Because of its interactive character the web may promote more accurate survey data than do other modes. One way that interactivity can increase data quality is by providing feedback to respondents about their answers. For example, questions known as tally or constant sum items require that multiple answers sum to a fixed total, e.g. 24 hours or 100 percent; for these items a web questionnaire can inform the respondent about whether or not their individual answers add-up to the target value. Designers can provide this feedback when the respondent submits the entire set of responses or in a running tally that is incremented as the respondent enters each component answer. We report three experiments that investigate how this kind of feedback aects the quality of answers for tally items. In experiment 1 we found that if the computer displays a running tally this increased the number of totals equal to the target sum (100% of a respondents internet usage) relative to feedback that is delayed until after the total is submitted, which in turn led to more totals equaling the target than no feedback. While promising, this result really concerns the number of well-formed answers (i.e., those equal to the target) which is not necessarily the same as the number of accurate answers because some respondents may produce a well-formed answer by adjusting one or two component answers to make the total add to the target without improving the accuracy of the components. So in experiments 2 and 3 we asked whether feedback about the total also led to more accurate data. In experiment 2, respondents reported how long it had taken them to complete each section of the questionnaire they were just nishing and we compared this to their actual times; the difference between their self-reports and the server-recorded durations was smaller when a running tally was displayed than when it was not. In experiment 3 we asked respondents to report how much time they had spent on each of several activities in the previous day and compared their time use estimates to published population estimates. Again, the running tally led to smaller dierences between the self- and published reports than when this feedback was not provided. Taken together, the studies suggest that the interactivity of the web can improve both the form and the accuracy of answers for this type of item. It is not possible to provide feedback like this with self-administered paper questionnaires and it is not easy to do so in interviews so this may be a case where web administration adds clear value over other modes of data collection.