Although ordinary logistic regression is a widely - used tool, such models are often inappropriate given complicated data structures. We discuss methods to assess the quality of logistic regression models and explore alternatives to traditional regression models. To illustrate our findings, we investigate if characteristics of an address in the American Community Survey (ACS) can predict if a mailing is undeliverable as addressed (UAA) by the United States Postal Service. In 2012, local post offices reported that over 10% of mailed questionnaires in the ACS were UAA. By identifying the address and geographic characteristics for those mailings which are returned as UAA, we hope to identify certain types of addresses that are especially problematic and to provide suggestions for their improvement. To obtain this information, we will compare a variety of logistic regression approaches including mixed effects, generalized estimating equations, and spatial models. We will also investigate the use of classification trees for variable selection. We will discuss how to select an appropriate model and if our results can inform approaches to decrease the ACS UAA rate.