[CTPP] Census "Planning DataBase" -- predicting response and targeting to increase response

Target audience for this post: survey methodologists On Nov 19, I attended an ACS Data Users Group webinar hosted by the Population Reference Bureau and Sabre Systems. The topic was the Planning Database (PDB), which is a database at the tract and block group level, focusing on low response. The main objective is to figure out where the Census will have lower response with the 2020 Census and plan to implement different techniques to improve response rates in those areas. www.census.gov/research/data/planning_database<http://www.census.gov/res… The speakers were Nancy Bates and Travis Pape. The PDB includes the geography (tract and block group), demographic characteristics, and then, Census operations data, e.g. mail-back, bilingual rates, the 2010 Census Mail return rate AND the LOW RESPONSE SCORE, based on a model using 5-years of ACS data. The model includes 25 variables, of which the top 3 variables were: rent (vs own); age ( 18-24), and female head of household with no husband present. Other variables include household size, presence and age of children, poverty, education, race and Hispanic origin. The regression results are R2 of .56 at the block group level, and R2 of .55 at the tract level. These results are for mail-back returns, and the Census Bureau is planning on over 60% returns for the 2020 Census using Internet, so the results may be different after newer ACS results which included Internet response method are incorporated into the model. Currently, the 2014 PDB which includes 2008-2012 ACS estimates and 2010 Census operations data is available, and an updated PDB based on more recent ACS will be available approximately in March 2015. The files are only available as nationwide files, zipped CSV format, one file for tracts and one file for block groups. The tract level file is 86 Mb. The Block Group file is 152 Mb. Bottom line - the model variables seem consistent with the results of non-response to household travel surveys, so I am thinking that when a large household travel survey is being planned, these data could also be useful for targeting different approaches to improve response rates, or oversampling housing unit addresses in low responding areas. Hope I didn't make any grievous errors in my summary above! If I did, I hope that someone from the Census Bureau will post corrections. Elaine Murakami FHWA Office of Planning 206-220-4460 (in Seattle)