Survey Question Types: Closed, Open, Scales and When to Use Each

The families are not interchangeable. A five-point agreement scale cannot tell you why someone agreed. An open question cannot give you a quota cell. Most studies need two or three families together: closed questions for the numbers, one or two open questions for the language, and a task if the decision involves a trade-off.

Dichotomous questions offer two answers, the yes or no survey question being the common form: bought or did not buy, aware or not aware. They are fast on a phone and easy to analyse. Their weakness is that a forced two-way choice hides the middle. Use them for facts, not opinions.

Multiple choice questions, often written MCQ question type in a questionnaire spec, give one answer from a list. Checklists allow several answers. The difference matters for analysis: a checklist gives a percentage per option that sums past 100, while a multiple choice gives shares that sum to 100.

Three rules cover most failures here:

1. Make the list complete. Always offer "None of these", "Do not know" or "Other" where a person could reasonably need it. A missing option becomes a wrong answer.
2. Keep options mutually exclusive. "Once a week" and "two to three times a month" overlap for some readers.
3. Keep lists short on a phone. Six or seven options fit a screen; twenty do not. If the list is long, add a search or split the question.

An open question is the right tool when you cannot list the answers yet: why a brand was rejected, what a pack message meant, which word people use for a category. It is the wrong tool when you need a percentage you can defend, because a coded open answer depends on the code frame.

Two practical points for Indian fieldwork. First, phone keyboards make long answers unlikely, so ask one thing per box and keep the prompt short. Second, answers arrive in mixed languages and scripts, so plan how they will be read. Our guide to open-ended analysis covers coding and translation.

If you need a number and the reason, ask both: a closed question for the number, then an open follow-up of the people who gave a particular answer.

A Likert item gives a statement and a set of agreement levels, usually five or seven, all labelled. A numeric scale asks for a number, most often 0 to 10, and is the basis of NPS.

Agree or disagree statements carry a known bias. Pew Research describes an acquiescence bias, a tendency to agree with statements that runs stronger among less educated respondents. If your statement is "The pack is easy to open", some people will agree because agreeing is easier than disagreeing. A question that asks how easy or difficult it was, with a labelled scale, removes that pressure.

Three design choices change the score, so keep them fixed across waves: the number of points, whether there is a midpoint, and whether every point has a label. A four-point scale with no midpoint forces a side. A seven-point scale with only the ends labelled invites interpretation. For a longer treatment, see the Likert scale guide.

Ranking asks a respondent to order items. It works up to about five or seven items. Beyond that, the bottom of the list is noise.

Grids put many items on one scale, which is efficient and also the fastest way to produce straight-lining, where a respondent taps the same column down the page. Use grids for a short, related list and check for straight-liners afterwards.

MaxDiff shows sets of items and asks for the best and worst in each set. It handles twenty or more features and produces a preference score per item. It needs a balanced design, so build the sets in a tool rather than by hand. The MaxDiff guide covers the design.

Where the decision is about combinations, such as price with pack size with flavour, a rating cannot recover the trade-off. That needs a conjoint design, covered in the conjoint guide.

Run every question through this list before you field it:

• One idea per question. "How satisfied are you with the price and the taste?" has two answers.
• Plain words. If a respondent would not say it aloud, rewrite it.
• No loaded adjectives. Premium, healthy, innovative and advanced all push the answer.
• A named time frame. Replace "recently" and "often" with "in the last 30 days".
• A complete option list, including the escape options named above.
• Indian context. Rupees, local brands, city names, and no US address or zip code fields.
• A defined term on first use. If you must ask about a category, name the products in it.

Two more checks catch problems that wording alone cannot. Read the pilot's answer distribution: an option chosen by almost nobody may be missing a better answer. And ask someone who did not write the questionnaire to read it on a phone, in the language it will be fielded in. The survey bias guide covers the systematic errors behind these checks.

Mobile app survey questions follow the same families with tighter limits. A phone screen fits one short question above the fold, so a five-item grid becomes five screens and a twenty-option list becomes a search field.

Three rules for application survey questions: keep a rating to five points so the labels stay visible, replace long checklists with search or a shortlist, and ask one open question at most. If the survey runs inside the product, trigger it after the action you are measuring rather than on first open, and keep the whole thing under two minutes.

The same question set that works on web will usually fail in an app unless it is shortened. Test it on a small phone before the wave, not on a desktop preview.

The set mixes families on purpose: a checklist for repertoire, a scale for purchase intent, an open question for language, and a task for shelf behaviour. Each question feeds a different table in the report.

The survey builder supports single and multiple choice, ratings, grids, ranking, open text, yes or no, MaxDiff and image stimuli, so the set above can be built without leaving the platform. Routing handles screen-outs and carry-forward. Surveys run in as many Indian languages as the study needs.

The free plan includes 30 responses, enough to pilot the question set before you pay for a wave. Pricing is per response: ₹150 on Plus from 100 responses, ₹120 on Pro in bundles of 400 to 1,000. Results come back as cross-tabs, significance tests and coded open answers. If you want the questionnaire reviewed before fielding, the quality checks describe what happens to responses afterwards.