Recommendation 14: The Food and Nutrition Service should monitor the accuracy of American Community Survey estimates of eligibility for the school meals programs, the accuracy of administrative certification estimates, and the accuracy and stability of differences between American Community Survey and administrative estimates. Such monitoring should be conducted for a broad sample of districts that are potential candidates for adopting the American Community Survey Eligibility Option but are not operating under that option or any other special provision or option. The Food and Nutrition Service also should monitor the accuracy of American Community Survey eligibility estimates for districts that have already adopted the American Community Survey Eligibility Option.
The benchmarking approach proposed by the panel for adjusting the ACS eligibility estimates for a consistent difference from administrative certification estimates uses only district-specific ACS and administrative data. The predictive modeling approach described in Chapter 4 and documented in Appendix F, which uses data from all districts as well as district-specific data to develop a benchmarking adjustment for each district, has some technical appeal and would benefit from additional research. In particular, FNS might sponsor research to develop and test a global predictive model that might be used for benchmarking either alone or in conjunction with a subsequent benchmarking adjustment based entirely on district-specific data.
Recommendation 15: The Food and Nutrition Service should sponsor research to develop and test a global predictive model for developing American Community Survey estimates of eligibility that are benchmarked to certification data for the school meals programs.
The specification and implementation of ACS school district-level model-based eligibility estimates produced by the Census Bureau and their subsequent evaluation by the panel took place in a brief time period. Although we could not recommend the use of the model-based estimates in place of the direct estimates at this time, the current models and estimates can be viewed as a proof of concept, and we believe refinement of a model-based approach could lead to more accurate and timely estimates than direct ACS estimates for many if not all school districts. Should implementation of the AEO prove successful, FNS should consider working with the Census Bureau to develop improved ACS model-based estimates. Specific suggestions for potential improvements in the models and estimation procedures are provided in Appendix C.