The health planners dilemma and the problem of aggregation bias--an irresolvable conflict.

Abstract

This study examines the potential biases that may exist when data which are collected at one level of aggregation (e.g., the United States) are extrapolated to a smaller area of aggregation (e.g., a census region consisting of several states). Theory would predict that there would be inaccuracies resulting from such extrapolation but does not suggest whether the error would be of sufficient size as to have practical significance for a planner. This study clearly shows that attempting to predict smaller area hospitalizations with large area data is highly inaccurate. It is not only consistent with theory but also shows that the theory has practical value in assessing approaches to estimating hospitalization data.