|
Abstract:
|
Consider a parsimonious model that imposes several restrictions on contingency table probabilities. Two facts are agreed upon generally: (1) When the model holds, restricted-model estimation is more efficient than unrestricted-model estimation; and (2) profile confidence intervals often have better coverage properties than their Wald counterparts. Motivated by these facts, this talk will describe restricted-model profile likelihood confidence intervals for a broad class of contingency table parameters. In particular, a simple and broadly applicable computational algorithm will be described. As an illustrative example, counts resulting from a stratified random sample will be used to estimate a common gamma coefficient in K rxc contingency tables. The proposed method is based on the theory of multinomial-Poisson homogeneous models.
|
