Return to Ken Judd’s Home Page
Return to Ken Judd’s 2021 Penn State Numerical Methods Course Home Page
March 23: Not your grandfather’s confidence intervals
Confidence intervals have generally relied on the Wald method of using local second-order approximations of the criterion function at the estimate. This ignores the fact that the criterion function often deviates significantly from those approximations away from the estimate, particularly for samples not sufficiently large for asymptotic results to apply. The likelihood ratio test is superior to Wald in many ways but is computationally demanding …. at least it was computationally demanding. Reich and Judd show how to use constrained optimization methods to construct likelihood ratio tests and provide examples of where their performance is far superior to Wald and bootstrap methods of constructing confidence intervals.