On Ill-Conditioned Generalized Estimating Equations and Toward Unified Biased Estimation

I address the issue of ill-conditioned regressors within
generalized estimating equations (GEEs). In such a setting,
standard GEE approaches can have problems with: convergence, large
coefficient variances, poor prediction, deflated power of tests,
and in some extreme cases, e.g. functional regressors, may not even
exist. I modify the quasi-likelihood score functions, while
presenting a variety of biased estimators that simultaneously
address the issues of (severe) ill-conditioning and correlated
response variables. To simplify the presentation, I attempt to
unite or link these estimators as much as possible. Some
properties, as well as some guidelines for choosing the meta or
penalty parameters are suggested.