Several interviewer-administered U.S. Census Bureau surveys, like the Annual Social and Economic Supplement of the Current Population Survey and the Survey of Income and Program Participation, collect income and expense data via open-ended numeric questions. These questions are essential to producing accurate statistics about the financial health of the United States, but they can be challenging for respondents to answer. Respondents may not know the exact amounts, can view the information as highly sensitive, or may prefer to answer using range options since reporting discrete amounts can be cognitively burdensome. Currently, interviewers play a vital role in encouraging respondents to provide precise values for these questions. As major U.S. Census Bureau household surveys prepare to integrate web survey response modes, it is crucial to explore effective methods for displaying these questions online in ways that motivate accurate responses, maximize data quality, and reduce respondent burden.

Using data from the web-administered Household Pulse Survey, this experiment tested three approaches for reporting discrete values for mortgage, rent, utilities, and the down payment for home purchases. The control group received the interviewer-administered wording requesting respondents to provide a dollar amount in a write-in box. The first experimental group was given an additional instruction telling them to provide their best estimate if they were unsure. The second experimental group had an “I’m not sure” checkbox below the write-in box they could select instead of reporting an answer. This study shows that an instruction on the item encouraging respondents to estimate if they are unsure is an effective way to reduce item missingness and the risk of breakoffs on cognitively burdensome items when compared with a control question without instructions. There is also evidence that the “I’m not sure” checkbox and the instruction to estimate may reduce the risk of survey breakoffs when compared with a control group that does not provide motivational or encouraging statements; however, this pattern was less clear when breakoffs were compared between the checkbox and the instruction-to-estimate treatments. As survey practitioners consider adding instructions or using nonsubstantive response options, careful considerations should be made to these tradeoffs in data quality and respondent burden.