While univariate instances of binomial data are readily handled with generalized linear models,\r\ncases of multivariate or repeated measure binomial data are complicated by the possibility of\r\ncorrelated responses. Likelihood-based estimation can be applied by using mixture distribution\r\nmodels, though this approach can present computational challenges. The logistic transformation\r\ncan be used to bypass these concerns and allow for alternative estimating procedures. One popular\r\nalternative is the generalized estimating equation GEE method, though systematic errors can\r\nlead to infeasible correlation estimates or nonconvergence problems. Our approach is the coupling\r\nof quasileast squares QLSs method with a rarely used matrix factorization, which achieves a\r\nsimplified estimation platformââ?¬â?as compared to the mixture model approachââ?¬â?and does not suffer\r\nfrom the convergence problems in GEE method. A noncontrived example is provided that shows\r\nthe mechanical breakdown of GEE using several statistical software packages and highlights the\r\nusefulness of the QLS approach
Loading....