Spatial Analysis of Census Mail Response Rates: 1990–2010 - Space-Time Integration in Geography and GIScience - page 141

Chapter 9
Spatial Analysis of Census Mail Response
Rates: 1990-2010

Daniel A. Griffith and Yongwan Chun

9.1 Introduction

As early as 1990, the United States (US) Census Bureau recognized the importance
of analyzing the geographic variation in census mail response rates (MRRs)
at the county level across the country. Keller ( 1990 ) notes the importance of:
understanding where MRRs decline; gaining insights in preparation for the next
decennial census; and, the usefulness of developing a predictive equation that allows
a more accurate forecast of the MRRs for the subsequent census.
Although the MRRs for a given census can serve only as a covariate for subse-
quent census (i.e., these rates are not a causative indicator, but rather a covariate),
they are a very valuable piece of predictive information to include in an analysis.
In a multiple linear regression analysis using county level data, Keller ( 1990 ) found
that the MRR covariate ranked second only to the percentage of occupied housing
units. His equation contains 36 additional variables (e.g., household size, percent
post office boxes), with the total set of 38 variables accounting for roughly 69 % of
the variation in the MRR across US counties.
The geospatial data for an analysis of MRRs consist of a county map shapefile
for the coterminous US and county-level MRRs for the decennial census years
1990, 2000, and 2010. 1 Restricting attention to the coterminous US for analysis
purposes here, 25 of the 3,067 counties are classified as not being parts of states, and

;