Formulations are provided for correlated sets of univariate outcomes allowing for missing values, generalized linear modeling of means for such outcomes, correlation structures, and the standard generalized estimating equations (GEE) approach to modeling those outcomes. Four cases for generalized linear modeling of means are considered: linear regression with the identity link function for normally distributed continuous outcomes, Poisson regression with the natural log link function for Poisson distributed count outcomes and related rate outcomes, logistic regression with the logit link function for dichotomous 0/1-valued outcomes, and exponential regression with the natural log link function for exponentially distributed positive-valued continuous outcomes. Four directly specified correlation structures are considered: independent, exchangeable, spatial autoregressive order 1, and unstructured correlations. Formulations of a likelihood function for addressing standard GEE modeling and of likelihood cross-validation (LCV) scores for GEE model selection are provided. Two approaches for computing LCV scores are considered: matched-set-wise deletion and measurement-wise deletion. An overview of adaptive regression for modeling nonlinear relationships is provided as well as descriptions for four data sets to be used in later adaptive analyses.