The U.S. Census Bureau is committed to offering an Internet response option for the 2020 Census. We expect the majority of self-responses to come in through this medium. Decennial census field tests, such as the 2012 National Census Test and the 2014 Census Test, have used an online instrument to collect data in preparation for the 2020 Census. However, testing conducted through 2014 used online instruments that were designed for optimal view on a desktop or laptop. Although these surveys could be answered on tablets or smartphones, the design was not optimized for these smaller devices. On some mobile devices the screen display was very small and required the user to zoom or make other manipulations to enable the user to clearly read and answer the questions.

Mobile - ownership statistics show that, as of 2014, over half of adults owned a smartphone and some adults were dependent upon their smartphone for Internet access. These smartphone-dependent adults were more likely to be lower income, younger, and minority (Pew Internet Project, 2014). With the growth of mobile device ownership overall and the differences in device-dependent Internet access across subpopulations, the Census Bureau realized it must offer a responsive design for the online Census. That is, the questions and response categories must render optimally on the device, whether it is a large desktop computer or a small smartphone. A responsive design was developed for the 2015 Census Test. This means there was an optimized design for smaller devices such as smartphones and small tablets, and for larger devices such as large tablets, laptops, and desktops. Usability testing was conducted on different devices prior to fielding the survey.
This paper discusses the style rules we used to develop the mobile-optimized version of the 2015 Census Test instrument, and the issues that arose during usability testing. Additionally, we present device usage, completion time, and break-off data from the non-optimized 2014 and optimized 2015 Census Test online instruments. The two tests occurred in different geographic regions of the country with different population characteristics. So the typical usability metrics of time-on-task and task completion presented here to evaluate the effect of optimization are limited by sample confounds.