The 2020 Decennial Census will be the first to use the Internet as the primary (and preferred) means by which data are collected from households. As currently planned, households will receive by U.S. Postal Service a postcard that provides an Internet link as a text URL and as a scannable QR code. The postcard will also have a printed, human-readable unique ID that links the postal delivery address to an entry in the Master Address File (MAF). Respondents will be asked to enter this identifier on the web form accessed by the URL.

Non-ID forms (those received on the Internet without a census-provided ID) pose certain challenges. The main text of this report uses the concept of a ROC (receiver operating characteristic) curve to illustrate the tradeoff between undercounts and erroneous counts. Without some countervailing attention, there is a danger that non-ID processing will end up on the “wrong side of the ROC curve”, leading to unacceptable break-off rates and hence undercounts. In particular, JASON finds that “identity validation” as practiced in ordinary commerce is likely not a useful model for the Census, because its ratio of penalties for false positives to those of false negatives puts it in a vastly different regime from the Census.