Is ‘Don’t Know’ Missing Data?

A blank, “don’t know,” and “neutral” can look equally inconvenient in a spreadsheet. But they tell different stories.

A blank tells us no answer was recorded. “Don’t know” tells us the respondent could not, or chose not to, give a substantive answer. “Neutral” records neither agreement nor disagreement, although it does not explain why.

Missing data is an unavailable value for something we intended to measure. The important question is: which value?

If we ask whether someone supports a policy, “don’t know” is an observed response, but it leaves their support unmeasured. If we are studying uncertainty itself, that same response becomes a central finding.

Neutrality deserves equal care. A midpoint can express a considered position, mixed feelings, indifference, or uncertainty. Sturgis and colleagues used follow-up questions to show that some apparently neutral responses expressed a lack of opinion. The answer category and the underlying attitude are not necessarily the same thing.

Statistical theory adds another complication: why a value is missing matters. Randomly lost responses present a different problem from income answers withheld because incomes are unusually high. The latter can systematically distort what remains. Van Buuren explains how missing-data mechanisms affect the conclusions we can draw.

There is also a deeper assumption hiding here: that a value exists to be recovered. An unformed opinion may not be a settled belief waiting behind a blank cell. Yet “don’t know” alone cannot tell us whether an opinion is absent, uncertain, or simply difficult to express.

The practical lesson is to preserve these distinctions. Replacing “don’t know” with neutral invents a position. Deleting it changes whose views the analysis describes.

A response can leave our intended measurement incomplete while still telling us something important. Before cleaning it away, we should understand what we would be losing.