3-year periods. (For the same reason, these estimates are also the least responsive.) Interestingly, the between-year changes in state-provided shares are much larger in 2006-2007 than in 2007-2008. We do not have enough detailed information about changes in state practices to identify specific reasons for the changes that might cause this variation and predict whether results would be similar in future years.
The more detailed information by state share grouping sheds more light on patterns of volatility. In absolute terms, the largest part of annual changes in share occurs in the states with relatively large shares; as noted above, these encompass about 74 percent of allocations. However in relative terms, these states show the least volatility by any measure. Since volatility in ACS estimates is largely driven by sampling variation, it is consistently larger in relative terms for each group of successively smaller states. The pattern is less consistent in the state-provided estimates, although generally the larger states tend to have more stable numbers. This stability may reflect the greater effects on smaller states of rapid changes in ELL population in a few local areas, or it may reflect changes in reporting. Overall, the 3-year ACS estimates appear to be the most stable, at the cost of some loss of responsiveness. And as discussed in Chapter 2, 1-year ACS estimates do not capture year-to-year changes with acceptable precision. The evidence is ambiguous on comparative stability of single-year ACS estimates and state-based estimates.
CONCLUSION 5-4 The superior precision and stability of the 3-year American Community Survey (ACS) estimates outweigh their slower responsiveness to changes and make them superior to the ACS 1-year estimates as a basis for allocations.