Beginning in 2013, the American Community Survey (ACS) collected data from respondents via the Internet as part of its self-response data collection. The ACS designed its Internet instrument to work with any electronic or mobile device, which came with the tradeoff that the instrument could not limit the types of characters respondents entered into open item survey questions. To limit incoming characters that ACS sends to downstream data processing, management created the Automated Response Cleanup (ARC) program to blank responses with insufficient data and perform other cleanup that the keyers of the paper questionnaire would normally handle at the National Processing Center. ARC also removes invalid characters from and properly formats Internet open responses.

The programming of ARC was subject to the same testing standards used in the development of the 2013 Internet instrument. Teams from various divisions within the Census Bureau planned, created, and reviewed the ARC program. The testing confirmed that the ARC was programmed according to its specifications; however, it did not confirm that the cleanup rules addressed all possible inputs or that that interaction of individuals rules produced output that the analysts and researchers had intended.

Team members expected that ARC would change entries for very specific reasons (like monetary items containing commas or decimals), but also counted on the program to blank any inputs that the ARC did specifically address in its cleanup rules. Members agreed that they would look at the data coming out of ARC after it was in production to make any enhancements to the program.

This report assesses the total responses that ARC changed during the first four months of 2013 to help determine if the ACS should consider enhancing its current rules. We quantify the input patterns of the responses that the ARC program specifically addressed (like commas or decimals in monetary field) in addition to responses that the program blanked because they were either uninformative/insufficient or the program did not have specific instructions to salvage the data otherwise.