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

more »

... d 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.