An Early Start: Skill Grouping and Unequal Reading Gains in the Elementary Years

Abstract

Separating elementary school students into skill-based reading groups within classrooms affects a vast majority of young children in the United States. The impact of this institutional process on students' learning has important implications for sociological perspectives on education and stratification, yet a lack of studies comparing similar grouped and nongrouped students has prevented scholars from drawing conclusions as to the salience of this type of curriculum differentiation. Drawing on data from the first- and third-grade waves of the Early Childhood Longitudinal Study-Kindergarten Cohort, I use propensity score matching techniques to estimate the impact of low, middle, and high group placement on reading gains relative to nongrouped instruction. Findings suggest that high-grouped students learn more, and low-grouped students learn less, than comparable nongrouped students. These analyses, which significantly lessen the extent to which selection into groups may bias results, add strong evidence to the view that within-classroom skill grouping in the early elementary years promotes unequal reading gains compared to nongrouped instruction. I conclude by discussing the theoretical and policy implications of these findings.

ACKNOWLEDGMENTS

I thank Douglas B. Downey, Vincent J. Roscigno, Zhenchao Qian, and Kendralin J. Freeman for helpful comments and suggestions on earlier versions of this manuscript.

NOTES

Notes

¹ The process through which teachers place students into groups is complicated and has a literature of its own, and a thorough exploration of that process is beyond the scope of this study. For my analyses, the differences between grouped and nongrouped students are more relevant. I have, however, confirmed that most of these student background attributes are indeed related to group rank in the ways noted here (Condron 2007; see also Haller and Davis [1981] and Haller [1985] for examples of studies that focus specifically on group placement).

² The ECLS-K asked teachers the same set of questions about skill grouping in kindergarten, first grade, and third grade, but the survey administered in fifth grade is slightly different and does not allow researchers to determine students' group placements that year.

³ I actually used multiple imputation, which replaces missing values several different times in order to introduce random variability into the imputation process (see Allison 2002). The intent of multiple imputation is to use these multiple data sets to run multiple models and then combine (average) the results into one set of final estimates. However, since I use multiple data sets depending on the wave (first or third grade) and the comparison (nongrouped versus low, middle, and high grouped), and since I run models within 10 different propensity subclasses (five per wave), analyses drawing on multiply-imputed data sets become extremely cumbersome. Therefore, all analyses draw on only one data set with imputed missing values.

⁴ There are other possible ways to determine whether students were grouped, but I believe the method I use is the best. My measures of group placement are derived from Part C of the teacher questionnaires (“t4nordgp” and “t4chrdgp” for first grade; “t5nordgp” and “t5chrdgp” for third grade), where teachers were asked how many “achievement groups in reading” they use and which group the focal student is currently in. Other measures, from Part A of the teacher questionnaire, pertain to grouping but (1) sometimes use terms like “instructional groups,” which may be different than skill/achievement-based groups, and (2) do not ask the teacher which group the student is in. I therefore rely on the Part C questions to code students' group placements.

⁵ The data reveal neither the nature of the student's disability nor whether it influences his or her reading instruction. In supplemental models, I excluded students with a disability and found the same pattern of results as that presented below.

⁶ All analyses are unweighted. Supplemental weighted analyses yielded estimates very similar to those reported here.

⁷ I supplemented the analyses presented in the text with three alternative analytic strategies. Rather than modeling reading gains within propensity subclasses, I used an analysis of covariance framework controlling for all the covariates in Tables 1 and 2 and estimating the effects of low, middle, and high group placement on reading gains relative to nongrouped instruction. Results from OLS, generalized least squares, and HLM regressions are highly consistent and all point to the same overall conclusion: Skill grouping leads high-grouped students to learn more, and low-grouped students to learn less, than comparable nongrouped students. The pattern of results does not depend on the choice of models, suggesting that the findings are robust.

⁸ The ECLS-K asked teachers whether the focal child remained in the same group, moved to a higher group, or moved to a lower group. In first grade, roughly 20 percent of the students moved to a higher group while about 4 percent moved to a lower group. In third grade, roughly 7 percent of students moved up and 1 percent of students moved down. However, it is entirely unclear when this group mobility occurred, making it difficult to determine and interpret the impact of this group mobility. In supplemental analyses, the results presented here hold when I control for whether students changed groups and when I exclude students who changed groups from the models.

⁹ Relatedly, supplemental analyses testing for interactions between group placement and initial reading skills suggested that the positive effect of high group placement on gains is weaker among initially high-skilled students than it is among initially low-skilled students. Again, it is difficult to know if this is an indication of a ceiling effect in the reading test scale or rather an instructional phenomenon.