1. Obtaining the bounds on the second-stage regression line implied by the precinct bounds is straightforward for the bivariate case. Although an infinite number of hypothetical data sets—and corresponding feasible regression lines—could lie inside the p precinct bounds, these lines must lie in a finite (vertical) interval. Thus for any Zi there must exist a maximal (minimal) regression line above (below) which no feasible regression line exists. To find all maximal and minimal regression lines, we create a data set for each Zc ∊ (Z 1, … Zi , … Zp ) with equal to the lower bound if Zi ≥ Zc and equal to the upper bound otherwise, and vice versa, yielding 2p data sets and associated regression lines. The 100% confidence intervals plotted in Fig. 5 are the minimum and maximum of these 2p lines at each Zi . The next version of EI and EzI will include a feature to calculate these second-stage bounds, along with the other developments in HS, AK, and HS2. The software will be available at http://gking.harvard.edu.
2. Cross-Contamination in EI-R: Reply
3. Variance Estimation for Superpopulation Parameters;Korn;Statistica Sinica,1998
4. Some Current Trends in Sample Survey Theory and Methods;Rao;Sankhya: The Indian Journal of Statistics,1999
5. We have also not studied the accuracy of standard errors, but the usual model-based standard errors, even when they are accurate, would still be a relatively minor portion of the uncertainty in most applications of ecological inference, most of which is the result of specification uncertainty. Christopher Adolph et al.