Statistics Team Develops New Census Model

Jonathan Bradley
Jordan Yount
News Source: 
College of Arts & Science

For roughly a decade, the U.S. Census Bureau has been publishing the American Community Survey (ACS), which provides timely information on a number of key demographics. The ACS, which replaced the Census Bureau’s long-form decennial census, is published in one-year, three-year and five-year period estimates. This information is used by city planners, demographers, and researchers for a variety of applications such as understanding how a particular community changes over time. However, the Census Bureau recently announced it was ending the three-year period estimates, prompting a team of statisticians at MU to develop a new model that can be used to estimate values for years and locations in which the ACS does not publish three-year period estimates.

Jonathan Bradley, a postdoctoral fellow in statistics, is being supported on a National Science Foundation – United States Census Grant under the mentorship of statistics professors Scott Holan and Christopher Wikle to develop new spatio-temporal modeling techniques for the American Community Survey.  Bradley says their solution, Spatial–Temporal Change of Support with Application to American Community Survey Multi-Year Period Estimates, will allow demographers and researchers to find values for any time period and any geographic area they want.

“We take into account the survey variances—the margins of errors that they publish with the estimates,” Bradley says. “The reason why they have one-year, three-year, and five-year estimates is that the more information you average, the more precise your results will be. So one-year period estimates are very sparse, they don’t have much data, but they are the most recent, so there’s a balance between currency and precision. We take into account those measures of error to obtain more precise estimates and better quantify uncertainty.”

Bradley says the ACS’ one-year period estimates are only performed for communities of a certain size, so in Missouri, for example, household income data would only be available in the one-year estimates for St. Louis, Kansas City, Columbia, and other highly populated areas. He says the new model will allow researchers throughout Missouri to fill in the gaps in their data sets.

“In terms of spatial analysis, the way it works is nearby things tend to be more similar,” Bradley says. “Suppose you know it is raining at Point A and if Point B is right next to it, then you have a pretty good idea that it is raining at Point B. However, at a certain distance from Point A, you don’t really know whether it’s raining or not at point B. There is a spatial correlation between Point A and Point B we take advantage of to get estimates.” He says the same methodology works in a temporal context as well.

Bradley says the new methodology can be applied in areas as diverse as environmental science and ecological modeling, among others, and he says the potential impact to data-users and policy makers interested in ACS custom-designed tabulations is unparalleled.

Share This

Facebook icon