Collection of Census data by Web survey is anticipated to be an important component of Census 2020. To prepare for this, the 2010 Census Quality Survey (CQS) followed-up a sample of 2010 Census respondents with an internet survey (among other modes) The Web survey collected a variety of serverside paradata from respondents as they completed the online CQS Survey. These included measures of how long it took to complete the each question and the entire survey, which items were changed (including original and corrected values), whether the respondent could complete the survey in one sitting, and many other issues of interest to Web survey methodologists. We analyze these paradata, in combination with knowledge about the survey questions (e.g., which items are more complex or sensitive on face value), with an orientation toward interpreting the paradata as measures of respondent burden (e.g,. length of the time to complete questions), cognitive difficulty (e.g., changes to answers) and other psychological states that can lead to measurement error, item nonresponse error, or unit nonresponse error (in the form of break-offs). The relationship of post-hoc Web survey paradata analysis to traditional lab-based usability testing at the Census Bureau will be discussed. Implications for mode-specific Total Survey Error components that can be linked to the interaction of respondents‘ psychological states and Web survey design will be emphasized.