The presentation order isn’t random. In that case, the people who keep answering all the way to the later items would be a particular subset, so the results could end up biased. Also, was it the case that you distinguish dropouts from non-responses and subtract dropouts from N?
It’s not purely random, but it’s not fixed either. Each person sees a different order. So I don’t think the items that appear later suffer from position-based bias.
Dropouts are distinguished from neutral responses. In the detailed report, dropouts are shown in white and neutral responses in gray. Dropouts are excluded from the denominator when percentages are displayed. In the matrix computations, missing values are imputed with the average of other respondents’ responses.
Thanks. So there is some bias because it isn’t random, but since it isn’t completely fixed the bias is small—got it? Imputing with the mean of other responses is a bit questionable (it pulls things toward the center).