In epidemiologic and public health studies of disease incidence in geographic subpopulations, attention is properly directed toward the ascertainment of accurate numerators. Population or person-years denominators are generally given less consideration, under the assumption that estimates produced by sources other than the state health department are sufficiently accurate. Here, we report our experience in estimating person-years denominators in the highly urbanized, rapidly expanding population of Maricopa County, Arizona. The usual sources of population estimates were found to be of little use for public health purposes, and so we report on a method for obtaining smoothed person-years figures that can accurately reflect population acceleration which varies from one time period to another. Our method is to regress the logarithm of census enumerations on quadratic or tertic polynomials in time. We describe how differential reliability of census figures can be incorporated into our procedure, and how the problem of missing census data can be handled by an iterated regression method. Our evidence suggests that the logarithmic regression model works well, even in the face of rapid and erratic population growth or decline.
