Using a Geographic Segmentation to Understand, Predict, and Plan for Census and Survey Mail Nonresponse

Using a Geographic Segmentation to Understand, Predict, and Plan for Census and Survey Mail Nonresponse was written by Nancy Bates and Mary H. Mulry in 2011. It was published in the Journal of Official Statistics (vol. 27, no. 4).

The authors design a segmentation of Census tracts using the 2000 PDB. Specifically they use the SAS FASTCLUS proc. Guiding principles were interpretability and actionability. Segments must be large enough to support further analysis. Segments must be clearly distinguished in terms of characteristics to be useful in planning the census. Finally, segments must vary in terms of response rates to be useful. The authors arrived at 8 segments through exploratory analysis.

1. All around average I (homeowner skewed)
2. All around average II (renter skewed)
3. Economically disadvantaged I (homeowner skewed)
4. Economically disadvantaged II (renter skewed)
5. Ethnic enclave I (homeowner skewed)
6. Ethnic enclave II (renter skewed)
7. Young/mobile/singles
8. Advantaged homeowners

These segments are evaluated with respect to the 2000 Census, the 2010 Census, and the 2010 ACS. Largely, their predictive power is validated.

The authors elaborate that the segmentation is intended for highlighting tracts that may require special or additional attention when planning the Census. They are not necessarily intended for quantitative analysis. Furthermore, the segmentation is based on dated information and warrants revisiting with fresher sample.

CategoryRicottone CategoryReadingNotes