Abstract:
|
A weighted pseudolikelihood has been used to fit regression models for decades. The influence functions corresponding to this pseudolikelihood give estimating equations: for generalised linear models these are simply weighted versions of the score equations. In loglinear models for contingency tables the log likelihood ratio statistic gives one of the famous Rao--Scott tests, but at the time these were intended primarily as corrections to a common practice of doing naive association tests on estimated population tables. A large chunk of Alastair's post-retirement work in survey statistics, however, focused on this likelihood as a basis for regression inference: likelihood ratio tests, AIC, and BIC. I will also talk about what is was like as someone completely new to the survey field to work with Alastair and some of his contributions to surveys in New Zealand.
|