom level, independent variables consisted of measures of teacher preparation, teacher background characteristics, whether the class was of full-day or part-day duration, and indicators of classroom quality such as the ECERS-R Language scale and Caregiver Interaction Scale. Class-level variables also included measures of the demographic composition of the class, expressed as deviations from the average demographic characteristics of the program. At the child level, the independent variables were measures that represented demographic characteristics of the child; socioeconomic, cultural, and structural characteristics of the family; parent literacy levels; disability status of the child; and the frequency of parental reading to the child.

Statistical tests were made as to whether a given set of independent variables (program-level, class-level, and child-level) improved the model’s fit to the data, over and above simpler models that did not include that set. Tests were also done as to whether the regression coefficient for a given independent variable was reliably greater than zero. Details about variable definitions, means and ranges, reliability of measures, and statistical tests used to ascertain the reliability of findings are described in the Appendix of the report.

FINDINGS

Multilevel regression analysis of the assessment and ratings measures showed that there were significant relationships between some of the program and class characteristics identified above and variations in children’s cognitive or behavioral gains. The following factors seemed to make a difference for children’s progress:

programs using an integrated curriculum, particularly the High/Scope Curriculum;
programs having higher teacher salaries;
teachers having bachelor’s or associate’s degrees;
children attending full-day rather than part-day classes; and,
parents reporting that they read to their children more frequently.

None of these factors was related to increased gains in all cognitive or behavioral indices examined. But most were related to gains on two or more outcome measures, and the relationships were in the hypothesized direction. The following sections describe the nature and size of these relationships.

A. Use of an Integrated Curriculum

The High/Scope Curriculum is a comprehensive, integrated preschool curriculum that has a long history of research and development. It grew out of the Perry Preschool Project, an intensive, relatively small-scale intervention that was found to have long-term effects on children’s achievement in a random-assignment evaluation study. It is the second most popular curriculum employed in Head Start programs, as described in Chapter III of this report. In the present analysis, children in programs that employed the High/Scope Curriculum were found to have made greater gains than children in programs that did not employ one of the two integrated curricula that are most widely used by local Head Start programs. (In the interests of simplicity, henceforth we shall refer to these curricula other than Creative Curriculum or High/Scope by the term “other curricula.”) The greater gains were on measures in both the cognitive and social-emotional domains.

Gains in Pre-Reading and Oral Communication Skills
Children in programs that employed the High/Scope Curriculum made small but significantly greater gains in letter recognition skills than children in programs that employed other curricula. In IRT scale-score terms, the average Head Start child made a gain on the Wood cock - Johnson - Revised - Letter - Word Identification task of just under 10 scale points from fall to spring of the Head Start year. Children in programs employing the High/Scope Curriculum showed an average gain of 12.6 scale points on the WJ-R LWI from fall to spring (p < .001), whereas children in programs employing the Creative Curriculum or other curricula made gains of about 9 scale points (p < .001). The regression coefficient for the High/Scope Curriculum in the three-level regression analysis, which is an estimate of the difference in average gains between it and other curricula adjusted for the influence of related variables, was 3.66 (p = .01).

The other popular integrated curriculum, Creative Curriculum, had a positive coefficient in the regression analysis of Letter-Word Identification scores in the spring, but did not reach significance (2.66, p < .10). It also had a positive coefficient in the gain analysis, but one that was not significantly different from zero (1.92, n.s.).

When differences were expressed in standard score terms, the High/Scope group showed a mean gain of 1.6 standard score points, going from a mean standard score of 92.2 in the fall to 93.8 in the spring. (Figure 5.2) By comparison, children in programs employing the other widely popular integrated curriculum, the Creative Curriculum, or other curricula held their own against national norms.

Children in programs using the High/Scope Curriculum were found to make greater gains as well on a criterion-referenced measure of oral communication skills. This difference was only at the trend level, however. The “Social Awareness” measure assessed children’s ability to tell an adult basic information about themselves such as their age, and month and year of birth.The regression coefficient for the High/Scope Curriculum showed a gain 0.21 points greater than that for other curricula (p = .098).

Improvement in Cooperative Behavior
The use of an integrated curriculum was found to be associated with greater gains for children in the social-emotional realm. Children in programs employing the High/Scope Curriculum showed larger gains in cooperative classroom behavior from fall to spring of the program year than children in programs employing other curricula. They also showed more pronounced declines in hyperactive behavior from fall to spring.

Children in programs following the High/Scope Curriculum showed a mean increase of 2.3 points on the Cooperative Classroom Behavior rating scale completed by Head Start teachers (a change equivalent to .48 of a standard deviation).They went from a mean score of 14.5 in the fall to a mean of 16.8 in the spring (p <.001). By comparison, children in programs following the Creative Curriculum showed an increase of 1.8 points (p <.001), and those in programs following other curricula increased by 1.7 points (p <.001) (.37 and .36 of a standard deviation, respectively). The regression coefficient for the High/Scope Curriculum in the three-level regression analysis, which is an estimate of the difference in average gains between it and other curricula adjusted for the influence of related variables, was 1.26 (p < .05).

Figure 5.2. Children in Head Start Programs Using High/Scope Curriculum Show Greater Gains in Letter Recognition Skills
[D]

Decline in Hyperactive Behavior
Children in programs following the High/Scope Curriculum exhibited significant improvement in their scores on a Total Behavior Problems rating scale completed by Head Start teachers (p = .03). In particular, they showed greater improvement on the Hyperactive subscale of the Problem Behavior rating scale. They showed a mean decline of 0.19 points (p < .10) on this scale, going from a mean rating of 1.39 in the fall to a mean of 1.20 in the spring (a change of .13 of a standard deviation). By comparison, children in programs following the Creative Curriculum showed a non-significant mean decline of -.08 points, while children in programs following other curricula showed a non-significant mean decline of -.13 points. The regression coefficient for the High/Scope Curriculum in the three-level regression analysis, which is an estimate of the difference in average declines between it and other curricula, adjusted for the influence of related variables, was -0.32 (p < .05). In the regression analysis of Total Behavior Problems, the coefficient for the High/Scope Curriculum had a value of -1.19 (p = .01).

B. Higher Teacher Salary Levels

We explored whether children attending Head Start programs with higher average teacher salary levels would make greater progress in their cognitive and social-emotional development. The multivariate analysis showed this to be the case with respect to children’s pre-reading and oral communication skills, and their cooperative and problem behavior in the classroom.

Gains in Pre-Reading and Oral Communication Skills
Average Annual Salary for Lead Teachers was associated with greater gains in letter recognition. The regression coefficient for Mean Teacher Salary Level in the three-level regression analysis of gains, which is an estimate of the difference in LWI scale scores associated with each $10,000 increment in average teacher salaries, adjusted for the influence of related variables, was 1.96 (p = .009). In standard score terms, the highest teacher salary group (top quartile) showed a gain of less than one standard score point. Children in programs with lower average teacher salary levels (bottom three quartiles) showed a slight and non-significant decline in their standard scores.

Children in programs with higher average teacher salaries made greater gains as well on the criterion-referenced “Social Awareness” measure.The regression coefficient for the Mean Teacher Salary Level showed an increased gain of 0.18 points for every $10,000 increment in mean salary (p = .008).

Improvement in Cooperative Behavior
Higher teacher salaries were found to be associated with greater gains for children in the social-emotional realm. Children in programs with higher average teacher salary levels showed larger gains in cooperative classroom behavior from fall to spring of the program year than children in programs with lower teacher salary levels. They also showed more pronounced declines in hyperactive behavior from fall to spring.

Children in programs in the highest quartile of teacher salaries showed a mean increase of 3.1 points on the Cooperative Classroom Behavior rating scale completed by Head Start teachers (a change equivalent of .65 of a standard deviation). They went from a mean score of 14.6 in the fall to a mean of 17.7 in the spring (p <.001). By comparison, children in programs in the middle two quartiles of teacher salary showed an increase of 1.7 points (p <.001), while those in programs in the lowest quartile increased by 1.5 points (p <.001) (equivalent to .34 and .30 of a standard deviation, respectively). (Figure 5.3.) The regression coefficient for Mean Teacher Salary Level in the three-level regression analysis, which is an estimate of the increase in average gains for every $10,000 increment in mean salary, adjusted for the influence of related variables, was 1.18 (p < .001).

Decline in Hyperactive Behavior
Children in programs with higher teacher salary levels exhibited significant improvement in their scores on the Hyperactive subscale of the Problem Behavior rating scale completed by Head Start teachers. Children in programs in the highest quartile on teacher salary levels showed a mean decline of 0.35 point (p = .013) on this scale, going from a mean rating of 1.39 in the fall to a mean of 1.04 in the spring (an effect size of .23 of a standard deviation). By comparison, children in programs in the middle two quartiles on teacher salary level showed no change (-0.12 point), while children in programs in the lowest quartile also showed no change (-.01 point). (Figure 5.4.) The regression coefficient for Mean Teacher Salary Level in the three-level regression analysis, which is an estimate of the difference in average declines for every $10,000 increment in mean teacher salary, adjusted for the influence of related variables, was -0.18 point (p < .05).

C. Teachers With Bachelor’s or Associate’s Degrees

The possession of a four-year college degree or an associate’s degree in education or a closely related field is among the most widely accepted indicators of teacher preparation. One of the current performance goals of the national Head Start program is to have all local programs staffed by teachers of whom a majority have bachelor’s degrees or associate’s degrees. We explored whether the lead teacher having a BA or AA degree made a difference in children’s progress on cognitive or social measures.

Figure 5.3. Children in Head Start Programs With Higher Teacher Salaries Show Larger Gains in Cooperative Classroom Behavior
[D]

Figure 5.4. Children in Head Start Programs With Higher Teacher Salaries Show Larger Declines in Hyperactive Behavior
[D]

On several of the cognitive assessment measures, children in classes taught by teachers with BA or AA degrees ended the program year with mean scores that were higher than those of children in classes taught by teachers with less than an AA degree. However, these children had also had higher mean scores at the beginning of the year. This may reflect a situation in which Head Start programs that hire teachers with college credentials tend to serve families with higher parent education and income levels than are typical for Head Start nationwide. What was less certain was whether the children taught by teachers with higher educational credentials made greater gains from fall to spring than children taught by teachers with lesser credentials. Early writing skills was one cognitive area in which there was evidence of greater gains as well as higher achievement levels. However, the evidence was not unambiguous.

Gains in Early Writing Skills
In standard score terms, and without adjustment for the effects of related variables, the picture was reasonably clear. Children in Head Start classes taught by lead teachers with bachelor’s degrees or associate’s degrees had higher mean scores on the Woodcock Johnson-Revised Dictation task in the fall—86.3 and 84.5, respectively—than children in classes taught by teachers with less than an associate’s degree, who had a mean score of 83.9. But the children in the former classes also made significant gains toward national averages, whereas children in the latter group did not. The gains were 2.48 standard score points (p = .03) for children whose teachers had bachelor’s degrees; 2.55 standard score points (p = .03) for children whose teachers had associate’s degrees; and 1.67 standard score points (p = .21) for children whose teachers had less than an associate’s degree. (Figure 5.5.) The respective gains represented effect sizes of .18, .19, and .12 of a standard deviation.

Figure 5.5. Children in Head Start Classes Taught by Teachers With Bachelor’s or Associate’s Degrees Show Gains in Early Writing Skills
[D]

In the three-level analysis of spring Dictation scores, children in classes taught by teachers with BAs or AAs had a significant regression coefficient of 6.14 IRT scale points (p = .01). This meant that children in these classes had a mean score in the spring that much higher than the mean for children in classes taught by teachers without those credentials. In the fall analysis, children in classes taught by teachers with BAs or AAs had had a regression coefficient that was also significant, though apparently smaller (5.46 scale points, p = .009), which meant that they started with higher scores in the fall. The class-level variables as a set did not improve model fit in the fall, whereas they did in the spring. However, in the multilevel analysis of gains on the Dictation task, the regression coefficient for full-day classes (0.63) was not significant (p = .77).

D. Full-Day Versus Part-Day Classes

As of the 2000-2001 school year, the majority of children who attended Head Start participated in part-day classes that were conducted in morning or afternoon sessions only.We explored whether children benefited more from the program in terms of academic achievement if they attended full-day classes. Children in FACES 2000 who did attend full-day Head Start programs made greater gains in several areas than children who attended part-day.

Gains in Pre-Reading and Early Writing Skills
Children in full-day Head Start classes made larger gains in letter recognition skills than children in part-day classes. Children in full-day Head Start classes showed a mean gain on the Woodcock-Johnson Revised Letter-Word Identification task of 12 points in IRT scale-score terms (p < .001). Children in part-day classes showed a mean gain of 8.7 points (p < .001). The regression coefficient for full-day classes in the three-level regression analysis, which is an estimate of the difference between these classes and part-day classes adjusted for the influence of related variables, was only significant at the trend level (1.81, p = .067).

In standard score terms, the full-day group showed an average gain on the WJ-R LWI task of 1.2 standard score points (p = .06), whereas the part-day group merely held their own against national averages, showing a non-significant decline of 0.9 points (p = .12). (Figure 5.6.)

Children in full-day Head Start classes made greater gains as well in early writing skills, although the statistical evidence here was more ambiguous. In standard score terms, and without adjustments for the effects of related variables, children in full-day classes had a mean gain from fall to spring of 3.5 standard score points (p = .004) on the Woodcock-Johnson Revised Dictation task.They went from a mean of 84.8 in the fall to a mean of 88.3 in the spring. The gain was equivalent to an effect size of .25. By contrast, children in part-day classes went from a mean of 85.0 in the fall to a mean of 86.1 in the spring, a non-significant difference of 1.1 standard score points (p = .162). (Figure 5.7.)

In the three-level analysis of spring Dictation scores, children in full-day classes had a significant regression coefficient of 7.80 IRT scale points (p = .005). This meant that children in full-day classes had a mean score in the spring that much higher than the mean for children in part-day classes, with related factors controlled. In the fall analysis, children in full-day classes had had a regression coefficient that was only marginally higher (4.10 scale points, p = .086). Furthermore, the class-level variables as a whole did not improve model fit in the fall, whereas they did in the spring. However, in the multilevel analysis of gains on the Dictation task, the regression coefficient for full-day classes (3.85) was not significant (p = .129).

E. More Frequent Parental Reading to Children

Children are in preschool programs for only a limited time, both in terms of hours of each day and months out of the child’s life. However, preschool programs may extend their influence by encouraging parents to engage in more frequent and more effective educational activities at home with their children. The national Head Start program recognized the importance of this function by stating, in its performance measures framework, that one of the major objectives of the program is, to “strengthen parents as the primary nurturers of their children.” Therefore, we decided to consider frequency of parental reading as an additional variable of interest in considering children’s outcomes. Analysis showed that more frequent parental reading in the fall was associated not only with higher initial achievement for children as they entered the program, but also with larger gains during the program year. Larger gains were observed both in vocabulary knowledge and letter recognition skills.

Figure 5.6. Head Start Children in Full-Day Classes Show Larger Gains in Letter Recognition Skills Than Those in Part-Day Classes
[D]

Gains in Vocabulary Knowledge
Parents were asked whether they read to their children, “not at all,” “once or twice,” “three to six times,” or “every day” during the previous week. Parental responses to the question were entered into the three-level regression analysis as a set of dichotomous variables, with the most frequent response, “three to six times,” as the omitted reference category. The reading responses were entered as child-level independent variables.

In the regression analysis of fall vocabulary test scores, children whose parents reported reading to their children “not at all” or “once or twice” had significantly lower mean scores than children whose parents reported reading “three to six” times. The mean score for children whose parents said they read “every day” was not significantly different from that of the “three to six times” group. In terms of IRT scale scores on the Peabody Picture Vocabulary Test, Third Edition, the mean for the “not at all” group was 1.93 points lower (p < .05), and the mean for the “once or twice” group was 1.83 points lower (p < .001), than the mean for the “three to six times” reference group.

Figure 5.7. Children in Full-Day Head Start Classes Show Greater Gains in Early Writing Skills
[D]

In the regression analysis of the spring PPVT-III scores, the “not at all” and “once or twice” groups again had mean scores that were significantly lower (by 1.83 scale points, p < .05 and 1.35 points, p < .01, respectively) than that for the “three to six times” reference group. But now the “every day” group had a mean score that was significantly higher than the reference group mean (by 1.17 scale points, p < .01). Thus, there was a 3 scale-point difference between the vocabulary means of the highest and lowest reading groups. In the regression analysis of vocabulary gains, the “every day” group had a larger gain than the reference group (by 0.68 scale points), although the difference was only significant at the trend level (p = .093).

These differences related to frequency of parental reading were obtained even after controlling for parent education level, the mother’s score on a measure of adult literacy (the K-FAST), and an indicator of the presence of books in the home. These measures were also entered into the regression analysis as child-level variables, and all were significantly related to children’s vocabulary test scores in the fall and spring of the program year. None was related to the size of fall-spring gains in vocabulary scores, however.

The picture was similar, though not identical, when looked at in terms of mean standard scores for the parental reading groups without adjustments for the effects of related variables. (Figure 5.8.) All four groups showed significant gains in their vocabulary standard scores from fall to spring of the Head Start year. But the gain was smallest for the group whose parents said they read to the child “not at all” in the previous week (2.1 standard score points, p < .05). And the gain was largest for the group whose parents said they read to the child “every day” (4.6 standard score points, p < .001). However, the gain for the group whose parents read only “once or twice” was also sizable (3.8 standard score points, p < .001). When the gains were seen in terms of effect sizes, they ranged from .14 to .32 of a standard deviation, with the “every day” reading group having the largest effect size.

Gains in Pre-Reading Skills
The parental reading groups showed differences in the gains children made on the Woodcock-Johnson Revised Letter-Word Identification task. Children whose parents reported reading to them only once or twice a week or less did not make as large gains in letter recognition skills as children whose parents reported reading to them three times a week or more. In the regression analysis of fall scores on the LWI test, the group means lined up in a fashion similar to that seen in the vocabulary analysis, but differences were not statistically significant. In the regression analysis of spring LWI scores, however, both the “not at all” reading group and the “once or twice” reading group had significantly lower means than the “three to six times” reference group. The respective differences, in terms of IRT scale scores, were -3.74 points (p < .10) and -2.95 points (p < .05). The “every day” reading group had a mean score that was not significantly different from that of the reference group.

Similar results were obtained in the regression analysis of fall-spring gains on the LWI task. Both the “not at all” reading group and the “once or twice” reading group had significantly smaller gains than the “three to six times’ reference group. The respective differences in gains, in terms of IRT scale scores, were -3.23 points (p < .10) and -2.30 points (p < .05). The “every day” reading group had a mean gain that was not significantly different from that of the reference group. Again, these results controlled for the effects of parent education, parental literacy level, and the presence of books in the home.

Figure 5.8. Head Start Children Whose Parents Read To Them More Often Show Larger Vocabulary Gains
[D]

F. Classroom Quality Indicators Not Found to Relate to Gains

The five program and class characteristics described above showed significant or marginally significant relationships with children’s gains in Head Start. But there were two sets of characteristics that did not show the relationships with children’s gains that were hypothesized. These were the indicators of classroom quality, and the indicators of child:staff ratio and individualized attention to the needs of each child.

As described in detail in Chapter IV and the Appendix, the FACES 2000 classroom observation battery contained a number of widely accepted indicators of the general quality of Head Start (or other preschool or childcare) environments. Two of these measures were chosen for inclusion in the three-level regression analysis of children’s gains. One was the Language scale, a component of the Early Childhood Environment Rating Scale-Revised (ECERS-R). This component scale consists of a series of observational items and ratings that deal with the frequency and quality of class activities related to oral language development, vocabulary building, and the nurturing of pre-reading and early writing skills. Thus, it seemed likely to relate to children’s gains in these areas.

The second classroom quality indicator included in the three-level models was the Caregiver Interaction Scale, an observational rating measure of the emotional tone of teacher-child interaction and the lead teacher’s sensitivity to children’s needs and feelings. It seemed likely to relate to children’s gains in the social-emotional domain. Neither of these expectations was supported by the FACES findings.

ECERS-R Language Scale
Children in classrooms with higher ECERS-R Language scores had higher vocabulary test scores in the fall than children in classrooms with moderate or lower ECERS-R Language scores. In terms of standard scores, the mean score on the PPVT-III in the fall for children in Head Start classes in the highest quartile on the ECERS-R Language scale was 84.4. For children in classes in the middle two quartiles on the Language scale, the mean score was 80.5, while for children in classes in the lowest quartile, the mean score was 79.4. But children in higher, moderate, and lower quality classes all showed similar gains in vocabulary knowledge from fall to spring. Children in the highest quartile classes showed standard score gains of 4.26 points (p < .001, effect size of .25). Children in the middle two quartile classes showed standard score gains of 4.32 points, (p < .001, effect size of .25). And children in the lowest quartile classes showed standard score gains of 4.59 points (p < .001, effect size of .27). Thus, the same differences in vocabulary knowledge between higher and lower quality classes that were observed in the fall were still present in the spring. (Figure 5.9.) The higher quality language environments were not associated with larger gains in vocabulary knowledge.

Indeed, when relationships with other variables were controlled in the three-level regression analysis, even the differences in fall vocabulary scores associated with ECERS-R Language scores proved not statistically significant. This implies that the differences were better accounted for by the operation of other, related variables.

Higher ECERS-R Language scores were not associated with larger gains in letter recognition, early writing skills, or early math skills. Nor were they associated with improvements in children’s cooperative or problem behavior in the classroom. Although the ECERS-R Language scale was the classroom quality indicator entered in the three-level regression models reported here, other correlation and regression analysis showed that the conclusions would not have been different had another quality indicator, such as the ECERS-R Total Score, been used instead. A Quality Factor weighted composite score was developed that incorporated the Assessment Profile Learning Environment and Scheduling scales as well as the ECERS-R Language scale. Analysis with that score produced non-confirmatory results as well.

Figure 5.9. Children in Head Start Classes With Higher, Moderate, and Lower Quality Language Activities Show Parallel Gains in Vocabulary Knowledge
[D]

Caregiver Interaction Scale
We hypothesized that the Caregiver Interaction Scale, an observation-based rating scale that reflects the sensitivity and emotional tone of teacher-child interaction, would be associated with improvements in children’s cooperative and problem behavior in the classroom. This proved not to be the case. Higher CIS scores for the classroom teacher were associated with higher cooperative behavior ratings in the spring 2001 at only the trend level (p < .10). In the three-level analysis of gains in cooperative behavior, the coefficient for CIS scores had a value of zero. CIS scores were not significantly associated with reductions in Hyperactive Behavior or Total Behavior Problems either.

Higher CIS scores were not associated with greater gains in vocabulary, letter recognition, or early math skills. There was a significant positive association between higher CIS scores and higher mean scores on the WJ-R Dictation task in the spring of 2001. The regression coefficient signified a Dictation score that was .29 IRT scale points higher for each unit increase in the lead teacher’s CIS score (p = .03). However, higher CIS scores were not significantly associated with greater fall-spring gains in this measure of early writing skills (p = .20).

Child:Staff Ratio and More Individual Attention
We hypothesized that lower child:adult ratios in Head Start classroom activities, and more attention to the needs and preferences of individual children would result in greater gains for children. These hypotheses were not confirmed by the data. Indeed, on some outcome measures, children actually showed greater gains in classrooms with more children per adult.

Child:Staff Ratios
The mean child:staff ratio was a figure derived from counting the number of children in the Head Start classroom and dividing that number by the number of teachers or other adult staff members actively interacting with children. These counts were taken at two separate occasions on the day that the classroom was observed, and the two resulting ratios were averaged. When the mean child:staff ratio was entered into the three-level regression analysis of children’s cognitive gains as a class-level variable, it proved not to be significantly associated with gains in vocabulary knowledge, early writing, or early math skills. In the vocabulary analysis, the results were much like those for the ECERS-R Language score. That is, children in classes with lower, moderate, or higher child:staff ratios all showed roughly equivalent gains in vocabulary knowledge from fall to spring of the Head Start year.

In the analysis of children’s pre-reading skills, results were opposite to what was predicted. Children in classes with higher child:staff ratios made significantly larger gains in letter recognition skills than those in classes with lower ratios. In terms of standard scores, children in the highest quartile of classes on the child:staff ratio showed an average gain of 1.2 standard score points on the WJ-R Letter-Word Identification task from fall to spring. They went from a mean standard score of 92.1 in the fall to a mean score of 93.3 in the spring. By comparison, children in classes in the middle two quartiles and lowest quartile merely held their own against national norms, showing non-significant declines of -0.6 standard score points (p = .28) and -0.9 standard score points (p = .55), respectively. (Figure 5.10.)

Figure 5.10. Children in Head Start Classes With Higher Child:Adult Ratios Show Larger Gains in Letter Recognition
[D]

In the three-level regression analysis of LWI gains, the value of the regression coefficient for the mean child:staff ratio was 0.57 (p < .05). This may be interpreted as the change in gain in LWI IRT scale scores that would be expected for every unit increase in the child:staff ratio, net of the effects of other related variables. If the original hypothesis had been confirmed, this coefficient would be negative. Instead, it was reliably greater than zero in the positive direction.

Higher child:staff ratios were also associated with behavioral gains. In the three-level regression analysis of gains in Cooperative Classroom Behavior, the coefficient for mean child:staff ratio was 0.15 (p = .086). In the analysis of declines in the Total Behavior Problems, the coefficient for mean child:staff ratio was -0.20 (p = .03). Both of these relationships were in the direction opposite to the expected one.

Assessment Profile Individualizing Scale
The Individualizing scale of the Assessment Profile instrument uses both observational and interview methods to assess the degree to which preschool teachers track the accomplishments of children in their classes and provide activities suited to the capabilities and interests of individual pupils. Class-level scores on this instrument did not relate to gains in any of the cognitive development areas. Nor did they show associations with improvements in the measures of social-emotional development.

SUMMARY AND DISCUSSION

This chapter explored variations in child achievement and behavior across local Head Start programs and classes. It used multilevel modeling to test hypotheses about early education program and class characteristics that many child development scholars believe to be associated with enhanced cognitive growth or emotional maturation in preschool children. The conceptual framework posited that the gains a child makes in Head Start depend on the nature of the learning environment that he or she experiences in the local program. The nature of the learning environment depends in turn on the training and experience of teachers in the program and the resources available to them in terms of facilities, materials, and teaching assistants. Programs with more resources are likely to be better able to provide adequate facilities and materials and recruit and retain well-prepared teachers. Another hypothesis was that children would make larger gains in programs employing curricula that are comprehensive and integrated in terms of educational activities and assessment methods. Other expectations were that children would make more sizable gains in higher quality classrooms, in full-day as opposed to part-day classes, in classes with better child:staff ratios and more individualized attention to pupils, and in families where parents engaged in more educational activities with their children.

Analysis of longitudinal data from FACES 2000 showed that children’s gains in Head Start were significantly related to several of the hypothesized characteristics of programs and classes. Specifically:

Use of an integrated curriculum was linked to greater gains in several cognitive and social-emotional areas. Children in Head Start programs using High/Scope showed larger fall-spring gains in letter identification and cooperative classroom behaviors than children in programs using other curricula. Children in programs using High/Scope also showed greater improvement in total behavior problems and hyperactive problem behavior.
Higher teacher salaries were linked to greater gains in several cognitive and social-emotional areas, including letter identification and cooperative classroom behavior. Children in programs with higher teacher salaries also showed greater improvement in hyperactive problem behavior during the Head Start year.
Teachers’ educational credentials were linked to greater gains in early writing skills. Children taught by Head Start teachers with bachelor’s degrees or associate’s degrees showed gains toward national averages in an assessment of early writing skills, whereas children taught by teachers with lesser credentials merely held their own against national norms.
Provision of preschool services for a longer period each day was linked to greater cognitive gains. Children in full-day classes in Head Start showed larger fall-spring gains in letter recognition and early writing skills than did children in part-day classes.
There was indirect evidence that encouraging parents to engage in more educational activities with their children at home could serve as a pathway to greater cognitive gains. Children whose parents reported reading to them every day showed larger fall-spring gains in vocabulary knowledge and letter recognition skills than children whose parents reported reading once or twice or less frequently per week.

Other analytic results were not in line with expectations. In particular:

Within the generally good quality range of Head Start classrooms, variation in quality as measured by the ECERS-R Language scale or the Caregiver Interaction Scale was not associated with differences in fall-spring achievement gains across classes.
Within the narrow range of child:staff ratios in Head Start, variation in child:staff ratios was not associated with or was negatively associated with differences in fall-spring achievement gains across classes.

The analysis results were generally supportive of the conceptual framework that the amount of resources available to a Head Start program and the curricular approach it uses can make a difference for children’s progress in the program. The results also supported the notions that children could make greater gains if they had more exposure to comprehensive, integrated preschool activities. While we were not able to consider the program’s effect on parental reading, it appears that benefits accrue from increased frequency of educational activities at home. At the same time, some provisos about the results should be noted.These include the following points:

Differences in cognitive gains, while statistically significant, were relatively modest in magnitude. By itself, each of the differences was not large enough to close the gap between where Head Start children typically end up at the end of the program year and the average achievement levels of American children at the start of elementary school. If several of the positive characteristics could be implemented simultaneously in a program, they might jointly make a more sizable difference, however.
Program- and classroom-related gains varied across cognitive areas. Significant gains were seen primarily with respect to letter recognition and early writing skills. The important areas of vocabulary and early math skills showed little variation in gains that was linked to specific program or class characteristics. Rather, children in programs and classes with different characteristics tended to show parallel gains (in vocabulary) or similar lack of gains (in early math skills).
Differences in achievement levels at the end of the Head Start year between children in Head Start programs with differing socioeconomic and ethnic composition were substantial. This was particularly the case with respect to vocabulary knowledge.The program and class characteristics studied here did little to narrow these gaps. This was partly because, as just noted, the studied characteristics were not linked to differential gains in vocabulary and math skills. But it was also because local programs with higher average parent education and income levels tended to have more of the desirable program and class characteristics.
Findings of significant links between program and class characteristics and improvements in children’s behavior have to be tempered by the realization that the measures of children’s behavior made use of ratings by teachers and parents. Thus, it is possible that the observed relationships were partly due to differences in the rating patterns of different groups of teachers rather than (or as well as) to actual behavioral differences between groups of children.
The failure to find significant links between children’s cognitive gains in Head Start and class-level scores on the ECERS-R Language scale may have to do with the generally good quality of Head Start classrooms and the limited range of variation in classroom quality that FACES found in its national samples of programs and classes (Peisner-Feinberg et al, 2001). The same may be said of the failure to find significant links between children’s improvements in the social-emotional domain and class-level scores on the Caregiver Interaction Scale. Studies encompassing broader ranges of quality of childcare and early education facilities have shown greater variations in classroom quality measures and significant relationships between quality measures and children’s gains (Peisner-Feinberg & Burchinal, 1997; Bryant, Burchinal, Lau, & Sparling, 1994; NICHD Early Child Care Research Network, 2000; Phillips, McCartney, & Scarr, 1987; Whitebook, Howes & Phillips, 1989; but for another failure to find a relationship, see Kontos & Fiene, 1987).

At the same time, the FACES results should make us wary of claims that Head Start could produce dramatically larger achievement gains in children from low-income families simply by raising ECERS scores or other indicators of classroom quality. It may be that good classroom quality is a necessary but not sufficient condition for practically significant gains in specific cognitive or behavioral areas. It may be that further progress depends on discovering and applying instructional approaches that can bolster gains in specific areas. Preliminary findings from randomized intervention studies conducted in Head Start programs in New York state as part of the Head Start Quality Research Consortium studies suggest that children in Head Start can make strikingly larger gains in letter recognition and related skills with appropriate, research-based supplementary curricula (Fischel, Storch, Spira & Stoltz, 2003).

REFERENCES

Bryant, D. M., Burchinal, M., Lau, L. B., & Sparling, J. J. (1994). Family and classroom correlates of Head Start children’s developmental outcomes. Early Childhood Research Quarterly, 9, 289-309.

Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear models: Applications and data analysis methods. Newbury Park, CA: Sage Publications.

Fischel, J., Storch, S. A., Spira, E. G., & Stoltz, B. M. (2003). Enhancing emergent literacy skills in Head Start: First year curriculum evaluation results. Poster presented at Society for Research in Child Development Biennial Meeting,Tampa Bay, Florida, April 26.

Kontos, S., & Fiene, R. (1987). Child care quality, compliance with regulations, and children’s development: The Pennyslvania Study. In D. Phillips (ed.), Quality in child care: What does the research tell us? (pp. 57-79). Washington, DC: National Association for the Education of Young Children.

NICHD Early Child Care Research Network (ECCRN). (2000). The relation of child care to cognitive and language development. Child Development, 71, 823-839.

Peisner-Feinberg, E. S. & Burchinal, M. R. (1997). Relations between preschool children’s child-care experiences and concurrent development: The Cost, Quality and Outcomes Study. Merrill-Palmer Quarterly. 43, 3, 451-477.

Peisner-Feinberg, E. S., Burchinal, M. R., Clifford, R. M., Culkin, M. L, Howes, C., Kagan, S. L., & Yazejian, N. (2001). The relation of preschool child-care quality to children’s cognitive and social developmental trajectories through second grade. Child Development, 72(5), 1534-1553.

Phillips, D., McCartney, K., & Scarr, S. (1987). Child care quality and children’s social development.
Developmental Psychology, 23, 537-543.

Phillips, D., Mekos, D., Scarr, S., McCartney, K. & Abbott-Shim, M. (2000). Within and beyond the classroom door:Assessing quality in child care centers. Early Childhood Research Quarterly, 15 (4), 475-496.

Phillips, M., Brooks-Gunn, J., Duncan, G. J., Klebanov, P., & Crane, J. (1998). Family background, parenting practices, and the black-white test score gap. Pp. 103-145 in The Black-White Test Score Gap, edited by C. Jencks & M. Phillips, Washington DC: Brookings Institution Press.

Singer, J. (1998). Using SAS PROC MIXED to fit multilevel models, hierarchical models, and individual growth models. Journal of Educational and Behavioral Statistics, 24(4), 323-355.

Whitebook, M., Howes, C., & Phillips, D. (1989). Who Cares? Child care teachers and the quality of care in America (Final report of the National Child Care Staffing Study). Oakland, CA: Child Care Employee Project.

**Table 5.1. Three-Level Regression Models of Assessment Scale Scores of Head Start Children in Fall and Spring of Program Year, and Fall-Spring Gains, 2000-2001**
	Unstandardized Regression Coefficients
	Fall Letter ID	Spring Letter ID	Letter ID Gains
Program-Level Predictor Variables
High Scope Curriculum	1.01	4.48 *	3.66 *
Creative Curriculum	0.88	2.66 †	1.92
Mean Teacher Salary Level	1.21 †	3.10 **	1.96 **
Proportion Non-Minority Children	-0.07	-0.38	-0.22
Program Mean Parent Education Level	3.55 **	4.76 *	1.20
Program Mean Family Income Level	-0.67	0.02	0.30
Proportion Language-Minority Children	5.20	3.07	-2.06
Class-Level Predictor Variables
Full-Day Class	0.25	1.99	1.81 †
Average ECERS Language Score	-0.49	-0.92	-0.41
Average Child-Adult Ratio	0.34	0.91 **	0.57 *
AP Individualizing Score	0.68 †	0.37	-0.27
Average Lead Teacher Arnett Score	0.03	0.02	-0.01
Teacher BA or AA	1.93 *	0.19	-1.71 †
Years Teaching Experience	-0.02	0.00	0.02
Teacher DAP Beliefs Score	-0.36	-0.76 *	-0.40
Black Teacher	-2.50 *	-0.16	2.33 †
Hispanic Teacher	-0.13	-0.41	-0.36
Teacher Salary Deviation Score	-0.15	-1.55	-1.39 †
Class Parent Education Level deviation	1.62 *	2.72 *	1.09
Class Family Income Level deviation	-0.73	-0.42	0.32
Class Proportion Non-minority deviation	1.86	5.23	3.54
Proportion Language-Minority deviation	-1.59	2.34	3.84
Child-Level Predictor Variables
Parent Literacy Standard Score KFAST	0.10 **	0.06	-0.04
Parent Education Level deviation	0.98 **	1.39 **	0.39
Family Income Level deviation	0.39	0.66	0.26
Welfare Status	-1.40	-1.51	-0.08
Books In Home	1.50	0.79	-0.72
Frequency of Reading to Child
Not At All	-0.55	-3.74 †	-3.23 †
Once or twice	-0.68	-2.95 *	-2.30 *
Every day	0.47	0.65	0.22
Age of Child in Months	0.51 ***	0.68 ***	0.17 †
Sex of Child	0.46	1.24	0.79
Black Child	1.77	1.97	0.29
Hispanic Child	-0.55	-1.92	-1.51
Language Minority Family	1.07	-1.49	-2.45
Disability Status	-2.12 *	-4.07 **	-2.00 †
Mother-Father Family	-1.07	-0.49	0.64
Neither Birth Parent In Home Continued	2.13	-1.55	-3.69 †
Intercept	301.22 ***	297.66 ***	-2.90
Proportion of Variance Accounted For:
Between-Programs Variance	100% ***	95% ***	100% **
Between-Classes Variance	100% ***	92% ***	99% **
Within-Classes Variance	7% ***	6% ***	1% **
Total Variance	15% **	20% ***	12% **

***p < .001
**p < .01
*p < .05
† p < .10
N = 957

**Table 5.2. Three-Level Regression Models of Assessment Scale Scores of Head Start Children in Fall and Spring of Program year, and Fall-Spring Gains, 2000-2001**
	Unstandardized Regression Coefficients
	Fall Vocabulary	Spring Vocabulary	Vocabulary Gains
Program-Level Predictor Variables
High Scope Curriculum	0.38	1.26	0.91
Creative Curriculum	0.88	0.71	-0.15
Mean Teacher Salary Level	0.57	0.34	-0.17
Proportion Non-Minority Children	4.65 ***	5.77 ***	1.25
Program Mean Parent Education Level	2.22 *	2.59 **	0.31
Program Mean Family Income Level	0.39	-1.14	-1.61
Proportion Language-Minority Children	-1.11	2.16	3.01 †
Class-Level Predictor Variables
Full-Day Class	-0.41	-0.36	0.10
Average ECERS Language Score	0.06	-0.13	-0.24
Average Child-Adult Ratio	0.20	0.20	0.00
AP Individualizing Score	0.17	0.06	-0.04
Average Lead Teacher Arnett Score	-0.09	0.03	0.01
Teacher BA or AA	0.07	-0.01	0.00
Years Teaching Experience	-0.04	-0.05 †	-0.02
Teacher DAP Beliefs Score	-0.05	-0.02	0.02
Black Teacher	-0.28	-0.07	0.04
Hispanic Teacher	0.55	0.73	0.36
Teacher Salary Deviation Score	-0.25	-0.24	0.04
Class Parent Education Level (deviation)	1.64 ***	1.61 ***	0.02
Class Family Income Level (deviation)	-1.08	-0.51	0.44
Class Proportion Non-minority (deviation)	3.25 *	3.87 **	0.69
Proportion Language-Minority (deviation)	-0.59	-2.46	-2.38
Child-Level Predictor Variables
Parent Literacy Standard Score KFAST	0.13 ***	0.12 ***	-0.02
Parent Education Level deviation	0.64 ***	0.65 ***	0.01
Family Income Level deviation	0.20	0.05	-0.13
Welfare Status	-0.66	-0.92 †	-0.21
Books In Home	1.63 **	1.26 **	-0.31
Frequency of Reading to Child
Not At All	-1.93 *	-1.83 *	0.12
Once or twice	-1.83 ***	-1.35 **	0.44
Every day	0.57	1.17 **	0.68 †
Age of Child in Months	0.81 ***	0.73 ***	-0.08 **
Sex of Child	-0.23	0.06	0.32
Black Child	-3.31 ***	-3.73 ***	-0.33
Hispanic Child	-1.97 ***	-2.62 ***	-0.64
Language Minority Family	-8.51 ***	-7.19 ***	1.32 †
Disability Status	-2.47 ***	-2.59 ***	-0.13
Mother-Father Family	-0.65	-0.83 *	-0.17
Neither Birth Parent In Home	2.05 *	0.46	-1.52 †
Intercept	0.55	13.65 **	13.23 **
Proportion of Variance Accounted For:
Between-Programs Variance	99% ***	97% ***	56% *
Between-Classes Variance	72% ***	80% ***	21%
Within-Classes Variance	34% ***	33% ***	1% *
Total Variance	53% ***	53% ***	3% †

***p < .001
**p < .01
*p < .05
† p < .10
N =1, 984

**Table 5.3. Three-Level Regression Models of Behavior Rating Scores of Head Start Children in Fall and Spring of Program year, and Fall-Spring Gains, 2000-2001**
	Unstandardized Regression Coefficients
	Fall Cooperative Behavior	Spring Cooperative Behavior	Cooperative Behavior Gains
Program-Level Predictor Variables
High Scope Curriculum	-0.01	1.30 *	1.26 *
Creative Curriculum	-0.53	0.06	0.57
Mean Teacher Salary Level	-0.58 †	0.59 *	1.18 ***
Proportion Non-Minority Children	-1.00	-0.47	0.45
Program Mean Parent Education Level	0.04	-0.20	-0.26
Program Mean Family Income Level	-0.80	-1.79 *	-0.92
Proportion Language-Minority Children	-0.08	0.37	0.31
Class-Level Predictor Variables
Full-Day Class	-0.29	0.13	0.45
Average ECERS Language Score	-0.06	0.04	0.06
Average Child-Adult Ratio	0.03	0.17 *	0.15 †
AP Individualizing Score	-0.12	-0.35 *	-0.22
Average Lead Teacher Arnett Score	0.03	0.03 †	0.00
Teacher BA or AA	-0.10	-0.53	-0.43
Years Teaching Experience	-0.03	-0.02	0.00
Teacher DAP Beliefs Score	-0.06	0.01	0.07
Black Teacher	-0.73	-0.96 †	-0.33
Hispanic Teacher	-0.06	-0.39	-0.28
Teacher Salary Deviation Score	0.23	-0.02	-0.24
Class Parent Education Level (deviation)	-0.44	-0.40	0.07
Class Family Income Level (deviation)	0.62	0.10	-0.48
Class Proportion Non-minority (deviation)	1.03	1.16	-0.02
Proportion Language-Minority (deviation)	0.34	-0.47	-0.79
Child-Level Predictor Variables
Parent Literacy Standard Score KFAST	0.00	0.01 *	0.01
Parent Education Level deviation	0.14 *	0.13 †	-0.01
Family Income Level deviation	0.22 *	-0.01	-0.23 *
Welfare Status	-0.27	-0.77 ***	-0.50 *
Books In Home	0.40 †	0.07	-0.34
Frequency of Reading to Child
Not At All	-0.08	0.22	0.32
Once or twice	-0.13	-0.14	-0.01
Every day	0.44 *	0.26	-0.17
Age of Child in Months	0.22 ***	0.17 ***	-0.05 ***
Sex of Child	1.31 ***	1.48 ***	0.18
Black Child	-0.02	-0.26	-0.24
Hispanic Child	0.23	-0.18	-0.42
Language Minority Family	0.08	0.50	0.40
Disability Status	-1.46 ***	-0.99 ***	0.47 †
Mother-Father Family	-0.11	0.38 *	0.49 *
Neither Birth Parent In Home	-0.54	0.07	0.60
Intercept	5.00	6.79 *	1.66
Proportion of Variance Accounted For:
Between-Programs Variance	77%	100%	65% *
Between-Classes Variance	11%	22% *	8%
Within-Classes Variance	14% ***	12% **	2% **
Total Variance	16% ***	17% ***	7% ***

***p < .001
**p < .01
*p < .05
† p < .10
N =2,138

**Table 5.4. Three-Level Models of Behavior Rating Scores of Head Start Children in Fall and Spring of Program Year, and Fall-Spring Gains, 2000-2001**
	Unstandardized Regression Coefficients
	Fall Hyperactive Behavior	Spring Hyperactive Behavior	Hyperactive Behavior Declines
Program-Level Predictor Variables
High Scope Curriculum	0.08	-0.23	-0.32 *
Creative Curriculum	0.08	-0.07	-0.15
Mean Teacher Salary Level	0.08	-0.11	-0.18 *
Proportion Non-Minority Children	-0.31	-0.09	0.24
Program Mean Parent Education Level	0.12	-0.33 *	-0.45 **
Program Mean Family Income Level	0.00	0.47 †	0.47 *
Proportion Language-Minority Children	0.25	0.29	0.06
Class-Level Predictor Variables
Full-Day Class	0.20 †	0.14	-0.04
Average ECERS Language Score	0.00	0.06	0.06
Average Child-Adult Ratio	-0.02	-0.06 *	-0.04
AP Individualizing Score	0.09 †	0.06	-0.03
Average Lead Teacher Arnett Score	-0.01 †	-0.01	0.00
Teacher BA or AA	-0.12	0.04	0.17 †
Years Teaching Experience	0.01 †	0.01	0.00
Teacher DAP Beliefs Score	-0.02	0.01	0.03
Black Teacher	-0.24	-0.10	0.14
Hispanic Teacher	-0.21	0.00	0.21
Teacher Salary Deviation Score	0.04	0.12	0.09
Class Parent Education Level deviation	0.14	0.17 †	0.03
Class Family Income Level deviation	-0.03	-0.03	0.00
Class Proportion Non-minority deviation	-0.28	-0.16	0.13
Proportion Language-Minority deviation	0.39	0.18	-0.21
Child-Level Predictor Variables
Parent Literacy Standard Score KFAST	0.00	0.00	0.00
Parent Education Level deviation	-0.06 *	-0.06 *	0.00
Family Income Level deviation	-0.06 †	-0.01	0.05
Welfare Status	0.01	0.04	0.03
Books In Home	-0.17 *	-0.04	0.14 †
Not At All	-0.18	-0.01	0.17
Once or twice	-0.11	0.11	0.22 **
Every day	-0.19 *	-0.10	0.10
Age of Child in Months	-0.05 ***	-0.03 ***	0.02 ***
Sex of Child	-0.41 ***	-0.53 ***	-0.13 *
Black Child	-0.21 †	-0.05	0.17
Hispanic Child	-0.12	-0.07	0.07
Language Minority Family	-0.09	-0.20	-0.11
Disability Status	0.62 ***	0.35 ***	-0.26 **
Mother-Father Family	-0.07	-0.27 ***	-0.19 **
Neither Birth Parent In Home	0.02	0.09	0.07
Intercept	4.38 ***	3.84 ***	-0.61
Proportion of Variance Accounted For:
Between-Programs Variance	0%	100%	100% *
Between-Classes Variance	22%	16%	11%
Within-Classes Variance	10% ***	9% ***	3% ***
Total Variance	11% ***	12% ***	6% **

***p < .001
**p < .01
*p < .05
† p < .10
N =2,045

21In the multilevel regression modeling, assessment scores were converted to "W-ability scores," based on IRT scaling of item difficulties carried out by the test developers. These scale scores are purported to have equal-interval properties that are desirable in regression modeling, particularly of gain scores. In other analyses, standard score versions of the assessment scores were used. These scores show how Head Start children performed compared to national norms. But they do not have as strong equal-interval properties as the W-ability scores.(back)

Table of Contents | Previous | Next