Comments
Eric Hanushek: This is a high-quality paper of the type that I have come to expect from these researchers. More importantly, they have turned their attention to a set of intellectual issues that have quite large policy importance. Murnane and his colleagues have picked up on Fryer and Levitt's very provocative analyses, which could be interpreted as suggesting that schools contribute to a growing racial gap in achievement. Murnane and others' analysis in this paper suggests that Fryer and Levitt's data and analysis are in question.
The most important Fryer and Levitt findings (in simplest terms) are that schools appear to contribute to a widening black-white achievement gap, but obvious school policies show little hope for improving the situation. Given the continued policy concerns about the distribution of a student outcomes, these findings that take the focus to the earliest school experiences lead to a new round of questions about what options are available.
Fryer and Levitt's unique study, however, needs some corroboration before policy is based on it. Specifically, much of the evidence is indirect—that differences in school achievement can be explained by neither measured differences between family background characteristics nor school factors, even though preschool differences can be explained by these factors. Further, the results could depend on the measures and samples of a new effort to study achievement—the Early Childhood Longitudinal Study (ECLS-K). This new education sample may in itself have some peculiarities.
This background motivates the study by Murnane and others. They have a simple but important research plan. First, they introduce a new and highly detailed data set from the National Institute of Child Health and Human Development (NICHD) in order to study the sensitivity of the results to the ECLS-K sample. Second, they investigate the sensitivity of the results to various aspects of the model specification.
This activity is extraordinarily valuable. There are many examples of alleged facts that come from one study or one sample proving to be much [End Page 128] less than facts when they are overturned by another study. This situation is especially problematic when policies are swirling around as they are in the area of the racial achievement gap and there is the possibility of making premature policy judgments. The authors of this paper show the merits of careful and imaginative replication. Their findings put a different twist on the racial gap. First, while they confirm the finding that black students enter school less prepared than whites, they find that the early achievement gap looks larger and less explicable than that portrayed by Fryer and Levitt. But second, they find that the gap shrinks with time in school, as opposed to expanding. Third, in the category of old results, they confirm that measured attributes of teachers and schools have little to do with either achievement or the racial gap (a Fryer and Levitt finding). Black and white students face roughly the same measured characteristics of schools, but they are not important in explaining achievement.
Murnane and his colleagues at the same time fail to confirm two common findings in previous work. They do not find that having a rookie teacher is a particular detriment, even though black students tend to get new teachers a disproportionate amount of the time. Further, these authors do not see that the more-segregated schools that black students are likely to attend are a particular disadvantage per se (although their measure of racial composition is quite imprecise). Instead, they find that racial composition is simply an imperfect proxy for the aggregate composition of student family income—that is, it is student socioeconomic status and not race.
Finally, the authors do find support for the idea that time on task has a positive effect. To people outside of education research, this seems like something that does not need to be explained. Nonetheless, this seemingly trivial finding has been difficult to document in a convincing way, and even...
