A new approach to estimating non-sampling errors using the analytic hierarchy process

This dissertation explains a study in which three approaches to estimating the relative proportion of different non-sampling error types in a very large database were examined. The U.S. Census Bureau designed the Master Address File (MAF) and the Topological Integrated Geography Encoding and Referencing (TIGER) database to contain the address and location for every residential housing unit in the United States. The first approach to estimating the non-sampling error types, the direct approach, asked experts experienced with the contents of the database to estimate the relative numbers for each type of error. The second approach uses pair-wise relative judgments in an Analytic Hierarchy Process (AHP) model without taking into account influencing factors. The third approach developed in group sessions with Geography Division experts is an Analytic Hierarchy Process model with influencing factors. The research hypothesized that a pair-wise approach would be more accurate than a direct approach and that a pairwise approach with influencing factors would be more accurate than the simple pair-wise approach. The results indicate that the pairwise with influencing factors approach provided better estimates than either of the two approaches. Also, that the pairwise without influencing factors approach was better than the direct approach.