Abstract
This study used ECLS-K 1998–1999 data to evaluate whether specific kindergarten teaching practices predicted school-year learning gains differently, depending on children’s ethnicity, SES, and fall test scores. Exploratory factor analyses guided the creation of four literacy and five math instruction composites from teachers’ reports of their teaching practices. The composites differentiated teaching activities that were child-centered versus skills-based and those that targeted basic versus advanced skills. Additionally, they served as predictors of year-end skill level in three-level hierarchical linear models. Literacy instruction composites showed no differential benefits. More frequent exposure to Applied Math (e.g., use math for word problems, use measuring instruments) was associated with higher average spring math scores among children who entered kindergarten with relatively high math skills but not among children who entered kindergarten with relatively low math skills.
There is no shortage of opinions about effective instruction for young children. The traditional debate in early childhood pits teacher-directed instruction focused on basic skills against instruction that involves children in more open-ended, child-initiated activities. Embedded in this conversation are claims made about differential benefits of particular practices for children from different socioeconomic or ethnic groups. The present study uses a nationally representative data set to examine evidence on associations between different math and literacy instructional practices and children’s academic skill development according to their ethnicity, socioeconomic status (SES), and initial skill level.
Lisa Delpit (1995) claims that White, middle-class children’s home cultures match the instructional practices of most schools and that this match gives them an advantage over low-income children and children of color. She explains that White, middle-class children come to school with a better understanding of what is expected of them, both academically and in terms of their behavior. They are therefore in a better position than their minority and low-income peers to benefit from instruction that allows students a fair amount of discretion and independence, or, to employ a term used in the literature, “child-centered instruction.” She describes how, as a public school teacher, the White children in her classroom made substantial academic strides when she used child-centered instruction. The progress of her Black students was slow in comparison, but it increased as she gradually shifted toward more direct, skills-focused instruction. Although Delpit does not advocate teachers’ exclusive use of skills-based instruction for minority children and children from low-income families, she suggests that child-centered instruction with little explicit teaching of basic academic skills will disproportionately benefit White, middle-class children while leaving their less affluent and minority peers behind, thus exacerbating the achievement gap.
Delpit’s view has many adherents. Without endorsing such an approach or referring explicitly to race and SES, Means and Knapp (1991) note that “the most widely accepted prescriptions for compensatory education sought to remedy the students’ deficiencies by teaching ‘the basics’ through curricula organized around discrete skills taught in a linear sequence” (p. 282). Consistent with this view, a few mathematics instruction researchers have warned that some “reform-minded” approaches, such as situating math in real-world contexts or failing to provide explicit teaching, may “unwittingly create, reproduce, or extend inequities among students” (Ball, Goffney, & Bass, 2005, p. 3).
Delpit (1995) bases her claims primarily on personal observations. But studies of parent interactions with young children support her contention that children have differential exposure to child-centered teaching. Extant research indicates, for example, that relatively high SES and White parents tend to be more child-centered (e.g., ask more open-ended questions, give more nonspecific suggestions) and less didactic when engaged in teaching interactions with their children than relatively low-SES parents and parents of color (Bee, Van Egeren, Pytkowicz Streissguth, Nyman, & Leckie, 1969; Diaz, Neal, & Vachio, 1991; Fuligni & Brooks-Gunn, 2013; Laosa, 1980; Stipek, Milburn, Clements, & Daniels, 1992). Children’s home experience may explain Lubienski’s (2000) finding that while the middle-SES students in her classroom tended to prefer an open-ended, problem-solving curriculum, students from low-SES families were more likely to prefer traditional, textbook math instruction, with teachers playing a more directive role.
Another rationale some authors have given for promoting skills-based instruction with low-income or minority children is that it provides a better fit, not with their culture but with their incoming skill level. They claim that economically disadvantaged children should master basic skills before moving on to more advanced skills (see Knapp, Shields, & Turnbull, 1992). A related argument is that skills-based instruction exposes children to large amounts of information in less time than is required for children to achieve mastery when given more discretion and opportunities to explore (Engelmann, 1999). Thus children who on average begin school academically behind (who are disproportionately from low-SES families and children of color) should benefit more from skills-based instruction than from child-centered instruction.
There are voices on the other side of this issue as well. Haberman (1991), for example, refers to the directed teaching approaches common in schools serving children of color and children from low-income families as “the pedagogy of poverty,” which he contrasts with “good instruction” that is more active, personally meaningful, and child-initiated. He claims that the prevalence of the traditional pedagogy-of-poverty approaches is in part the reason why urban students do not learn as well as more privileged students.
Although previous research has examined the association between various instructional approaches and learning, differential benefits of particular approaches to young children by SES status and ethnicity have received little attention, and only a few studies have assessed initial skill level as a moderator. The present study employs data from the Early Childhood Longitudinal Study, Kindergarten Class of 1998–1999 (ECLS-K). These data offer the advantage of a very large, nationally representative sample with information on both literacy and math instruction. The central question of the study is whether different kinds of instruction benefit kindergarten children more or less according to their demographic backgrounds. Of particular interest is whether children of color and children from low-SES families benefit relatively more from instruction focused on basic skills, and whether White children and children from more affluent families benefit relatively more from child-centered instruction.
We review below extant approaches to categorizing instructional practices that informed the strategies used in the present study for grouping the specific practices in the ECLS-K 1998–1999 data. We then summarize the scant evidence that exists on differential benefits of specific instructional approaches according to children’s ethnicity, SES, or prior skill level as well as evidence on differences in the kinds of instruction children from different ethnic and SES backgrounds receive.
Mirroring the debate on direct versus child-centered instruction in the early childhood literature, subject-matter researchers often distinguish between instruction that is focused on conceptual understanding and gives children some discretion related to learning activities and instruction that is largely teacher-directed and focused on discrete skills. Although many terms have been used (e.g., constructivist, scaffolded, reform-minded, progressive, child-directed versus didactic, traditional, conventional, teacher-directed; Early et al., 2010; Knapp et al., 1992; Stipek, 2004), heretofore we use the terms “child-centered” for the former and “skills-based” for the latter instructional approach. At the risk of oversimplifying the various versions of this distinction, typically in descriptions of child-centered instruction, students play an active role in personally meaningful learning activities; the focus is on developing underlying understanding more than on discrete skills, and teachers do not adhere to the belief that skills must be learned in a set order (Stipek et al., 1992). In contrast, in descriptions of skills-based instruction, the teacher plays a central, directive role, and instruction focuses on developing discrete academic skills in a predetermined order (Dahl & Freppon, 1995). In the case of literacy for young children, for example, some researchers differentiate between “whole language” instruction (where children engage in independent reading and writing one-on-one with teachers or in small groups) and instruction focused on phonics, a specific skill central to reading. In math, the terms “basic skill” (featuring problems with one correct solution) and “inferential” (emphasizing students’ deductive reasoning and problem solving) have been used (Crosnoe et al., 2010).
Although some scholars have endorsed primarily one or the other approach, the current consensus is for some balance. Reports such as that of the National Mathematics Advisory Panel (2008) warn against promoting one approach of instruction over the other due to the need for more rigorous research. Authors of other national reports recognize that both approaches—child-centered and skills-based—have value (Snow, Burns & Griffin, 1998; Neuman, Copple, & Bredekamp, 2000), and that instruction is too complex to be captured by defining it primarily as either skills-based or child-centered (Kilpatrick, Swafford, & Findell, 2001). Like the instruction itself, teachers cannot typically be classified as either skills-based or child-centered as most teachers use more than one approach, albeit to varying degrees (e.g., Stipek, 2004). What is of interest, therefore, is the relative use of these different teaching strategies.
Despite being a common distinction in the research literature, the two dimensions of child-centered and skills-based are broadly defined and thus limited in identifying the most effective instructional strategies. A few studies have employed more differentiated categorizations of instruction. In some studies categorization is based on a priori criteria. Engel, Claessens, and Finch (2013), for example, used the ECLS-K 1998–1999 data to create categories of math instruction that aligned with proficiency levels of math skills as determined by the math test administered at the beginning of kindergarten. Other researchers have used empirical analyses of specific practices to generate categories of instruction (e.g., Bodovski & Farkas, 2007; Georges, 2009; Guarino, Hamilton, Lockwood, & Rathbun, 2006). Bodovski and Farkas (2007) conducted two separate factor analyses of the ECLS-K 1998–1999 kindergarten math teaching practices—one for math instructional processes (i.e., math teaching activities) and the other for math instructional content (i.e., instruction targeting math skills). The categories from the factor analysis of instructional processes largely reflected the distinction between child-centered (e.g., group/interactive, manipulatives) and skills-based (e.g., traditional) instruction, whereas the factor analysis of instructional content yielded categories that reflected specific content areas of math (e.g., writing numbers, data/approximations), the level of difficulty of the material (e.g., advanced practical math), or both (e.g., basic numbers, advanced counting). Factor analyses by Guarino and colleagues (2006) of the ECLS-K 1998–1999 kindergarten literacy and math data resulted in categories that differentiated child-centered (e.g., student-centered mathematics) and skills-based (e.g., didactic instruction in literacy) practices as well as instruction targeting different skill levels (e.g., advanced numbers and operations in math) and involving different content areas (e.g., comprehension in literacy) among other features. However, the exact criteria for determining the number of factors to retain, which can impact the categories created, were not clear.
An empirical approach is used to categorize instructional practices in the current study because this approach yields groupings of practices that cohere in real classrooms. Unlike the study by Bodovski and Farkas (2007), the current study includes both instructional content and process variables in the same factor analysis, which avoids artificially constraining the categories created. Additionally, the current study uses parallel analysis, an objective and accurate method for determining the number of factors to retain (Zwick & Velicer, 1986).
Hundreds of studies have examined associations between specific curricula or teaching approaches and student learning. Many of these studies have compared the effectiveness of two or more curricula or teaching strategies (Cohen & Hill, 2000; Dahl & Freppon, 1995; Education Consumers Foundation, 2011; Hamilton et al., 2003; Klein, Starkey, Clements, Sarama, & Iyer, 2008; Knapp et al., 1992; Milesi & Gamoran, 2006; National Early Literacy Panel, 2009; Preschool Curriculum Evaluation Research Consortium, 2008; Schweinhart, Weikart, & Larner, 1986; Stebbins, St. Pierre, Proper, Anderson, & Cerva, 1977; Wenglinsky, 2004). But these studies have either focused on a particular population (e.g., mostly children from low-income families), or no analyses were conducted to assess differential associations with academic skill gains as a function of children’s demographic characteristics. It is difficult to assess differential associations looking across studies because they involve different curricula, teaching strategies, student samples, and measures of academic achievement.
Very few studies were found that examined, as the present study does, differential associations between instructional approaches and student learning. One of the studies found, by Hickey, Moore, and Pellegrino (2001), reported that a reform-minded math curriculum produced the same gains in high- and low-SES fifth-grade classrooms in math problem-solving ability, but greater gains in conceptual knowledge and estimation skills were seen in high-SES than in low-SES classrooms. The program was not implemented the same in both contexts, however, which makes it impossible to determine whether the findings are explained by the instructional approach or its implementation. In another study that used NAEP (National Assessment of Educational Progress) data, Wenglinsky (2004) found that the instructional practices that significantly affected the Black-White achievement gap in fourth grade were not the same as the practices that affected the Latino-White achievement gap, but the instructional practices that affected achievement gaps concerned topics (measurement and estimation versus data analysis), not the instructional approach. Moreover, he did not compare the effect of instructional practices on White, Black, and Latino students, and the data were cross-sectional. Also using NAEP cross-sectional data, Lubienski (2006) examined the interaction between specific teacher practices in math and students’ race and SES. Only one interaction was significant: a “non-number curricular” (geometry, measurement, data analysis, algebra functions) emphasis correlated more positively with achievement for higher-SES students than lower-SES students.
One study is particularly relevant to the present study because it used the same nationally representative data set, the ECLS-K 1998–1999. Georges (2009) employed five categories of instructional practices in math: (1) worksheets, textbooks, and chalkboard, (2) activities using manipulatives, (3) activities in collaborative groups, (4) aesthetic activities, and (5) activities that build on analytic and reasoning skills. She found that instruction emphasizing analytic and reasoning activities (e.g., performing simple data collection and graphing) positively predicted math performance on most of the math subtests for students in high-poverty but not low-poverty classrooms. For low-poverty but not for high-poverty classrooms, activities in collaborative groups (e.g., solving problems in small groups or with a partner) had a positive association, and activities involving manipulatives (e.g., using rulers, measuring cups, spoons, or other measuring instruments) had a negative association with most of the math subtest gains.
Georges’s (2009) findings suggest that some math practices may affect students in classrooms with a substantial proportion of children from low-income families differently from students in classrooms serving more affluent children. But separate analyses of the two sets of classrooms did not allow for statistical tests of differences in the associations. She also differentiated family income levels at the classroom rather than at the student level. Other variables associated with classroom-level poverty could account for the differences in student outcomes. In addition to using statistical tests for differential associations, the present study also assesses the associations between different kinds of instruction and children’s learning as a function of their own level of SES, holding constant other differences in classrooms associated with the student population.
In summary, there is scant empirical evidence on whether particular kinds of instruction affect children differently depending on their ethnic background or socioeconomic status. Extant studies do not provide strong or consistent evidence favoring a particular approach, and no study has compared directly the associations between different kinds of instruction and learning across different groups of children.
A few studies have addressed the possibility that initial skill level rather than ethnicity or socioeconomic status should be the basis for teachers emphasizing child-centered versus skills-based instruction. A handful of studies on the teaching of literacy suggest that skills-based instruction favors low-skilled children and that child-centered instruction favors more highly skilled children. For example, Juel and Minden-Cupp (2000) compared reading gains in four first-grade classrooms—two that prominently featured phonics instruction and two that mainly employed spelling and lists of familiar words (i.e., word walls). Children from the relatively low-skill reading groups made substantially greater gains in the more phonics-based classrooms than in the other two classrooms, whereas mid-level readers appeared to benefit more from reading text and writing than from phonics-based instruction. Using ECLS-K 1998–1999 data, Xue and Meisels (2004) categorized literacy instruction based on how frequently teachers reported using phonics (e.g., alphabetizing, matching letters to sounds) and integrated language arts (e.g., using context cues for comprehension, publishing own writing). Children with relatively low levels of initial literacy skills benefited less from integrated language arts instruction (e.g., using context cues for comprehension, publishing own writing) than children with higher initial literacy skills. Connor and colleagues (Connor, Crowe, & Meadows, 2009; Connor, Morrison, & Katch, 2004; Connor, Piasta, et al., 2009) conducted several studies suggesting differential benefits of literacy instruction depending on children’s initial skill level in first grade. For example, one study found that first graders with initially high vocabulary scores benefited more than children with initially low vocabulary scores from learning activities focused on meaning with little involvement from the teacher (e.g., students writing in their journals) and less from phonics activities managed by the teacher (e.g., teacher writing a word on the board for children to sound out together; Connor et al., 2004).
Note that in all three of these studies the difference between the more skills-based and child-centered instruction is in some ways confounded with the level of difficulty. Skills-based instruction involved phonics, which is a very elementary skill. Phonics is contrasted with a focus on spelling whole words (Juel & Minden-Cupp, 2000), using context cues for comprehension and writing (Xue & Meisels, 2004), and writing in journals (Connor et al., 2004)—all requiring more advanced skills. It is possible that the low achievers benefited more from instruction that was targeted appropriately at their low level of skill, not necessarily because it was child-centered. These studies suggest the importance of differentiating the level of instruction (low versus advanced) from the kind of instruction (teacher-directed and focused on discrete skills versus child-centered).
In contrast to literacy, one study in math found benefits of child-centered instruction for relatively low-skilled students. Crosnoe and colleagues (2010) observed the frequency of skills-based (problems with one solution) and of inferential (focusing on deductive reasoning and problem solving) instruction in third- and fifth-grade classrooms. Controlling for children’s maternal education, race, and age, they found in both grades that children with low initial math skills made more progress toward narrowing the achievement gap when they received frequent inference-based instruction than when they received frequent instruction targeting basic skills, provided that they did not have high levels of conflict with their teachers; children with low initial math skills who did have high levels of conflict with their teachers did not appear to reap the same benefits from inference-based instruction.
There are too few studies to draw strong conclusions about the benefits of particular instructional practices on low- versus high-achieving students, and most confound the nature of instruction with the level of difficulty. The different results for literacy and math could stem from differences in the kind of skills required for learning math and literacy as well as the kinds of assessments of learning that were used. Studies that examine both subjects simultaneously could help untangle the associations.
Despite the dearth of studies on student demographic characteristics as moderators of the association between instructional approaches and student learning, evidence suggests that minority, low-income, and low-skilled students receive less child-centered and more skills-based instruction than their White, middle-class peers in preschool (Early et al., 2010; Lee & Ginsburg, 2007), the elementary school grades (Anyon, 1981; Bargagliotti, Guarino, & Mason, 2009; Lubienski, 2006; Rathbun & Hausken, 2003; Smith, Lee, & Newmann, 2001; Stipek, 2004), and middle school (Swanson & Stevenson, 2002; Wenglinsky, 2002). Smith and colleagues (2001) found also that more didactic and less interactive teaching (involving explaining and discussing answers) was seen in classrooms where all children performed below grade level, suggesting that the higher use of more skills-oriented instruction in schools serving predominantly low-income children of color may be based in part on beliefs about the needs of children who are behind academically rather than on their race or social class. These findings that suggest differences in the type of instruction children typically receive underscore the value of assessing whether different instructional approaches actually benefit some children more than others.
Using a nationally representative sample, the present study creates empirically based distinctions in teaching approaches to assess whether specific approaches to teaching in kindergarten benefit students differently depending on their ethnicity, SES, and initial skill level. Empirically based groupings of teaching practices have the benefit of providing information on which teaching practices are implemented with which other teaching practices in a nationally representative sample of teachers. It extends previous studies by examining a variety of approaches to instruction across child background characteristics instead of focusing primarily on schools with a high concentration of low-performing, low SES, and minority students. Although other studies have used the ECLS-K 1998–1999 data to examine instructional approaches in kindergarten (e.g., Georges, 2009; Guarino et al., 2006; Xue & Meisels, 2004), this study offers contributions beyond previous work with the same data set. Unlike the work by Guarino and colleagues (2006), who examined the link between instruction generally and gains in test scores, the current study seeks evidence of differential benefits of instructions based on child characteristics. Additionally, it includes statistical tests that assess differential associations according to students’ own ethnicity and SES rather than that of the class composition, which distinguishes it from research by Georges (2009). Finally, it does so in both math and literacy, giving it a wider scope than that of Xue and Meisels (2004), which focused exclusively on literacy.
Specifically, the study addressed the following questions of interest: (1) Are distinctions between child-centered and skills-based teaching supported by an empirical analysis of how different teaching practices cluster, or do alternative characterizations of teachers’ practices emerge? (2) Does a nationally representative sample of teachers replicate previous findings suggesting that children of color and children from low-income families receive less child-centered and more skills-based instruction? These two questions relate to the primary research question for the current study: (3) Does student race/ethnicity, SES, or fall skill level moderate the association between teaching strategies and students’ spring skill level?
All data for this study come from the ECLS-K, a nationally representative data set giving detailed, longitudinal information on 21,260 children, their parents, their teachers, and their schools, beginning in their kindergarten year (Tourangeau et al., 2001). This study makes use of the first two waves of data collection, the fall and the spring of kindergarten.
The ECLS-K data set includes weight variables. Due to selective oversampling and to higher rates of nonresponse from some groups of participants compared to others, use of a weight is necessary for sample descriptive statistics to be nationally representative (Tourangeau et al., 2001). In this study, child descriptive statistics use the weight BYCW0. Teacher descriptive statistics, exploratory factor analyses of teaching activities, Cronbach’s alphas for the groups of teaching activities, and correlations of teacher-level variables use the weight B2TW0. School descriptive statistics use the weight S2SAQW0. Although the full sample has 21,260 children, a portion of those children have zero weights and thus are not included in the reported sample size.
The final sample for use in analyses consisted of 15,393 children taught by 2,741 teachers in 799 schools, averaging about 6 children per teacher (minimum: 1; maximum: 15) and 3 teachers per school (minimum: 1; maximum: 25). Children with missing spring teacher or school ID numbers and children who changed schools or teachers during kindergarten were excluded from the sample. Since this study focuses on statistical interactions involving race/ethnicity, we included only children belonging to a racial or ethnic category with sufficient numbers to detect a statistically significant interaction; these included White, Black, Hispanic, and Asian categories.
Table 1. Weighted Descriptive Statistics for Full and Analytic Sample
| Base | Analytic | T test p value | |
|---|---|---|---|
| Children:a | |||
| Sample size | 17,915 | 15,393 | |
| Female (%) | 48.57 | 48.48 | .83 |
| White (%) | 57.43 | 63.78 | <.001 |
| Black (%) | 15.94 | 16.82 | .003 |
| Hispanic (%) | 18.88 | 16.44 | <.001 |
| Asian (%) | 2.97 | 2.96 | .95 |
| Other ethnicity (%) | 4.79 | – | |
| In full-day program (%) | 56.15 | 56.43 | .49 |
| Non-English household (%) | 12.65 | 8.93 | <.001 |
| Mean SES (SD) | −.02 (.80) | .04 (.79) | <.001 |
| Literacy mean spring IRT score (SD) | 45.97 (14.05) | 46.30 (14.09) | .003 |
| Literacy mean fall IRT score (SD) | 34.97 (10.21) | 35.23 (10.23) | .001 |
| Math mean spring IRT score (SD) | 35.90 (12.05) | 36.63 (12.01) | <.001 |
| Math mean fall IRT score (SD) | 25.70 (9.11) | 26.30 (9.12) | <.001 |
| Teachers:b | |||
| Sample size | 3,305 | 2,741 | |
| Female (%) | 97.84 | 98.23 | .12 |
| Elementary certified (%) | 82.67 | 83.65 | .17 |
| Hispanic (%) | 6.29 | 5.53 | .09 |
| Other ethnicity | 1.81 | 1.02 | <.001 |
| Black (%) | 7.13 | 7.01 | .81 |
| Asian (%) | 2.29 | 2.13 | .56 |
| White (%) | 89.31 | 90.28 | .10 |
| Full-day program (%) | 62.91 | 61.83 | .25 |
| Mean years teaching kindergarten (SD) | 8.19 (7.40) | 8.44 (7.46) | .09 |
| Schools:c | |||
| Sample size | 865 | 799 | |
| Private (%) | 34.89 | 33.55 | .42 |
| % free lunch eligible (SD) | 25.83 (27.81) | 25.98 (27.59) | .89 |
The results show that the analytic sample had significantly higher student fall and spring test scores, student SES level, and percentages of Black and White students and significantly lower percentages of Hispanic students and students from a non-English household than the base sample. At the teacher level, the percentage of teachers of other ethnicity was significantly lower in the analytic sample than in the base sample. The ECLS-K only administered the literacy test to students who proved proficient in English in the pretest (Tourangeau et al., 2001). The exclusion of non-English proficient students from testing contributed to the differences reported between the base and analytic samples. Additionally, some of the differences in the percentage of Black and White students can be attributed to our retention in the analytic sample of students belonging to racial and ethnic categories with sufficiently high numbers to detect a statistical interaction, which shifted the percentages for the categories. Despite the reported differences, the minimum and maximum values of SES in the base (−4.75 to 2.75) and analytic (−4.47 to 2.69) samples are nearly identical, and the minimum and maximum scores on the fall literacy test were identical for the two samples (21.01 to 138.51). Thus, in terms of SES and literacy ability at the beginning of kindergarten, the analytic sample includes a full range of students. Nevertheless, the results of the t tests in Table 1 show that caution is needed when generalizing the results of the current study.
The ECLS-K 1998–1999 used all sources of information on children’s race and ethnicity—teacher, parent, and school questionnaires and interviews—to form a composite variable of race/ethnicity (Tourangeau et al., 2001). Although the ECLS-K distinguishes between Hispanic, race specified, and Hispanic, race not specified, we have collapsed these two categories into one given that no additional information on race is available for children in either category. The result is a collection of dummy-coded race/ethnicity variables (Asian, Hispanic, Black, and White, where 0 = child was not of that race/ethnicity and 1 = child was of that race/ethnicity). Socioeconomic status (SES) is a composite item in the ECLS-K combining household income, maternal/female guardian and paternal/male guardian education, and the prestige scores for mother’s/female guardian’s and father’s/male guardian’s occupation. For most participants, values for these variables came from items from the spring parent interview, but in the case of nonresponse to the relevant items, the ECLS-K data included imputed values using the value given by a parent with similar characteristics or, in the case of income, partial income data (Tourangeau et al., 2001). Child gender, determined by parent questionnaires or, if missing, by data entered in the field by ECLS-K staff, is given by the dummy-coded variable female (male = 0, female = 1). The ECLS-K determined whether or not English was the primary language spoken at home by consulting school records or the child’s teacher; in the data, it is given by a dummy-coded non-English household variable (0 = English spoken at home, 1 = non-English language spoken at home).
Children took tests in literacy, mathematics, and general knowledge; the first two, administered individually both in the spring and in the fall of the kindergarten year, were analyzed for this study (Tourangeau et al., 2001). All participating children completed the same first phase of both the literacy and math tests (Tourangeau et al., 2001). For the second phase, children received one of three levels of difficulty, depending on their performance on the first phase. Children whose primary language was not English completed a language pretest. As mentioned above, only those who proved proficient in English in the pretest took the literacy test; 54% of children whose home language was not English were missing data for one or both of the literacy tests. The analytic sample thus includes only English-proficient children, and results do not generalize to children who are not proficient in English.
The literacy test targeted letter and word recognition, familiarity with print, rhyming sounds, beginning and ending sounds, receptive vocabulary, listening comprehension, and word context. The math test covered number sense and properties, operations and functions, measurement, spatial sense, geometry, knowledge about data (including data collection, probability, and statistics), patterns, and algebra. Both tests were created for use in the ECLS-K. The test scores used in analyses were the IRT (Item Response Theory) scale scores, which take into account the difficulty of each test item and whether children who got a difficult item correct likely did so by guessing. The IRT scale scores for the kindergarten year have reliabilities ranging from .88 to .95 across subject areas tested (Tourangeau et al., 2001). For literacy, they range from 21.01 to 128.27 and from 22.89 to 156.85 in the fall and spring, respectively. The math scores range from 10.51 to 96.04 and from 11.57 to 106.86 in the fall and spring, respectively.
As part of the ECLS-K 1998–1999, teachers completed a questionnaire asking information about themselves and their classrooms, including the number of years they had been teaching kindergarten, whether they taught full-day or half-day classes (dummy coded where 0 = half-day and 1 = full-day), and whether they had elementary certification (dummy coded where 0 = certification and 1 = no certification; Tourangeau et al., 2001).
School administrators completed a questionnaire that asked whether the school was public or private (dummy coded where 0 = public and 1 = private) and the percentage of students eligible for free lunch. Table 1 gives summary statistics on teachers and their classrooms and on schools.
As part of data collection for the ECLS-K data set, kindergarten teachers filled out questionnaires that included items on their teaching activities in literacy, mathematics, social studies, and science (Tourangeau et al., 2001). The present study used data on literacy and mathematics teaching. Responses to two types of questions were used to assess how teachers spent their instructional time. The first asked, “How often do children in this class do each of the following reading and language arts activities?” (Guarino et al., 2006, p. 10). Teachers gave responses for each specific activity ranging from never to daily with intermediate scale responses of once a month or less, two or three times a month, once or twice a week, and three or four times a week. The second type of question was: “For this school year as a whole, please indicate how each of the following reading and language arts skills is taught in your class(es)” (Guarino et al., 2006, p. 10). As with the first type of question, teachers indicated how frequently they taught each particular skill. The scale was identical to the one described above with the exception of the never response; instead teachers specified either taught at a higher grade level or children should already know. For the purposes of the analyses in the present study, we coded both the taught at a higher grade and the children should already know responses as equivalent to never. The mathematics questions used the same format with the same response options as those used for the literacy questions.
We began the analysis by creating groupings of teaching activity variables—referred to henceforth as teaching strategies. The groupings, guided by exploratory factor analysis, addressed the research question of whether an empirical analysis supports the distinction between child-centered and skills-based teaching, and they provided the teaching strategy variables for statistical models addressing other research questions.
In preparation for the building of the statistical models, we computed correlations separately for literacy and for math, using listwise deletion, between teaching strategies and the classroom mean for each teacher of students’ SES and race/ethnicity, partialing out the mean for each teacher of children’s fall test scores in the relevant subject area. Since variables giving children’s race/ethnicity were dummy coded (0 = child did not belong to racial/ethnic category, 1 = child belonged to racial/ethnic category), the teacher-level mean for these variables represented the proportion of children of the given race/ethnicity out of all sampled children for each teacher.
To deal with missing data, prior to building the multilevel models, we performed multiple imputation on the analytic sample. Multiple imputation creates copies of the data set with each missing value in each data set imputed. Analyses performed on multiply imputed data sets give the average of the coefficients over the set of copies of the data set (Rubin, 1987). We included the outcome variables (spring math and literacy scores) in the multiple imputation models but subsequently deleted observations with imputed values on the outcome variables. This method has been found to have less bias than omitting the outcome variables from the multiple imputation models and less noise than keeping the imputed values on the outcome variables in analyses of the multiply imputed data (von Hippel, 2007).
To address our research question regarding the differential effects of instructional approaches on children’s learning, we constructed a series of multilevel models using the multiply imputed data with students (level 1) nested within teachers (level 2) nested within schools (level 3). Data with a nested structure such as the ECLS-K are well-suited for multilevel models to ease concerns about aggregation bias, problems with estimating precision, and issues related to the unit of analysis (Bryk & Raudenbush, 2002). The assumptions for multilevel modeling appeared to be appropriate for the data.
The first step in the multilevel modeling was to build an unconditional model to get an estimate of the amount of variance between children with the same teachers, between teachers in the same school, and between schools. The second step was to include individual child background variables—fall test scores, SES, and dummy-coded variables for Asian, Black, and Hispanic (with White as the reference category)—and covariates in the model. The level 1 covariates were (1) the number of days between the fall and spring assessments, (2) non-English household, and (3) female. For level 2, the covariates were (1) the number of years the teacher had taught kindergarten, (2) whether the teacher was certified, (3) full-day versus half-day program, (4) mean fall test score for each teacher, (5) mean SES for each teacher, and (6) mean race/ethnicity for each teacher. Including the teacher-mean fall test scores, SES, and race/ethnicity allowed us to control for the collective child demographic characteristics and school-entry skill level of children with the same teacher. Level 3 had the covariates (1) private versus public and (2) percentage of students eligible for free lunch.
For the third step, we added the teaching strategies. The fourth step was to pare down the model by removing nonsignificant covariates. The final step assessed interactions between individual-level fall test scores, race/ethnicity, or SES and teacher-level teaching strategies. In the interest of space, results from steps 2, 3, and 4 are presented in tables, and only when interaction terms were statistically significant are models from step 5 presented.
All continuous variables are standardized—centered at the grand mean and in units of standard deviation to give effect sizes. Additionally, since the multilevel models were built in Stata, which allows for weights at both the highest and lowest levels of the model, the child-level weight BYCW0 and the school-level weight S2SAQW0 were used in the models. To produce unbiased estimates, Stata requires level 1 weights to be conditional on the highest level weights. Thus, we created a new weight for use at level 1 by dividing the child-level weight BYCW0 by the school-level weight S2SAQW0 (see the Stata documentation for the command xtmixed for more information on Stata’s handling of weights in multilevel models).
To determine which teaching practices clustered together, we performed separate exploratory factor analyses with maximum likelihood estimation and varimax rotation on all 42 literacy and on all 44 math teaching activity variables, dropping teachers with missing values on any of the variables. We ran parallel analysis (Horn, 1965) to find the appropriate number of factors to retain, resulting in five factors for literacy and eight for math. Evidence suggests that parallel analysis is one of the most accurate methods for determining the number of factors to retain (Zwick & Velicer, 1986).
Table 2. Standardized Alphas and Factor Loadings for Literacy Teaching Activities (N = 2,098)
| Alphac | Group Name | Teaching Activities in Group | I | II | III | IV | V |
|---|---|---|---|---|---|---|---|
| .82 | Child-Centered Literacy | Write with encouragement to use invented spellings | .72 | ||||
| Write stories or reports | .68 | ||||||
| Write in journal | .67 | ||||||
| Publish children’s own writing | .56 | ||||||
| Read books children have chosen themselves | .51 | ||||||
| Dictate stories to teacher | .50 | ||||||
| Do a project related to book | .43 | ||||||
| Retell stories | .40 | ||||||
| Perform plays or skitsa | .38 | ||||||
| Listen to teacher read stories while seeing the printa | .35 | ||||||
| Read silentlya | .30 | ||||||
| Peer tutoringa | .28 | ||||||
| Mixed group literacy worka | .27 | ||||||
| .77 | Comprehension | Make predictions based on text | .70 | ||||
| Identify main idea of story | .64 | ||||||
| Use cues for comprehension | .64 | ||||||
| Communicate ideas orally | .47 | ||||||
| Common prepositionsb | .40 | ||||||
| Follow complex directionsa | .39 | ||||||
| New vocabularya | .32 | ||||||
| Rhyming words and word familiesa | .31 | ||||||
| .79 | Skills-Based Advanced Literacy | Reading aloud fluently | .63 | ||||
| Conventional spelling | .53 | ||||||
| Use basal reading textsb | .52 | ||||||
| Alphabetizing | .47 | ||||||
| Write from dictation | .46 | ||||||
| Reading multisyllable words | .44 | ||||||
| Vocabulary | .42 | ||||||
| Write stories with a beginning, middle, and end | .40 | ||||||
| Read alouda | .36 | ||||||
| Use workbooks or worksheetsa | .34 | ||||||
| .74 | Skills-Based Basic Literacy | Matching letters to sounds | .71 | ||||
| Alphabet and letter recognition | .68 | ||||||
| Work on phonics | .54 | ||||||
| Work on letter names | .53 | ||||||
| Writing alphabet | .50 | ||||||
| Writing own namea | .37 | ||||||
| Conventions of printa | .32 | ||||||
| .80 | Advanced Writing | Compose or write complete sentences | .72 | ||||
| Use capitalization and punctuation | .59 |
Table 3. Standardized Alphas and Factor Loadings for Math Teaching Activities (N = 2,054)
| Alphab | Group Name | Teaching Activities in Group | I | II | III | IV | V | VI | VII | VIII |
|---|---|---|---|---|---|---|---|---|---|---|
| .81 | Basic Math Skills | Ordering objects | .83 | |||||||
| Sort into subgroups using rule | .80 | |||||||||
| Name geometric shapes | .52 | |||||||||
| Identify relative quantity | .49 | |||||||||
| Making or copying patterns | .46 | |||||||||
| Write numbers one to tena | .33 | |||||||||
| .80 | Child-Centered Math | Solve math with a partner | .58 | |||||||
| Counting manipulatives | .58 | |||||||||
| Play math-related games | .53 | |||||||||
| Mixed group math work | .50 | |||||||||
| Use geometric manipulatives | .45 | |||||||||
| Peer tutoring | .45 | |||||||||
| Solve real-life math | .44 | |||||||||
| Explain how to solve math | .43 | |||||||||
| Relation between number and quantitya | .37 | |||||||||
| Count out louda | .28 | |||||||||
| .75 | Advanced Number Skills | Reading three-digit numbers | .75 | |||||||
| Counting beyond 100 | .58 | |||||||||
| Place value | .57 | |||||||||
| Reading two-digit numbers | .56 | |||||||||
| Counting by 2s, 5s, and 10s | .48 | |||||||||
| Write all numbers 1–100a | .29 | |||||||||
| .74 | Applied Math | Use measuring instruments accurately | .65 | |||||||
| Recognizing fractions | .52 | |||||||||
| Telling time | .48 | |||||||||
| Use measuring instruments | .43 | |||||||||
| Estimating quantities | .40 | |||||||||
| Estimating probabilitya | .33 | |||||||||
| Know value of coins and casha | .32 | |||||||||
| Use math for word problemsa | .27 | |||||||||
| .81 | Number Representations | Reading simple graphs | .73 | |||||||
| Simple data collection/graphing | .71 | |||||||||
| Recognizing ordinal numbersa | .30 | |||||||||
| .61 | Skills-Based Math | Do math on chalkboard | .52 | |||||||
| Use math textbooks | .50 | |||||||||
| Do math worksheets | .48 | |||||||||
| .88 | Operations | Add single-digit numbers | .83 | |||||||
| Subtract single-digit numbers | .66 | |||||||||
| .76 | Math through Music and Movement | Use movement to learn math | .80 | |||||||
| Use music to learn math | .70 |
The exploratory factor analysis revealed that several groupings could be characterized as either child-centered or skills-based. As their names suggest, Child-Centered Literacy and Child-Centered Math both featured largely child-centered teaching activities. Skills-Based Basic Literacy, Skills-Based Advanced Literacy, and Skills-Based Math all had teaching activities that were largely skills-based in nature.
Another, somewhat orthogonal, distinction emerged as well—whether teaching activities targeted low-level or high-level skills. Skills-Based Basic Literacy and Basic Math Skills consist of teaching activities targeting mainly low-level skills. Skills-Based Advanced Literacy and Advanced Number Skills feature activities focusing on high-level skills.
Two teaching strategies—Comprehension and Applied Math—did not conform to either the distinction between skills-based and child-centered instruction or the distinction between instruction targeting low-level versus high-level skills. Because the practices making up each of these two categories formed reliable factors, in keeping with our goal of studying practices that are grouped in real classrooms, we retained both the Comprehension and Applied Math teaching strategies.
Table 4. Partial Correlations for Teacher-Level Variables, Controlling for Teacher-Mean Fall Test Scores
| Mean Spring Score | Mean SES | Prop. Asian | Prop. Black | Prop. Hispanic | Child-Centered Lit./Basic Math Skills | Comp./Child-Cent. Math | Skills-Based Adv. Lit./Adv. Number Skills | Applied Math | Skills-Based Math | |
|---|---|---|---|---|---|---|---|---|---|---|
| Spring score | – | .05* | .01 | −.17*** | .06** | .02 | .09*** | .13*** | .10*** | .16*** |
| SES | .03 | – | .01 | −.19*** | −.12*** | −.03 | −.03 | −.02 | .01 | −.15*** |
| Asian | .08** | .02 | – | −.08*** | .01 | −.04 | −.04 | .03 | −.06** | −.02 |
| Black | −.11*** | −.28*** | −.09*** | – | −.27*** | .02 | .12*** | .04* | .08*** | .11*** |
| Hispanic | .04 | −.17*** | .02 | −.21*** | – | .05* | .06** | .07** | .03 | .01 |
| Child-Centered Lit./Basic Math Skills | .11*** | −.06** | .02 | .06** | .07** | – | .52*** | .27*** | .49*** | .11*** |
| Comprehension/Child-Cent. Math | .06** | −.10*** | .02 | .07** | .06** | .52*** | – | .30*** | .52*** | .15*** |
| Skills-Based Adv. Lit./Adv. Number Skills | .18*** | −.13*** | .02 | .19*** | .14*** | .45*** | .43*** | – | .33*** | .05* |
| Skills-Based Basic Lit./Applied Math | .07** | −.09*** | −.01 | .07** | .02 | .20*** | .21*** | .19*** | – | .19*** |
The significant and small negative correlations between mean SES and all four literacy teaching strategies—Child-Centered Literacy, Comprehension, Skills-Based Advanced Literacy, and Skills-Based Basic Literacy—signal that teachers implemented all of these strategies more frequently if they taught relatively more students from low-SES families, holding constant fall literacy scores. The only significant correlation between a math teaching strategy—Skills-Based Math Instruction—and mean SES was negative, though this correlation was small. Controlling for fall math scores, teachers with relatively high-SES students tended to use Skills-Based Math less frequently than those with low-SES students. The remaining associations between children’s level of SES and the literacy and math teaching strategies were not statistically significant.
All of the literacy teaching strategies correlated modestly but significantly with the proportion of Black students. Child-Centered Literacy, Comprehension, and Skills-Based Advanced Literacy all correlated positively with the proportion of Hispanic students, though in all cases these correlations were also small. None of the correlations between literacy teaching strategies and the proportion of Asian students were statistically significant. In summary, teachers with higher proportions of Black and Hispanic students in the sample tended to use Child-Centered Literacy, Comprehension, and Skills-Based Advanced Literacy more frequently. Teachers with higher proportions of participating Black students in their classroom employed Skills-Based Basic Literacy more frequently. All of these associations are net of fall literacy scores.
In math, Basic Math Skills had a small positive correlation with the proportion of Hispanic students. Both proportions of Black and Hispanic students were significantly correlated with Child-Centered Math and Advanced Number Skills. These correlations were positive and small. Additionally, the correlations between the proportion of Black students and both Applied Math and Skills-Based Math were positive and small. The one statistically significant correlation between a math teaching strategy—Applied Math—and the proportion of Asian students was negative and small. These results show that teachers with higher proportions of Hispanic students generally used Basic Math Skills more frequently, and teachers with higher proportions of Black students generally used both Applied Math and Skills-Based Math more frequently. Teachers with high proportions of Black and Hispanic students used more frequent Child-Centered Math and Advanced Number Skills. Finally, those with higher proportions of Asian students tended to employ Applied Math less frequently. Fall math scores are held constant in these associations.
The unconditional model for literacy yielded an intraclass correlation coefficient (ICC) for teachers of .51 and an ICC for schools of .32. The variance components for spring literacy scores revealed that 49% of the variance was between children with the same teacher, 19% was between children with different teachers in the same school, and 32% was between schools. For math, the ICCs were .33 and .51 for schools and teachers respectively, based on the unconditional model. The variance components showed that 49% of the variance in spring math scores was between students with the same teacher, 19% was between children with different teachers in the same school, and 33% was between schools.
For both literacy and math, the statistically significant level 1 covariates retained in the model were female and the number of days between the fall and spring tests. For math, the level 2 covariates uncertified, teacher-mean Asian, teacher-mean Black, and teacher-mean Hispanic were also retained. Although teacher-mean Asian and teacher-mean Hispanic were not on their own statistically significant in the math models, a linear hypothesis F test of the null hypothesis that the coefficients for teacher-mean race/ethnicity variables were jointly 0 rejected (p < .05) and thus indicated that the collection of teacher-mean race/ethnicity variables was statistically significant and should be retained in the model.
Table 5. Literacy Multilevel Models Predicting Spring Literacy Scores
| Model 1 | Model 2 | Model 3 | Model 4 | |||||
|---|---|---|---|---|---|---|---|---|
| Coefficient | SE | Coefficient | SE | Coefficient | SE | Coefficient | SE | |
| Intercept | −.116 | .095 | −.103 | .096 | −.006 | .011 | −.006 | .011 |
| Level 1 variables (N = 13,605): | ||||||||
| Fall lit. score | .796*** | .013 | .796*** | .013 | .798*** | .012 | .798*** | .012 |
| SES | .037*** | .006 | .037*** | .006 | .037*** | .006 | .037*** | .006 |
| Asian | .090** | .034 | .090** | .034 | .103** | .032 | .103** | .032 |
| Black | −.083*** | .022 | −.083*** | .022 | −.090*** | .020 | −.090*** | .020 |
| Hispanic | −.008 | .021 | −.008 | .021 | .001 | .019 | .001 | .019 |
| Non-English household | .045 | .029 | .045 | .029 | – | – | – | – |
| Days between tests | .103*** | .020 | .103*** | .020 | .102*** | .020 | .102*** | .020 |
| Female | .041*** | .010 | .041*** | .010 | .041*** | .010 | .041*** | .010 |
| Level 2 variables (N = 2,393): | ||||||||
| TM fall lit. score | .042 | .027 | .039 | .027 | – | – | – | – |
| TM SES | .006 | .019 | .007 | .019 | – | – | – | – |
| TM Asian | −.092 | .071 | −.091 | .071 | – | – | – | – |
| TM Black | −.100 | .058 | −.090 | .057 | – | – | – | – |
| TM Hispanic | −.018 | .050 | −.018 | .050 | – | – | – | – |
| Uncertified | −.026 | .026 | −.027 | .026 | – | – | – | – |
| Full-day program | −.030 | .046 | −.039 | .046 | – | – | – | – |
| Years teaching kindergarten | .011 | .007 | .010 | .007 | – | – | – | – |
| Child-Centered Lit. | – | – | .010 | .012 | .012 | .012 | .012 | .012 |
| Comprehension | – | – | .003 | .011 | .005 | .011 | .005 | .011 |
| Skills-Based Adv. Lit. | – | – | .017 | .012 | .017 | .012 | .017 | .012 |
| Skills-Based Basic Lit. | – | – | .004 | .009 | .003 | .009 | .003 | .009 |
| Level 3 variables (N = 798): | ||||||||
| Private | .014 | .031 | .023 | .032 | – | – | – | – |
| % free lunch eligible | .041** | .013 | .039** | .013 | .015 | .011 | – | – |
| Random effects: | ||||||||
| Level 3 intercept | .262 | .012 | .2537 | .011 | .253 | .011 | .253 | .011 |
| Level 2 intercept | .180 | .008 | .179 | .008 | .180 | .008 | .180 | .008 |
| Level 1 residual | .457 | .008 | .457 | .008 | .457 | .008 | .457 | .008 |
Note. For race/ethnicity variables, 0 = not race/ethnicity, 1 = belongs to race/ethnicity; for non-English household, 0 = English household, 1 = non-English household; for female, 0 = male, 1 = female; for uncertified, 0 = certified, 1 = uncertified; for full-day program, 0 = half day, 1 = full day; for private, 0 = public, 1= private; TM = teacher-mean.
**. p < .01.
***. p < .001.
Table 6. Math Multilevel Models Predicting Spring Math Scores
| Model 1 | Model 2 | Model 3 | Model 4 | |||||
|---|---|---|---|---|---|---|---|---|
| Coefficient | SE | Coefficient | SE | Coefficient | SE | Coefficient | SE | |
| Intercept | −.024 | .075 | −.012 | .074 | .066*** | .019 | .066*** | .019 |
| Level 1 variables (N = 13,605): | ||||||||
| Fall math scores | .778*** | .010 | .778*** | .010 | .779*** | .010 | .780*** | .010 |
| SES | .045*** | .008 | .045*** | .008 | .047*** | .007 | .047*** | .007 |
| Asian | .045 | .037 | .045 | .037 | .044 | .035 | .045 | .035 |
| Black | −.119*** | .021 | −.119*** | .021 | −.118*** | .021 | −.117*** | .021 |
| Hispanic | −.056** | .021 | −.056** | .020 | −.056** | .019 | −.054** | .019 |
| Non-English household | −.003 | .024 | −.003 | .024 | – | – | – | – |
| Days between tests | .110*** | .016 | .110*** | .016 | .111*** | .016 | .111*** | .016 |
| Female | −.025* | .011 | −.025* | .011 | −.025* | .011 | −.025* | .011 |
| Level 2 variables (N = 2,393): | ||||||||
| TM fall math scores | .026 | .024 | .025 | .024 | – | – | – | – |
| TM SES | .037 | .025 | .037 | .024 | – | – | – | – |
| TM Asian | −.053 | .077 | −.055 | .077 | −.048 | .076 | −.052 | .076 |
| TM Black | −.108 | .061 | −.098 | .060 | −.130* | .060 | −.130* | .060 |
| TM Hispanic | .095 | .057 | .094 | .056 | .061 | .055 | .059 | .055 |
| Uncertified | −.056* | .025 | −.058* | .025 | −.059* | .025 | −.059* | .025 |
| Full-day program | .003 | .008 | −.005 | .047 | – | – | – | – |
| Years teaching kindergarten | .008 | .048 | .002 | .008 | – | – | – | – |
| Basic Math Skills | – | – | −.015 | .011 | −.015 | .011 | −.015 | .011 |
| Child-Centered Math | – | – | .013 | .012 | .013 | .012 | .013 | .012 |
| Adv. Number Skills | – | – | .009 | .012 | .010 | .011 | .010 | .011 |
| Applied Math | – | – | .020 | .013 | .020 | .013 | .022 | .013 |
| Skills-Based Math | – | – | .022 | .013 | .022 | .013 | .023 | .013 |
| Level 3 variables (N = 798): | ||||||||
| Private | −.010 | .029 | −.017 | .029 | – | – | – | – |
| % free lunch eligible | .016 | .014 | .012 | .015 | – | – | – | – |
| Interaction: | ||||||||
| Applied Math × fall math scores | – | – | – | – | – | – | .017* | .009 |
| Random effects: | ||||||||
| Level 3 intercept | .245 | .012 | .237 | .012 | .238 | .011 | .238 | .011 |
| Level 2 intercept | .181 | .008 | .180 | .008 | .180 | .008 | .180 | .008 |
| Level 1 residual | .479 | .006 | .479 | .006 | .479 | .006 | .479 | .006 |
Note. For race/ethnicity variables, 0 = not race/ethnicity, 1 = belongs to race/ethnicity; for non-English household, 0 = English household, 1 = non-English household; for female, 0 = male, 1 = female; for uncertified, 0 = certified, 1 = uncertified; for full-day program, 0 = half day, 1 = full day; for private, 0 = public, 1= private; TM = teacher-mean.
*. p < .05.
**. p < .01.
***. p < .001.
The main effect of SES was statistically significant and positive in all models predicting spring literacy and math scores (see Tables 5 and 6). Higher levels of SES were associated with higher spring test scores, controlling for all covariates included in the model.
In the models in Tables 5 and 6, we ran linear hypothesis F tests on the null hypothesis that the coefficients for Black, Hispanic, and Asian were jointly 0. Results for both math and literacy were highly significant (p < .001), indicating that there was an overall effect of race/ethnicity and that these variables should be retained in the model. Additionally, we ran linear hypothesis F tests to determine whether spring test scores for Black, Hispanic, and Asian students differed from each other, controlling for all predictors in the models. This step was warranted because the coefficients for these variables in the model only test whether they differ from the reference category, White. For literacy spring scores, Black students scored significantly lower than students from all other racial/ethnic groups (p < .001). Hispanic students’ spring literacy scores were significantly lower than those of Asian (p < .01) but not White students, and White students’ scores were significantly lower than Asian students’ scores (p < .01). For math spring scores, Black students scored significantly lower than students from all other racial/ethnic groups (p < .001), and Hispanic students scored significantly lower than both White (p <. 01) and Asian (p <. 01) students. Spring math scores for Asian and White students did not differ from each other.
The interaction term between Applied Math and children’s fall math test scores when predicting spring math test scores was significant and positive, controlling for children’s race/ethnicity, SES, and gender; the number of days between tests; whether teachers were certified; the teacher-mean level of children’s ethnicity; and all other math teaching strategies. The association between Applied Math and spring math scores is more positive for students who started the school year with strong math skills than for those whose skills were low at the start of the year (see Fig. 1). Statistical tests revealed that the slope for children who started the school year with high math skills, but not for children who started the year with low math skills, reached statistical significance (p < .05). For children with high initial skill level only, the more frequently teachers employed Applied Math, the higher children’s spring math scores were on average.
The central goal of this study was to assess evidence for claims that some groups of children benefit more than others from either child-centered or skills-based instruction—a claim that has been made more often than it has been studied. This study extends previous research, which has primarily examined the effects of instructional approaches within a population (e.g., only students from low-income families), within a teaching strategy (e.g., only child-centered teaching), or within a subject matter (e.g., only literacy) by examining the effects of a broad array of teaching strategies on students with diverse ethnic and economic backgrounds and different initial skill levels.
We begin by reviewing evidence concerning the two preliminary study questions. First, does an empirical analysis of teaching practices reveal clusters of practices that fall neatly into the child-centered or skills-based categories? Second, does a nationally representative sample of teachers replicate previous findings suggesting that children of color and children from low-SES families and with relatively low skills receive less child-centered and more skills-based instruction? We then discuss results pertaining to the primary focus of the study, asking whether race/ethnicity, SES, or fall skill level moderates the association between teaching strategies and students’ spring skill level.
Clusters of instructional approaches that could be characterized as child-centered or skills-based did emerge in the empirical analyses. For both literacy and math, analyses revealed groups of teaching activities that were primarily child-centered and some that were largely skills-based. The Applied Math cluster was not clearly either child-centered or skills-based, although some researchers associate learning in authentic or meaningful contexts with more child-centered approaches.
A second potentially important dimension also emerged—the degree to which practices focused on low-level versus more advanced skills. In literacy, two clusters—Skills-Based Basic Literacy (e.g., alphabet and letter recognition, matching letters to sounds) and Skills-Based Advanced Literacy (e.g., reading multisyllable words, reading aloud fluently)—reflected skill-focused instruction targeting basic and advanced literacy skills respectively. The analyses of math practices also revealed clusters of teaching activities aligned with skill level. Specifically, Basic Math Skills (e.g., count numbers 1–10, name geometric shapes) aligned with low-level skills while Advanced Number Skills (e.g., count by 2s, 5s, and 10s; place value) aligned with more advanced skills.
Differentiating the level of difficulty of instruction may be a particularly important dimension to assess in future research designed to examine associations between instructional practices and learning. But level of instruction should be examined in the context of students’ skill levels. Engel and colleagues (2013) concluded from their analyses of the ECLS-K data that substantial amounts of time were focused on skills that a large proportion of the students had mastered by the beginning of kindergarten. Research that examines the match between the level of skills children are being taught and their current skills would be consistent with the emphasis in learning theory on teaching in children’s “zone of proximal development” (Vygotsky, 1978) and may be as important in future research as examining other qualities of instruction, such as whether it is child-centered versus skills-based.
Teachers did appear to adjust their instructional strategies according to children’s demographic backgrounds, but the findings were mixed. Controlling for fall test scores, some of the findings aligned with and in other cases contradicted previous studies suggesting that students from relatively high-SES families experience more child-centered instruction while students of color and students from relatively low-SES families experience more skills-based instruction (Anyon, 1981; Bargagliotti et al., 2009; Dahl & Freppon, 1995; Early et al., 2010; Rathbun & Hausken, 2003; Smith et al., 2001; Stipek, 2004).
Students in classrooms in which the mean SES was relatively low received more frequent Skills-Based Math, Skills-Based Advanced Literacy, and Skills-Based Basic Literacy, but they also received more frequent Child-Centered Literacy and Comprehension. The findings related to race/ethnicity were equally mixed. For example, although teachers with higher proportions of Black students more frequently employed Skills-Based Basic Literacy and Skills-Based Math, they also employed more frequent Applied Math. Teachers with relatively high proportions of Black and Hispanic students also implemented relatively more frequent Child-Centered Literacy and Child-Centered Math activities as well as engaging more frequently in Skills-Based Advanced Literacy. These results were net of children’s fall test scores and so do not confound SES and race/ethnicity with skill level at the beginning of kindergarten.
The results indicate a complex picture with very little evidence in this nationally representative sample of systematic or widespread bias toward child-centered or skills-based teaching related to children’s backgrounds. Rather, taken on the whole, the results suggest that teachers who had a relatively high proportion of low-SES children and children of color tended to report more varied instruction (i.e., to engage more frequently in a number of different teaching strategies). One possible explanation is that teachers with more economically disadvantaged children and children of color may feel pressure to focus on literacy and math instruction using a variety of methods to help close the achievement gap. Regardless of the explanation for them, the present findings do not obviate the need to identify how different teaching practices affect different groups of students. They suggest instead that whatever is learned about effective instructional strategies may not have to contend with widespread preferences for using particular strategies for particular groups of children in classrooms.
The findings provide no evidence to support the view that White, middle-class children benefit more from child-centered instruction than children of color or children from low-SES families, nor does it support the view that children of color or low-SES children benefit more from skills-based instruction (e.g., Delpit, 1995; Engelmann, 1999). No significant interactions emerged between teaching strategies and either children’s race/ethnicity or their family SES. The results therefore do not offer support for teachers providing more skills-based instruction to children of color and low-SES children than to middle-class White children, as has been found to be the case in some previous studies (e.g., Anyon, 1981; Bargagliotti et al., 2009; Dahl & Freppon, 1995; Early et al., 2010; Rathbun & Hausken, 2003; Smith et al., 2001; Stipek, 2004).
There was evidence, however, that children who began kindergarten with relatively high math skills benefited more from one instructional strategy—Applied Math—than children who began with relatively low skills. In fact, the association between Applied Math and spring test scores was statistically significant and positive for children with relatively high math skills in the fall of kindergarten but was not statistically significant for children with relatively low fall skills. The implication is that for classrooms with a large number of children who start the year struggling in math, the teaching activities that make up Applied Math may exacerbate the gap between the low- and high-performing students. We can only speculate about why Applied Math activities appeared less effective for the children who entered kindergarten with relatively low math skills. Children may not have had the prerequisite skills to benefit from instruction focused on such tasks as measurement and telling time.
The greatest limitation of this study is its reliance on teacher self-reports of their instructional practices and the use of frequency, not amount of time, to capture instruction. Prior research indicates that measures of instruction from surveys asking teachers to report the frequency they engage in specific teaching activities have good validity and reliability (Mayer, 1998, 1999). Teachers in this study, however, may not have used the same criteria when making their frequency judgments. Teachers were asked about instruction late in the school year. Most likely there was variation in how frequently any given strategy was employed over the course of the academic year, and teachers had to adjust their responses to reflect this variation. Some teachers may have focused on how frequently they engaged in an activity in the most recent months (easier to remember), while others adjusted their estimate by averaging over the course of the year. The fact that the different teaching strategies were often positively correlated with each other provides evidence that the measures of instruction were problematic. A higher frequency of some practices would be expected to result in a lower frequency of others, which, in contrast to what we found, would yield some negative correlations among teaching strategies.
We highly recommend that in future research if classroom observations cannot be used to assess teaching practices for the entire sample of classrooms, they are used to validate teacher self-reports and identify potential biases. Classroom observations that are conducted several times over the course of the school year are ideal because they too are affected by the particular time of the year they are done.
A second limitation is that the study provides no information on the quality of implementation of any of the teaching strategies measured. The quality of implementation may be more important for children’s learning than the nature of activities teachers implement. Relatively new observation protocols that assess the quality of teaching directly and are validated by significant associations to student learning (e.g., Hill, Kapitula, & Umland, 2011; Pianta, Belsky, Vandergrift, Houts, & Morrison, 2008) are useful tools for future research of this kind.
Finally, the fact that the ECLS-K 1998–1999 data do not include literacy test scores for children who were not proficient in English serves as an additional limitation. Since these children were excluded from the analytic sample, the results for analyses of spring literacy and math scores do not represent all students. Given that a focus of the present study was on children’s ethnicity, the impact of the missing data for Hispanic students is of particular concern since many of these students were dual language learners.
Although this study has limitations, research on the effects of instruction on children’s learning gains is important. Too often debates about effective instruction are based on ideology. Empirical evidence is needed to inform the debate and, more importantly, to inform teachers’ practices. The findings reported here contribute to the literature by providing evidence that categorization of instruction only along the lines of child-centered or skills-based does not adequately capture patterns in how teachers teach. The level of skill targeted is also an important characteristic of instruction to include in future studies. The finding that teaching strategies did not differentially predict spring test scores based on children’s race/ethnicity or SES but did differentially predict spring math scores based on children’s skill level in math is further evidence of the importance of focusing on skill level in research and in the classroom, when making instructional decisions.
This research was made possible in part through the financial support of a Stanford Graduate Fellowship awarded to the first author. Tara Chiatovich is a data specialist at the Passaic Board of Education and is also a Strategic Data Project Fellow, a program of the Center for Education Policy Research at Harvard University. Deborah Stipek is a professor of education at Stanford University.
