Assessing quality improvement initiatives when expert judgements are uncertain

A new approach for examining quality improvement initiatives regarding errors in the U.S. Census Bureau’s Master Address File (MAF) and the Topologically Integrated Geographic and Referencing System (TIGER) databases is presented. A stochastic multi-criteria decision-making method involving Bayesian weighted hierarchical multinomial logit models is used to conduct inference on the priorities in a multiple-expert decision scenario. Quality initiatives regarding basic street addresslevel address matching, geocoding completeness, and geocoding quality were judged to be the most important for MAF/TIGER improvement at the 95% probability level. The approach allows managers to go one step further in understanding the relative impact of various types of errors on overall quality and thus be better prepared to select approaches to reduce these errors.