Analysing survey results without measuring who answered.

A survey with two thousand responses and a forty per cent completion rate among unhappy respondents is less trustworthy than one with four hundred responses and even completion. Size does not fix a systematic absence.

The check takes minutes: compare who finished against who started, split by whatever your headline comparison is about. If dropout correlates with the answer, the headline is measuring dropout.

The decision about partial responses then changes the result, so it belongs in the write-up rather than in a setting nobody looked at. Discarding them is the default in most survey tools and it silently removes the people most likely to be dissatisfied.

Negatively worded items have to be reversed before anything is averaged, or they cancel out the items they were meant to corroborate. That much is standard.

The step people skip is checking afterwards. Respondents frequently answer a reverse-worded item as though it were positive, and an item that still correlates oddly with the rest of the scale after recoding is not measuring what its neighbours measure. Including it drags the score toward noise.

A select-all question is stored as one binary column per option. Treated as separate questions, the percentages sum to something meaningless and every option looks like a minority view.

The base has to be respondents rather than responses, and it should be stated, because forty per cent of respondents and forty per cent of selections are different claims that get reported identically.

Written answers are usually summarised separately from the scores, which wastes both. A theme appearing in forty comments means something different depending on whether those forty people rated you two or five.

Grouping the text by theme and showing the mean score per theme is the single most useful thing available, and it takes very little work once the responses are coded.