Written and maintained by CASRAI Editorial Board
Last updated
A double-barrelled question asks about two things at once but allows only one answer. It is one of the most common defects found in survey instruments during pre-testing, and one of the easiest to fix once it is actually spotted — the difficulty is almost entirely in spotting it, not in the fix itself. This guide gives a single mechanical test for finding double-barrelled items in a draft instrument, then works through real before/after rewrites across the three question formats where the problem shows up differently: Likert, yes/no, and frequency.
What Makes a Question Double-Barrelled
A question is double-barrelled when it packages two (or more) distinct claims, attitudes, or behaviors into one item, but the response scale only lets the respondent register a single answer. The defect is not that the sentence is long or that it contains the word “and” — plenty of well-formed items contain “and.” The defect is that the two halves are separable: a real respondent could reasonably hold a different view of each half, and the instrument gives them no way to say so.
“Is the customer service polite and effective?” is double-barrelled because politeness and effectiveness are independent dimensions — a support agent can be warm and unhelpful, or brusque and highly effective. A respondent who experienced one but not the other has no correct box to check. By contrast, “Do you regularly exercise and eat vegetables?” on a pre-screening item asking whether someone follows a general wellness routine is arguably still double-barrelled if the two behaviors are meant to be scored separately, but a question like “Did you find the checkout process fast and easy?” is double-barrelled precisely because speed and ease are the kind of experience dimensions users routinely rate in opposite directions.
The Diagnostic: the “And/Or” Test
Run every draft item through this two-step test before it goes near a respondent:
- Find every conjunction. Scan the item for “and,” “or,” and implicit conjunctions hidden inside a compound noun phrase or list (“the price and packaging,” “staff who are knowledgeable, friendly, and available”). Flag every one.
- Ask whether a respondent could reasonably answer each half differently. Split the item at the conjunction and read each half as its own question. If you can construct a plausible respondent — not a contrived edge case, but a realistic one — who would answer “yes” to one half and “no” to the other, or who would rate one half a 5 and the other a 2, the item is double-barrelled and needs to be split.
The second step is the part that actually catches problems; the first step just tells you where to look. A conjunction alone is not disqualifying — “Are you a full-time or part-time employee?” uses “or” but is a single, mutually exclusive classification question, not two separable claims. The test is whether the halves are independently answerable, not whether the sentence is grammatically compound.
A useful secondary check: try to imagine the respondent’s internal monologue. If answering the item plausibly requires them to silently average, pick whichever half feels more important, or guess which one you actually meant, the item has failed the test even if no single respondent would consciously notice the ambiguity. That silent-averaging behavior is exactly what corrupts the data — a resulting “4” on a satisfaction scale can mean “5 and 3, averaged” for one respondent and “4 and 4” for another, and the instrument cannot tell them apart.
Worked Before/After Examples, by Question Type
Double-barrelling shows up differently depending on the response format, because each format constrains what a single answer can actually communicate.
Likert-Type Items
Likert items are especially prone to the problem because the natural language of attitude statements invites compound claims (“the product is affordable and reliable”), and a single agreement scale absorbs the ambiguity silently — there is no visible sign in the response data that the item was compound.
| Before (double-barrelled) | Why it fails the and/or test | After (split) |
|---|---|---|
| “The training was well-organized and delivered by knowledgeable instructors.” (Strongly Disagree–Strongly Agree) | Organization and instructor knowledge are independent; a well-run session can have a weak instructor, or vice versa. A respondent forced to pick one score is silently averaging or guessing which half you care about. | Item 1: “The training was well-organized.” Item 2: “The instructors were knowledgeable.” |
| “This app is fast and easy to use.” | Speed and usability are commonly dissociated in real usage — a fast app can have a confusing interface. | Item 1: “This app is fast.” Item 2: “This app is easy to use.” |
Yes/No (Dichotomous) Items
Yes/no items make double-barrelling more visible once you look for it, because the binary answer has nowhere to hide a partial response — but it is also the format where the underlying data damage is worst, since there is no partial-credit score to signal that something was off.
| Before (double-barrelled) | Why it fails the and/or test | After (split) |
|---|---|---|
| “Did you experience side effects and stop taking the medication?” (Yes/No) | A respondent may have experienced side effects and continued the medication, or stopped for an unrelated reason. “Yes” and “No” both lose information about which happened. | Item 1: “Did you experience any side effects?” (Yes/No) Item 2: “Did you stop taking the medication before the course was complete?” (Yes/No) |
| “Have you read and agreed to the terms of service?” | This one is a genuine and common instrument defect, not just a theoretical one: a respondent can click through without reading, meaning “Yes” conflates two very different states (informed consent vs. procedural compliance). | Item 1: “Have you read the terms of service?” (Yes/No) Item 2: “Do you agree to the terms of service?” (Yes/No) |
Frequency Items
Frequency scales (“Never” to “Always,” or a numeric count) are the format where double-barrelling is easiest to miss during drafting, because the two behaviors being asked about often co-occur often enough in ordinary life that the compound phrasing feels natural — right up until a respondent for whom they don’t co-occur answers the survey.
| Before (double-barrelled) | Why it fails the and/or test | After (split) |
|---|---|---|
| “How often do you exercise and track your meals?” (Never / Rarely / Sometimes / Often / Always) | These are two distinct behaviors with independent frequencies — someone can exercise daily and never track meals. A single frequency answer cannot represent both rates at once. | Item 1: “How often do you exercise?” Item 2: “How often do you track your meals?” |
| “How often do you contact support by phone or email?” | If the two channels are meant to be understood separately (e.g., to inform staffing by channel), collapsing them into one frequency question loses exactly the information the survey needs. If the intent is genuinely channel-agnostic — total contact frequency regardless of method — this is not double-barrelled, which is why the and/or test has to be applied with the actual analysis plan in mind, not on the sentence alone. | Item 1: “How often do you contact support by phone?” Item 2: “How often do you contact support by email?” (Or, if total volume genuinely is the construct of interest, keep the item as a single “how often do you contact support, by any method” question and say so explicitly.) |
The Sibling Pitfalls: A One-Pass Screening Checklist
Double-barrelling rarely travels alone. A draft instrument benefits from running all of the following checks in a single pass over every item, since a rewrite that fixes one often introduces or reveals another.
- Double-barrelled (this guide). The item asks two independently-answerable things and provides one response. Test: the and/or test above.
- Double negatives. The item requires the respondent to negate a negative to answer correctly — “Do you disagree that the policy should not be changed?” A respondent who wants the policy changed has to parse two negations correctly before selecting “disagree,” which is itself a double negative relative to their actual view. Test: read the item aloud and count negation words (“not,” “never,” “disagree,” “un-”/“dis-” prefixes); two or more in one item is a rewrite trigger.
- Presupposition (loaded questions). The item assumes a fact not yet established — “How satisfied are you with the improved checkout process?” presupposes the process improved. A respondent who thinks it got worse still has to answer a satisfaction question built on a premise they reject. Test: ask whether the item would still make sense to a respondent who disagrees with an unstated claim buried in its wording; if not, isolate and remove the presupposition, or ask about it directly first.
- Leading wording. A close cousin of presupposition: the item’s phrasing signals which answer is socially or organizationally preferred (“Don’t you agree that our support team resolved your issue quickly?”). See response bias for the fuller taxonomy of wording effects, including social desirability bias, which compounds leading wording rather than substituting for it.
Running this checklist against a draft instrument — ideally alongside a small cognitive-pretest or expert-review pass, not as a substitute for one — catches the class of wording defect that a pilot-test sample size is too small to detect statistically but that still systematically distorts every response it touches.
Why It Matters for Data Quality, Not Just Style
A double-barrelled item does not fail loudly. Respondents answer it; the survey platform records a valid-looking response; the analysis proceeds. The damage is that the resulting variable no longer measures a single, well-defined construct — which undermines construct validity and depresses measured reliability, because different respondents are, in effect, answering different questions while producing responses that look identical in the dataset. This is exactly the kind of instrument-level defect that questionnaire design best practice addresses through iterative drafting and cognitive pretesting before an instrument goes to field, and that a broader survey of survey question types and formats can help identify by format-specific failure pattern, as in the three formats worked through above.
Frequently Asked Questions
What is a double-barrelled question?
A double-barrelled question is a single survey item that asks about two (or more) distinct things but provides only one response option, forcing the respondent to either answer just one part, silently average across both, or guess which part the researcher actually cares about. “Is the product affordable and reliable?” is a standard example — affordability and reliability are separable dimensions that a respondent could rate very differently.
How do you identify a double-barrelled question?
Apply the and/or test: scan the item for conjunctions (explicit “and”/“or,” or an implicit one hidden in a list or compound phrase), then split the item at each conjunction and ask whether a realistic respondent could answer the two halves differently. If yes, the item is double-barrelled and should be split into separate items.
Is every question with “and” in it double-barrelled?
No. A question is only double-barrelled if the two halves joined by the conjunction are independently answerable. “Are you a full-time or part-time employee?” contains “or” but asks a single mutually-exclusive classification question, not two separable claims, so it passes the and/or test.
How do you fix a double-barrelled question?
Split it into separate items, one per independent claim or behavior, each with its own response scale. Where the two elements genuinely need to be understood together as a single combined construct (e.g., overall contact volume across channels), keep it as one item but state that combined framing explicitly, so the response is interpretable as intended rather than ambiguous by omission.
What other question defects commonly appear alongside double-barrelling?
Double negatives (requiring the respondent to un-negate a negative statement), presupposition or loaded questions (assuming a fact the respondent may not accept), and leading wording that signals a preferred answer. A single screening pass checking for all four at once, before an instrument goes to field, catches defects that a pilot sample is often too small to detect through statistics alone.
