• The Census Bureau could not use the SAIPE relative error methodology to evaluate the estimation error of the eligibility rates for the school meals programs because it requires an independent source of poverty estimates.

• The SAIPE model uses shares from the 2000 decennial census long form as an independent variable. These shares are now 10 years old. The Census Bureau has not evaluated the use of shares from the 5-year ACS but suspects that they are less reliable. The models for the school meals programs do not use the decennial census data as an independent variable.

• The SAIPE shares methodology for the 2008 estimates did not use the direct ACS current-year estimate, so there would be a potential loss of information over the school meals model.

• The shares methodology is a two-step process, adding estimation error at each step.

PANEL’S SUGGESTIONS FOR MODELING ELIGIBILITY PERCENTAGES FOR THE SCHOOL MEALS PROGRAMS

As noted previously, the models for the school meals programs were developed quickly as a proof of the concept that using SAIPE-like small-area models for the school meals programs might provide accurate and timely estimates of eligibility. The panel considers that the work done to date demonstrates the feasibility of such an approach. While the model-based eligibility estimates for the school meals programs are timely, they did not prove to be as accurate as the 5-year ACS direct estimates. Accordingly, the panel believes that this promising approach would benefit from further research, particularly if the ACS Eligibility Option (AEO) is adopted.

Among general topics that might warrant research are (1) variations in the synthetic method used to determine school district or school attendance area estimates, (2) consideration of transformations of the variables entering model equations to improve modeling of county data, and (3) variations on the use of partitioning of county data to improve performance at the school district and school attendance area levels. The following are the panel’s specific suggestions concerning approaches for improving the models:

  • While the school meals programs are not “fixed-pie” fund allocation programs, controlling estimates to higher levels of geography should give the estimates greater precision and lower bias, while also improving face validity.


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement