What is a double-barrelled question?

How to spot one in your own draft

Search the draft for the words and, or, as well as, and any slash. Each one is a candidate. Then ask whether somebody could hold different views about the two halves. Bread and butter is fine because nobody has a separate opinion about each; speed and accuracy is not, because plenty of people do. The trap is subtler when the two ideas hide in one word. Asking whether a service was professional bundles competence, manner and appearance, and asking whether a process was simple bundles the number of steps with how clearly they were explained. Those are not strictly double barrelled but they fail the same test, which is whether two respondents could give the same answer for opposite reasons.

Why the resulting data cannot be repaired

This is the reason the flaw matters more than most. A biased scale can sometimes be adjusted for, and a missing option can sometimes be inferred. A double barrelled answer contains no information about which half drove it, and no analysis can recover it. A score of two might mean the reply was slow, or wrong, or both, and the three situations need different responses. Worse, the item will look perfectly healthy in the results: it has a mean, a distribution and a trend line, and nothing about it signals that it is uninterpretable. That is how a broken question survives ten waves of a tracking survey and drives decisions the whole time.

When combining is legitimate

There are cases where a compound question is the right one. If the two parts are genuinely inseparable in the respondent's experience, such as asking about a delivery that either arrived on time and intact or did not, a single item matches how people think about it. If the aim is an overall verdict rather than a diagnosis, a deliberately broad question can be the point, as long as the report does not later attribute the result to one component. And on a very short form, one compound question may be a reasonable trade for a completion rate. The test is whether you would be tempted to explain a movement in the number by pointing at one half of it, because that is the moment the shortcut becomes a mistake.

Splitting without doubling the length

The objection to fixing these is that the survey gets longer, and there is a neat answer: a matrix. Several short statements sharing one scale take far less effort per item than the same statements as separate questions, because the respondent learns the scale once. Four aspects of a course in one grid is about as quick as one compound question and produces four usable columns. The limit is around eight rows before people begin straight lining down one column, so split the grid rather than extending it. Keep the row text short, because long row labels wrap badly on a phone and make the grid unreadable.

Building the split version in a form

A matrix is rows by columns, with radio style allowing one answer per row and checkbox style allowing several, so four aspects on a five point scale is four rows and five columns in radio style. A matrix exports as one column with each statement and its answer joined as Statement: Answer, which keeps one response on one row but means you will split that column before charting any single statement. If one of the aspects is the number you report on regularly, ask it as its own radio field as well so it arrives in its own column.

Where they come from

Almost every compound question in a business survey has the same origin: two people wanted their question included and the draft was running long, so the two were merged. Recognising that pattern is the best defence, because the fix is a conversation about priorities rather than a wording change. A second common source is a question copied from a report heading, since headings routinely pair ideas that no respondent holds a joint view about. A third is translation, where a phrase that reads as one idea in the source language splits into two in the target. Reviewing a survey with the specific instruction to mark every item that joins two ideas takes about ten minutes on a twenty question draft and is the single highest yield review anybody runs. Do it before the survey is styled, not after, because nobody wants to reopen a form that already looks finished.

Questions people ask