To the Editor:
For decades, “free and reduced-price school lunch” data have served as the major—and often the only—measure of child poverty in educational research and evaluation. “Analysts Rethinking Popular Indicator of Child Poverty” notes important recent criticisms of the adequacy of these data. These criticisms include analytic challenges posed by patterns of program participation, program-eligibility changes, and so on.
As statisticians say, these data are “noisy.” While readily available, “school lunch” data, based on eligibility and program participation, are at best a crude measure of poverty and disadvantage because students are merely “eligible” or not. And factors affecting participation mean that the data also include large amounts of random and nonrandom error due to social forces, program implementation, and so on.
Your article fails to note that the more random error in a measure, the smaller any real effects will appear to be in analyses. Even in well-designed studies, statistical noise can lead to underestimates of the impacts of child poverty on teaching and learning.
We now have decades of educational research and program evaluations built on school lunch data that often report findings of little or no significant effect. Because of their reliance on crude measures of key variables—such as child poverty—it is hard to know whether such findings are real or just statistical noise.
Good measurement matters. We must do better.
Danièle Rodamar
Associate Professor
World Languages and Culture
American University
Washington, D.C.
