Question Types in Surveys: 10 Formats and When to Use Them

Two questions can target the same thing and still return data that will not combine. “How satisfied are you with your degree program?” with five graded options beside it collects a number. “What would you change about it?” with an empty box underneath collects a sentence. Only the question type differs, and yet nothing converts one result into the other.

Question types are the structural forms a question uses to collect its answer: whether respondents pick from a fixed list, place themselves on a graded scale, rank several items against each other, or write freely in their own words. The choice also settles whether numbers or text land in your data file. By the end of this guide you will know the common question types, see an example of each, and know which one matches which research goal.

📌 Key takeaways

• A question type sets how an answer is collected, not what is asked.
• Response format and function in the questionnaire are two separate decisions.
• Single-choice, scale, matrix, ranking and open text cover almost every need.
• Forced-choice batteries produce higher and more accurate reporting than select-all-that-apply lists.
• Closed-ended answer lists do not just record opinion, they help shape it.

What is a question type?

A question type is the structural form of an item in a questionnaire, meaning the combination of question wording and prescribed response format. The question type fixes the shape an answer is allowed to take, and therefore the shape that answer has when you open your data file.

Question wording states what is being asked. Response format states how the answer may be given. Together the two produce the item a respondent actually sees on screen. The same underlying question about satisfaction with campus dining can be posed as a scale from 1 to 5, as a choice among four labeled categories, or as an empty text box, and each of the three measures something different.

Several words circulate for one idea. Question type, question format and item format are interchangeable in everyday use. Response format is narrower and names only the answer side, without the question wording. Level of measurement is a different concept again, because level of measurement describes which calculations your data permit and follows from the format you picked. Which arithmetic your level of measurement allows is therefore settled the moment you choose a question type.

Why the distinction matters beyond vocabulary becomes obvious during analysis. Satisfaction collected as a choice among “good”, “fair” and “poor” yields no meaningful average, even though a spreadsheet will happily compute one from the coded values 1, 2 and 3.

Response format and function: the two axes

Question types resist a single flat list, because two independent properties are in play: the response format of an item and the job that item does inside the questionnaire. Most overviews merge the two and set a screening question beside a matrix question as though you had to choose between them.

Treating the two as crossing axes works better. A screening question can be posed as a yes/no item or as a choice among six categories, and a demographic question about age can be a numeric field or a list of brackets. Once the axes are separated, the decision gets easier, because you make two small choices instead of one large one.

Three response formats, not two

Survey methodology offers a sturdier split than the familiar open versus closed pair. Lenzner and Menold (2015) separate three groups of response formats in the GESIS Survey Guidelines on question wording: open-ended questions, closed-ended questions with a rating scale, and closed-ended questions without a rating scale.

Consequences follow directly from that third category. On a rating scale the order of the points is fixed, since the order itself carries meaning, and the number of points becomes a separate methodological decision. In an unordered list, such as the majors offered by a college, order is free, and precisely for that reason the position of an option starts to influence how often it gets picked.

What a question does in the questionnaire

Function describes the task an item performs in the flow. A substantive question collects the content you are actually after, a screening question decides the route through the instrument, a demographic question supplies the characteristics for later subgroup analysis, and an attention check tests whether somebody is still reading carefully.

Function mostly determines where an item sits, not how it is answered. For building a sensible order from opening to close, see the guide on how to create a questionnaire.

Question types in a questionnaire: the overview

Ten question types cover nearly everything a questionnaire needs, from a class project to a customer study. The list below is sorted by response format, starting with selection questions and ending with free input.

Few questionnaires need all ten. A good instrument usually gets by with three to five types, and variety then appears on its own, because different content calls for different formats.

Two notes on the list. A yes/no question is strictly a single-choice question with exactly two options, and appears separately only because of the distinct role it plays in questionnaires. An NPS question is a rating scale with fixed labeling and a fixed scoring rule, so the NPS question is a convention rather than a new format.

Selection questions: single-choice and multiple-choice

Selection questions put a list of possible answers in front of respondents to pick from. The only difference between the two variants is how many boxes may be checked, and that small setting changes the analysis completely.

In a single-choice question the shares add up to 100 percent of respondents, because every person counts exactly once. In a multiple-choice question they do not, since one person can check three boxes and appear in three categories. Reporting that overlooks the difference produces percentages summing past 200 percent of respondents, which tells a reader nothing.

Single-choice: exactly one answer

A single-choice question fits wherever the categories genuinely exclude each other. Degree sought is a clean example, since an associate, bachelor’s, master’s and doctoral program cannot all apply at the same point in time. What matters is that the list really covers every case, otherwise respondents settle for a category that does not fit them. The article on single-choice questions shows how to build exhaustive lists.

Multiple-choice: several answers at once

A multiple-choice question fits where several answers can be true at once, such as modes of transportation used or campus events attended. The format is comfortable for respondents, and comfort is exactly what tempts researchers to let the list grow. As a rule of thumb, anything beyond roughly ten options gets skimmed rather than read. The article on multiple-choice questions covers analysis and the usual pitfalls.

Why the answer list shapes the result

How a category is defined moves the numbers as much as the topic does. According to the Pew Research Center, one half of the sample in a January 2002 poll was asked whether it was more important for President Bush to focus on domestic policy or foreign policy, and 52 percent of that half chose domestic policy against 34 percent for foreign policy. When “foreign policy” was narrowed to “the war on terrorism”, the picture reversed: 33 percent chose domestic policy and 52 percent chose the war on terrorism.

Position in the list matters as well, and it pulls in opposite directions depending on mode. The same Pew methods guide on writing survey questions reports that telephone respondents more often choose items heard later in a list, a recency effect, while respondents completing a survey themselves tend to choose items at the top, a primacy effect. For online studies the practical answer is randomizing the order of options per respondent, which Pew programs into many of its own lists.

Scale questions: rating, Likert and NPS

Scale questions supply a graded dimension on which respondents place themselves. Scale questions therefore capture not only whether something applies but how strongly, which makes them the most used format in satisfaction and attitude research.

What all scale questions share is a fixed order of points, an order that carries meaning by itself. What separates them is what they measure and what you may legitimately compute afterward.

The rating scale as the base form

A rating scale measures one single property across several graded points, for example satisfaction from “very satisfied” to “very dissatisfied”. The decisions that matter are the number of points, the labeling, and whether a neutral midpoint exists at all. Those three settings move results more than the question wording does, which is why the article on the rating scale treats them at length.

The Likert scale for attitudes

A Likert item presents a statement that respondents agree or disagree with in steps, most often five points from “strongly agree” to “strongly disagree”. Several such statements on one topic form a measurement instrument. Whether the steps are equally spaced, and whether averages are therefore permitted, is a much weaker claim than most textbooks suggest, and the article on the Likert scale explains why.

Agree-disagree wording carries a second cost that is easy to miss. According to the Pew Research Center, less educated and less informed respondents show a greater tendency to agree with a statement regardless of its content, an effect known as acquiescence bias, and the tendency grows stronger when an interviewer is present rather than when people answer on their own. Offering a choice between two alternative statements is the better practice where a topic allows it.

The NPS question as a fixed convention

The Net Promoter Score question asks a single recommendation question on a scale from 0 to 10 and condenses the answers into one metric by a fixed rule. Comparability across organizations and across time is the advantage. The drawback is the one every single-number metric carries: an NPS value reports where you stand without reporting why, which is why an open follow-up question almost always sits behind it.

💡 Tip

Stay with one scale direction and one number of points across a whole questionnaire. Running five points from positive to negative in the first block and seven points the other way round in the second collects answers that look like data but partly measure how quickly respondents adjusted to the switch.

Matrix and ranking questions

Matrix and ranking questions bundle several judgments into one item. Both save considerable space and time, and both demand noticeably more concentration from respondents than a plain selection question does.

Each of the two has its own trap, and in both cases the trap is length. A matrix with five rows gets filled in carefully, a matrix with twenty rows gets clicked through.

The matrix question for item batteries

A matrix question stacks several statements underneath each other and rates them all on one shared scale. The question wording appears once above, the scale once alongside, and respondents work down row by row. For statements that belong together, no format is more efficient, because respondents build the rating standard only once and then keep it for every row.

Forced-choice beats select-all-that-apply

Two constructions exist for lists where several answers can apply, and the two are not equivalent. A select-all-that-apply list lets people tick everything that fits in a single view, while a forced-choice battery asks each row separately with its own yes and no. According to the Pew Research Center (2019), the second construction is the more accurate one. In a randomized experiment among 4,581 U.S. adults, fielded July 30 to August 12, 2018, reporting rates were higher under forced choice for all twelve items tested, by an average of 8 percentage points. Being denied coverage by a health insurance company, for instance, was reported by 19 percent of the forced-choice group against 13 percent of the select-all group. The full report on forced-choice and select-all-that-apply formats sets out every item.

Order effects point the same way. Across 21 panel surveys run between November 2015 and August 2018, Pew measured an average order effect of 3 percentage points for select-all-that-apply questions against 0 percentage points for forced-choice questions, and has since adopted a policy of using forced choice in its online surveys wherever possible. Practically speaking: if you want defensible shares from a multiple-answer list, build the list as a matrix with yes and no columns rather than as a row of checkboxes. The price is a few extra seconds per question.

The ranking question for real priorities

A ranking question asks respondents to put objects in an order instead of rating each one separately. The format forces a decision that a scale never forces: somebody who rates five criteria all as “very important” has said nothing, while somebody who ranks them has. The practical ceiling sits around five to seven objects, above which sorting turns into busywork and the lower places get guessed.

⚠️ Caution

Matrix questions break first on narrow screens. Since a large share of responses arrives from phones, treat five answer columns and eight rows as the upper limit. Beyond that, split the matrix into two questions instead of shrinking the type.

Open-ended and semi-open questions

Open-ended questions supply no answer options at all, only the question wording and an empty field. Open-ended questions are the one format that returns content nobody anticipated while building the questionnaire, and that is where their value sits.

How much the format matters shows up most clearly in a split-half experiment. According to the Pew Research Center, a poll conducted after the 2008 presidential election asked one question in two versions, closed-ended with five options for one half of the sample and open-ended for the other: “What one issue mattered most to you in deciding how you voted for president?” Offered the economy explicitly, 58 percent of the closed-ended group chose it, while only 35 percent of the open-ended group named the economy unprompted. And 43 percent of the open-ended group gave an answer absent from the closed-ended list, against 8 percent of the closed-ended group who volunteered something outside the five options read to them. The methods guide on writing survey questions lays out the comparison in full.

Worth reading twice: a closed list does not simply record an opinion, a closed list helps produce one, because respondents reason with the options in front of them rather than the ones in their heads. Running a small open-ended pilot first and building the answer list from what people actually say is the standard remedy, and Pew describes exactly that sequence.

When an open-ended question earns its cost

Open-ended questions pay off in four situations above all: when you are testing knowledge and want to rule out guessing, when you do not yet know the range of possible answers, when a complete list would be unmanageable, as with occupations, and when you want to avoid steering respondents with options. The price is coding effort, since text has to be categorized before it can be counted. The article on open-ended questions walks through that coding step.

The semi-open question as a hybrid

A semi-open question offers a fixed list and adds a free field, usually labeled “other, please specify”. The hybrid is the normal case for any substantive question whose answers cannot be listed exhaustively, and it prevents the damage caused by leaving that category out: when an obvious answer is missing, the item is either skipped or something arbitrary gets checked. The comparison of all three forms sits in the article on open and closed questions.

How box size steers answer length

The size of a text box influences how much people write, and almost nobody uses that deliberately. Züll (2015) points to studies in the GESIS Survey Guidelines on open-ended questions showing that in self-administered questionnaires the field size provided correlates positively with the length and detail of answers. A small box signals that a keyword will do, a large one invites full sentences. Want keywords, make the box small and state the expected form.

Screening questions and demographic questions

Screening questions and demographic questions belong to the second axis from the third chapter: both are defined by their function in the questionnaire, not by their response format. Neither usually collects the thing you actually care about, and hardly any questionnaire works without them.

Take a study on housing costs. Whether somebody can report the monthly assessment on a condo depends on whether they own one at all, and that prior item decides the route through everything that follows.

Screening questions route respondents

A screening question decides which items come next. The mechanism protects respondents from having to comment on circumstances that do not apply to them, and protects you from answers nobody can interpret. Lenzner and Menold (2015) flag a difference that is easy to overlook in planning: computer-assisted surveys handle branching without trouble for respondents, while mail questionnaires should avoid numerous or complex filters entirely.

Demographic questions supply the comparison groups

Demographic questions collect characteristics such as age, gender, education or region. The purpose is not the value itself but later analysis by subgroup, since without those characteristics you cannot show that first-year students answer differently from seniors. Placement belongs near the end, because demographic items motivate least of all. According to the Pew Research Center, income, education and age should not appear near the start of a survey unless they are needed to establish eligibility or to route respondents. The article on demographic questions collects wording that has proven itself.

Restraint has a second reason on campus. Human subjects research at a US college normally passes an institutional review board before fielding, and the demographic block is the part reviewers examine most closely, because a narrow combination of major, class year and country of origin can single out one student in a small program. Every characteristic you drop is one fewer route to reidentification.

Attention checks and the closing question

Attention checks test whether somebody is still answering carefully, typically by requesting a specific response at one point in the questionnaire. Attention checks make sense in long instruments and are dead weight in short ones. The closing question, by contrast, is almost always open-ended and asks whether there is anything else the person wants to add. Züll (2015) describes that closing item explicitly as a motivational device, since it hands respondents the floor after a long run of prescribed options.

Which question type fits which goal?

The choice of question type follows from the statement you want to make at the end, not the other way around. Anyone who knows which sentence should appear in the results section has usually made the decision already without noticing.

The table below reads from left to right: find your goal in the first column and read off the matching format and the reason. Where several goals apply, the row closest to your research question wins.

One rule stands above every row: analysis first, question type second. Writing down in advance which table or which chart should exist at the end shows immediately whether the planned item can even produce it. Ten minutes of planning here will save you an entire afternoon later.

Common mistakes when choosing a question type

Most problems in questionnaires come not from poor question wording but from a question type that does not match the content. Four of them turn up again and again, and all four can be fixed in minutes before anything goes out.

What the four have in common is timing: they surface during analysis, and by then nothing can be done. A questionnaire that has already reached two hundred people cannot be corrected.

Mistake 1: the options do not match the question

Lenzner and Menold (2015) give an example you will recognize everywhere afterward. A question asks whether somebody was an active member of a club during the past twelve months, and the options underneath read “often”, “sometimes”, “rarely” and “never”. The wording calls for a yes or a no, the options call for a frequency, and respondents are left guessing. Either the wording or the options must change, and not both at once.

Mistake 2: the categories overlap

Answer categories have to be exhaustive and mutually exclusive, otherwise an answer cannot be assigned unambiguously. Time and age brackets are the classic case: somebody who gets up at exactly 6:30 can pick both “between 6:00 and 6:30” and “between 6:30 and 7:00”. The fix is unglamorous and still rarely applied, namely drawing the boundaries at 6:29 and 6:59. Further rules sit in the article on closed-ended questions.

Mistake 3: the question asks two things at once

Double-barreled questions ask respondents to judge more than one concept in a single item, and the answers cannot be interpreted afterward. The Pew Research Center uses the example “How much confidence do you have in President Obama to handle domestic and foreign policy?”, where an answer could refer to either half or to some blend of the two. Splitting the item into two questions costs one extra line and rescues the whole measurement.

Mistake 4: too many open-ended fields, or a scale that shifts

Open-ended questions cost respondents the most effort, and the effort accumulates: a questionnaire with eight text fields gets abandoned more often than one with two, and produces volumes of text nobody codes. A shifting scale is the quietest of the four mistakes and one of the most expensive. Four points in the first block, five in the second and a reversed direction in the third leaves answers that no longer compare, and part of your sample will read straight past the switch.

Conclusion

Question type is not a formatting detail settled at the end of writing but the first decision about which statements your questionnaire will ever support. Keeping the two axes in view, response format and function, splits that decision into two manageable steps. And writing down the table you want to end up with usually reveals the right format on its own.

What the American evidence adds is a warning against false comfort. Both Pew experiments make the same point from different angles: a format does not merely record an answer, a format helps produce one. Neither the 58 percent of the closed-ended group choosing the economy nor the extra 8 percentage points under forced choice came from anyone changing their mind.

Choose the question type while planning the analysis, not while writing the question.

Where to go next

Four follow-ups cover the questions that usually come next, in the order they tend to come up: building the instrument, choosing between open and closed formats, borrowing tested wording, and sorting out the vocabulary around surveys and questionnaires. Each one stands on its own, so pick whichever matches where you are.

• Building your instrument right now? How to create a questionnaire: steps, structure, example
• Weighing open against closed formats? Open and closed questions compared
• Looking for wording you can borrow? Questionnaire templates and examples
• Unsure how a survey differs from a questionnaire? Survey and questionnaire: the differences

Want to try every question type once?

With empirio.ai, an online survey tool from Germany, you can put single-choice, scales, matrix, ranking, NPS and open text into one questionnaire and watch the results come in live. Create a survey for free