Back
Multiple Choice Questions: Types, Examples and Best Practices
SmartInterview Team
A multiple choice question gives respondents a fixed list of answer options to pick from. Single-select means they choose exactly one. Multi-select means they can choose several. A good one follows two rules: the option list must cover every plausible answer, and no two options can be true at the same time.
Everything else is craft. Here is what that craft comes down to.
Use radio buttons for single-select and checkboxes for multi-select. The control tells people how many answers you want before they read a word of instruction.
Exhaustive means nobody is forced into an answer that is false for them. Mutually exclusive means no respondent can legitimately belong in two boxes.
Randomize option order to cancel position bias, but anchor "Other", "None of the above", "Don't know", and anything with a natural order.
Multi-select percentages sum to more than 100%. Decide up front whether your denominator is respondents or responses, and label the chart with it.
Long lists are a design smell. Past roughly 7 to 10 visible options, switch to a grouped list, a search field, or a two-step funnel question.
Multiple choice vs the alternatives
Multiple choice is the default for a reason, but it is not always the right instrument. Here is how it sits next to the closed-ended formats it competes with.
Format | What it measures | Respondent effort | Analysis output | Best used for |
|---|---|---|---|---|
Single-select multiple choice | One categorical choice: the single best fit | Low. One decision, list visible | Clean frequency table, shares sum to 100% | Segmentation, screeners, "which one" questions |
Multi-select multiple choice | Membership in a set: everything that applies | Medium. Every option needs its own yes or no judgment | One binary variable per option, shares exceed 100% | Brand awareness, feature usage, reasons and barriers |
Dropdown | One choice from a long, familiar, ordered list | High. Options are hidden until opened | Same as single-select, with more non-response | Country, year, canton, currency. Never attitudes |
Rank order | Relative priority between options | High. Every item is compared to every other | Mean ranks and top-box shares, harder to read | Trade-offs where "all of these matter" is the real answer |
Open text or voice | Unprompted, in the respondent's own frame | Highest. Nothing to recognize, everything to recall | Needs coding before it becomes countable | Discovery, reasons behind a choice, anything you might not have thought of |
The honest trade-off: multiple choice buys you speed and comparability by handing respondents your categories. Open questions keep their categories but cost you coding work. That is the same tension covered in qualitative vs quantitative research, and most solid questionnaires use both.
Single-select vs multi-select
These are two different measurements, not two settings of the same question.
Single-select forces a choice. It asks which one thing is truest, and it produces a clean distribution where the shares sum to 100%. Use it when the options genuinely compete: primary bank, main reason for leaving, preferred delivery slot. If you want the mechanics of forced-choice design, see single select questions.
Multi-select measures set membership. Each option becomes its own yes-or-no variable behind the scenes. Use it when several answers can be true at once: brands you have heard of, features you use, channels you saw an ad on.
The interface has to signal which one it is
Respondents read the controls before they read the instructions. Round radio buttons mean pick one. Square checkboxes mean pick as many as apply. Mixing them up produces silent damage: people who wanted to select three options select one, and you never see the answers you lost.
Practical rules:
Radio buttons for single-select, checkboxes for multi-select. No exceptions, no custom widgets that look like neither.
Put the instruction in the stem, not below the list: "Select all that apply" or "Select one".
If you cap selections, say the cap in the stem ("Select up to three") and enforce it in the interface.
On a phone, make the whole option row tappable, not just the small control.
The two rules every option list must pass
Exhaustive: every respondent can find themselves
If a respondent's true answer is not on the list, they do one of three things: pick something close but wrong, pick at random, or abandon. All three corrupt your data, and only the third is visible to you.
Common ways lists fail to be exhaustive:
Asking "Which of our plans are you on?" with no "I'm not sure" for people who genuinely do not know.
Listing competitor brands you care about and omitting the regional ones your respondents actually use.
Asking about employment status with Employed, Unemployed, and Retired, and nothing for students, carers, or self-employed people.
Offering demographic categories that assume a respondent fits one box when they do not.
The fix is either a complete list or a well-placed "Other (please specify)". More on that below.
Mutually exclusive: no respondent belongs in two boxes
This is the rule that gets broken most often, and almost always in the same place: numeric bands.
Broken | Why it breaks | Fixed |
|---|---|---|
18-25, 25-35, 35-50 | A 25 year old fits two bands. Shares depend on which one they happen to click | 18-24, 25-34, 35-49, 50 or older |
Under CHF 50, CHF 50-100, CHF 100+ | Both boundaries are double-counted | Less than CHF 50, CHF 50 to CHF 99, CHF 100 or more |
Price / Value for money / Cost | Three labels for one concept, so the vote splits three ways | Price, plus a separate option for something genuinely different |
Daily, Weekly, Regularly | "Regularly" overlaps both of the others | Daily or more, A few times a week, Weekly, Less than weekly, Never |
Overlap is not just untidy. It splits a real signal across two options, so a category that should have been your top finding shows up in third and fourth place. In single-select questions the damage is invisible in the output: the numbers still sum to 100% and look perfectly healthy.
Note that the mutual exclusivity rule applies to the options, not to the respondent's behavior in a multi-select. In multi-select, picking several options is the point. What must still be exclusive is the concepts: two options should never mean the same thing.
Writing the option set
Once the list is exhaustive and non-overlapping, the wording decides how much bias you introduce.
Keep the grammar parallel
Every option should be the same part of speech and follow the same shape as the stem. If the stem is "Why did you cancel?", every option should be a reason: "It cost too much", "I stopped using it", "The support was slow". Mixing a noun ("Price") with a full sentence ("I found that the support team took too long to respond") makes respondents slow down and re-read.
Keep the lengths similar
Longer options attract more selections in a scanned list, simply because they carry more detail and more chances to match something the respondent recognizes. If one option needs 20 words and the rest need three, shorten it or split it.
One idea per option
"Fast and affordable" is double-barreled. Someone who found it fast but expensive cannot answer honestly, and if they select it you cannot tell which half they meant. Split it into two options, or pick the one you actually want to measure. The same trap applies to question stems, and it is one of the classic errors in customer satisfaction survey questions.
No internal jargon
Options like "Tier 2 onboarding" or "post-purchase NPS flow" mean something in your team's Slack, not to a respondent. Write the option in the words a customer would use, even when it costs you precision.
Do not lead with the list
An option set is itself a form of priming. If you ask "What did you like about the service?" and every option is a positive attribute, you have taught the respondent that liking it is expected. Balance the set, or ask a neutral open question first and let people answer before they see your categories.
How many options is too many
There is no hard limit, but there is a practical one. Respondents scan lists rather than read them, and the longer the list, the more they satisfice: they read the first few options, find something acceptable, and stop. On a phone, anything below the fold is read by fewer people than anything above it.
As a working recommendation, keep a visible flat list to roughly 7 to 10 options. Past that, restructure instead of scrolling:
Group the list. Category headers turn 40 options into 6 scannable blocks. Works well for product catalogs and job roles.
Use search or autocomplete. For long, well-known, closed lists where people know their answer before they arrive: country, city, university, car model. Never for attitudes, since respondents cannot search for a feeling.
Split into a two-step funnel. Ask the broad category first, then show only the options inside it. This is the cleanest option for large brand or product lists, and it shortens the perceived question for everyone.
Cut the list. Often the honest answer. If 15 of your 30 options will each land under 2%, you are paying respondent attention for noise.
Length has a cost beyond bias. Every extra scroll is another chance for someone to drop out, which is part of why survey response rates keep falling.
Order effects, randomization, and what to anchor
Where an option sits in the list changes how often it gets picked. This is a real, well-documented effect in survey methodology, and it works in opposite directions depending on how the question is delivered.
Primacy and recency
Primacy bias appears in visual, self-administered surveys: web, mobile, paper. Respondents process the list from the top, and the early options get more attention and more selections.
Recency bias appears when options are read aloud: phone interviews, in-person interviews, voice surveys. The last options read are the freshest in memory when the respondent answers.
Both effects grow with list length and with respondent fatigue. Neither disappears because your options are well written.
Randomization and rotation
You cannot remove order effects, but you can stop them from landing on the same option every time. Randomizing the order per respondent spreads the advantage evenly across the list, so the bias becomes noise instead of a systematic tilt toward option one.
Full randomization shuffles options independently for each respondent. The default choice for unordered lists such as brands, features, or reasons.
Rotation keeps the list in a fixed sequence but changes the starting point per respondent. Useful when the list has an internal logic you want to preserve.
Block randomization shuffles groups of related options while keeping each group together. The right tool for categorized long lists.
Record the order each respondent saw. Without it you cannot check afterwards whether the randomization actually worked, and you cannot test for order effects in your own data.
The options you must anchor
Randomization applies only to the substantive options. Some options must stay pinned in place and be excluded from the shuffle:
"Other (please specify)": always last. It is a fallback, and floating it into the middle makes it read like a real category.
"None of the above": always last, after Other if both are present.
"Don't know" and "Prefer not to say": always at the bottom, visually separated from the substantive list where the interface allows it.
Ordered scales: never randomize. Agreement, frequency, and satisfaction scales carry meaning in their sequence. Shuffling a Likert scale makes it unreadable. If you want to counterbalance a scale, reverse the whole thing for half your sample rather than scrambling it.
Any naturally ordered list: time periods, age bands, income bands, sizes, days of the week. If a respondent would notice the order being wrong, it is ordered.
"Other", "None", and "Don't know" are three different things
Other (please specify)
"Other" is your insurance policy against a list that is not exhaustive. It keeps respondents honest instead of forcing a false answer, and it tells you what you left out.
What it costs you: every free-text entry has to be read and coded before it can be counted. That is real work, and it is why "Other" responses so often sit uncoded in the export while everyone reports the closed options as if they were the whole story.
How to use it well:
Run it in the pilot on purpose. Field the question to a small sample with "Other" open, read what comes back, and promote the recurring answers into named options for the main wave. That is what a soft launch is for.
Watch the size of the Other bucket. If it is large, your list is incomplete, and no amount of analysis on the named options fixes that.
Back-code before you report. Many "Other" entries are just a named option written differently. Merge those in and report the genuinely new categories separately.
Do not add it reflexively. For a genuinely closed list ("Did you use the app or the website?") it adds noise rather than coverage.
The coding cost is the reason many teams quietly avoid open-ended options. It does not have to be manual work: platforms that code open responses automatically, SmartInterview among them, will cluster "Other" text and free-text answers into themes so you can report them next to the closed options instead of leaving them in a spreadsheet.
None of the above
"None of the above" means the list does not apply to this respondent. It belongs in almost every multi-select, because without it the only way to say "none of these" is to leave the question blank, which is indistinguishable from skipping.
It must be exclusive. Selecting it should clear every other checkbox, and selecting any other option should clear it. If your tool does not enforce this, you will find respondents who checked four brands and also "None of the above", and you will have to decide what to do with them after the fact.
Don't know
"Don't know" is a real answer, not a refusal. It means the respondent lacks the information. Include it when the question asks about facts they may not have: pricing they did not choose, features they never opened, decisions someone else made in their household.
The trade-off is that offering it makes it easier to skip thinking. On attitude questions where everyone can form a view, it invites satisficing. On knowledge questions, omitting it manufactures false precision. Decide per question, not per survey.
Prefer not to say
This one is about consent, not knowledge. The respondent knows the answer and has chosen not to give it. Offer it on sensitive items: income, health, political preference, ethnicity, sexual orientation. Treat it as missing data, never as a category, and never merge it with "Don't know". They mean opposite things.
Survey multiple choice answers: designing the answer set
The question stem gets the attention. The answer options decide the data. A well-worded stem attached to a lopsided, overlapping, or badly ordered option list produces numbers that look clean and mean nothing. Treat the answer set as its own design object with its own checklist.
How many answer options to offer
There is no methodological constant here, so treat the following as practitioner guidance rather than a finding from a study. The useful way to think about it is what breaks at each extreme.
Answer set size | What happens | Where it shows up in the data | What to do |
|---|---|---|---|
2 to 3 options | Forced simplification. Real answers get flattened into the nearest available box | One option takes an implausible majority, and the "Other" bucket is large | Fine for genuinely binary facts. Otherwise split the dominant option into its real components |
4 to 7 options | The comfortable range. The whole list is visible on a phone without scrolling | Spread across options, low "Other", stable between waves | Default target for attitude, driver and frequency questions |
8 to 12 options | Workable but the bottom of the list starts losing attention, especially on mobile | Options below the fold underperform regardless of content | Randomize order, keep labels short, check the list fits one screen |
13 or more options | Satisficing. Respondents read until something acceptable appears, then stop | A long tail of options under 2%, and average selections per person falls | Group under headers, add search, use a two-step funnel, or cut the tail |
One more constraint that people forget: for scale questions the count is not a free choice. Odd numbers of points give a midpoint, even numbers force a side. Five and seven points are the common working defaults, and whichever you pick has to stay identical across every wave of a tracker.
Answer options must be mutually exclusive and exhaustive
These two rules are covered in depth earlier in this article, so here is the short version as it applies to the option list itself:
Mutually exclusive: no two options can be true for the same respondent at the same time. Read your list and ask whether any single person could honestly claim two entries. Numeric bands and near-synonyms like "price" and "value for money" are where this breaks.
Exhaustive: every respondent can find something true. If you cannot enumerate the full universe, an anchored "Other (please specify)" makes the set exhaustive by construction.
Parallel and comparable: same grammar, similar length, one idea each. An option that is twice as long as its neighbors is not competing on equal terms.
Ordering the answer options
Order is not cosmetic. It changes which options get picked. Four orderings cover almost every case, and picking the wrong one either introduces bias or makes the list unreadable.
Ordering | Use it when | Example | Risk if misapplied |
|---|---|---|---|
Scalar (fixed, low to high) | The options form a scale or a quantity | Satisfaction, agreement, frequency, age, income | Shuffling it makes the scale unreadable and the data unusable |
Logical (fixed, follows a real sequence) | The list mirrors a process, a timeline or a hierarchy | Stages of a purchase journey, days of the week, company size bands | Randomizing forces respondents to re-derive the order themselves |
Alphabetical (fixed) | Long, well-known lists where people arrive knowing their answer | Countries, cantons, languages, car makes | Systematically favors early letters if the list is not one people scan for a known answer |
Randomized (per respondent) | Unordered lists where every option competes with the others | Brands, features, reasons, barriers, channels | None, provided you anchor the special options and log the order shown |
The default decision rule: if a respondent would notice the order being wrong, keep it fixed. If they would not, randomize it. Either way, the special options at the bottom never move.
Special options, and which ones are missing data
The four options that live at the bottom of a list are not interchangeable, and the most consequential difference is that two of them are substantive answers and two of them are missing data.
Option | What it actually means | Substantive or missing? | Legitimate when |
|---|---|---|---|
Other (please specify) | The true answer exists but is not on your list | Substantive. Code the text, then report it | You cannot enumerate the full universe: brands, reasons, job titles, tools |
None of the above | A real "no" to every option listed | Substantive. Zero is a finding | Almost every multi-select, so that a blank is not the only way to say none |
Don't know | The respondent lacks the information to answer | Missing data. Exclude from the base, report the rate separately | Knowledge and factual questions where not knowing is plausible |
Prefer not to say | They know and have declined to tell you | Missing data. Never a category | Sensitive items: income, health, politics, ethnicity |
The practical consequence is in the denominator. "None of the above" stays in the base because it is an answer. "Don't know" and "Prefer not to say" normally come out of it, because including them shrinks every substantive percentage for a reason that has nothing to do with the topic. Whichever convention you pick, state it on the chart and keep it constant across waves.
Two failure modes worth naming. Merging "Don't know" with a neutral midpoint turns ignorance into a moderate opinion. Merging "Prefer not to say" with "Don't know" collapses a refusal and a knowledge gap into one meaningless bucket.
Balanced vs unbalanced answer sets
An answer set is balanced when the positive and negative options mirror each other in number and in intensity. It is unbalanced when one direction gets more room than the other. Unbalanced sets bias results because respondents read the option list as a signal of what a normal answer looks like, and because they simply have more ways to land on one side.
A balanced satisfaction set:
Very satisfied
Satisfied
Neither satisfied nor dissatisfied
Dissatisfied
Very dissatisfied
The same question, unbalanced:
Extremely satisfied
Very satisfied
Satisfied
Neutral
Dissatisfied
Three positive steps, one neutral, one negative. Someone mildly unhappy has exactly one box, while someone mildly happy has three. Top-two-box scores computed on the second set are not comparable to the first, and a set like this cannot be fixed in analysis because the underlying distinctions were never offered.
Unbalanced sets show up in three recognizable forms:
Numerically unbalanced: more positive options than negative ones, as above.
Verbally unbalanced: equal counts, but the labels are not symmetric. "Excellent, very good, good, fair, poor" is a positively tilted set pretending to be neutral, because four of its five labels are non-negative.
Content unbalanced: a driver or attribute list where every option is a benefit. If you ask what people think of a product and offer only positive attributes, you have told them the expected answer before they choose.
The check takes ten seconds: count the options on each side of the midpoint, then read the labels out loud and ask whether the negative words are as strong as the positive ones.
Answer set examples you can copy
Thirteen answer sets that pass the checks above, grouped by what they measure. Adjust the wording to your market, but keep the structure.
Demographics
1. Age. What is your age? Select one.
Under 18
18 to 24
25 to 34
35 to 44
45 to 54
55 to 64
65 or older
Prefer not to say
Bands do not touch, the top band is open-ended, and the fixed scalar order is never randomized.
2. Employment status. Which of these best describes your current situation? Select one.
Employed full time
Employed part time
Self-employed or freelance
Student
Not working and looking for work
Not working and not looking for work
Retired
Prefer not to say
The two "not working" options exist because the distinction matters and because a single "unemployed" box excludes carers and people who have stopped looking.
3. Household income. What is your total annual household income before tax? Select one.
Less than CHF 50,000
CHF 50,000 to CHF 74,999
CHF 75,000 to CHF 99,999
CHF 100,000 to CHF 149,999
CHF 150,000 or more
Don't know
Prefer not to say
Both missing-data options are present and separate, because in a household some people genuinely do not know the total and others simply will not say.
Frequency of use
4. Frequency. How often do you use the app? Select one.
Several times a day
About once a day
A few times a week
About once a week
A few times a month
Less often than monthly
Never
No vague labels. "Regularly" and "occasionally" mean different things to different people and overlap with everything around them.
5. Recency. When did you last use the app? Select one.
Today
In the past 7 days
In the past 30 days
In the past 3 months
More than 3 months ago
I have never used it
Recency and frequency answer different questions. Ask both when you need to separate heavy users from lapsed ones.
Satisfaction and agreement
6. Satisfaction. Overall, how satisfied are you with your most recent delivery? Select one.
Very satisfied
Satisfied
Neither satisfied nor dissatisfied
Dissatisfied
Very dissatisfied
7. Agreement. How much do you agree with this statement: the pricing is easy to understand. Select one.
Strongly agree
Agree
Neither agree nor disagree
Disagree
Strongly disagree
Don't know
Both sets are symmetric on either side of a labeled midpoint. "Don't know" sits outside the scale, never inside it, because not knowing is not a moderate opinion. For more question wording on this theme, see customer satisfaction survey questions.
8. Effort. How easy was it to resolve your issue? Select one.
Very easy
Easy
Neither easy nor difficult
Difficult
Very difficult
I did not contact support
The last option is a screen-out in disguise. Without it, people who never contacted support answer anyway and pollute the average.
Purchase drivers
9. Main driver, single-select. What was the single biggest reason you chose us? Select one.
Price
Product quality
Delivery speed
Range of choice
Recommendation from someone I know
Previous experience with the brand
Other (please specify)
Randomize the first six, anchor "Other" last. Note there is no "value for money" option, because it would split the price vote.
10. Drivers, capped multi-select. Which of these mattered in your decision? Select up to three.
Price
Product quality
Delivery speed
Returns policy
Customer reviews
Sustainability credentials
Loyalty scheme
Other (please specify)
None of these
The cap forces prioritization. Without one, respondents tick everything that sounds good and every option lands between 50% and 70%.
11. Barriers. What stopped you from completing your purchase? Select all that apply.
The total price was higher than I expected
Delivery cost too much
Delivery was too slow
The item I wanted was out of stock
I wanted to compare other options first
The checkout did not work properly
I was only browsing
Other (please specify)
"I was only browsing" is there because a large share of abandonment is not a problem to fix. Leaving it out inflates every genuine barrier.
Brand awareness
12. Aided awareness. Which of these brands have you heard of? Select all that apply.
Brand A
Brand B
Brand C
Brand D
Brand E
None of these
13. Consideration. Which of these would you consider buying next time? Select all that apply.
Brand A
Brand B
Brand C
Brand D
Brand E
None of these
Brand lists get randomized every time, "None of these" is anchored and exclusive, and the list must stay identical across waves. Adding a brand mid-tracker depresses every other brand's share and breaks the series. Ask the unaided question first if you want both, since showing the list contaminates recall for the rest of the interview.
Analyzing multiple choice data without misleading yourself
Percentages will sum above 100%, and that is fine
In a multi-select, each respondent contributes several selections. If 60% picked A, 45% picked B, and 30% picked C, the total is 135% and nothing is wrong. Say so on the chart. Someone will otherwise assume the data is broken, or worse, "fix" it by rescaling to 100% and destroying the meaning.
Decide what your denominator is
This is the single most common reporting error in multi-select analysis. There are two defensible bases and they give different numbers.
Base | What it divides by | Reads as | Use when |
|---|---|---|---|
Percent of respondents | Number of people who saw the question | "45% of customers use feature B" | Almost always. This is the default and the one stakeholders assume |
Percent of responses | Total number of selections made | "B accounts for 18% of all features named" | Share-of-mention analysis, where the mix of selections is the point |
Whichever you choose, print the base on every chart: the label should say what n is and what it counts. Mixing the two bases across slides in the same deck is how a finding quietly changes size between meetings.
Checkbox lists under-report
A multi-select checkbox list and a grid of separate yes-or-no questions ask the same thing, but they do not return the same numbers. Checkbox lists systematically produce fewer selections. Respondents treat the list as a whole, tick what comes to mind first, and move on once they feel they have answered. A forced-choice grid makes them consider every item individually, and item-level endorsement goes up.
Two consequences:
Do not read checkbox multi-select shares as absolute incidence. Read them as relative: which options lead, which trail.
Never change format between waves of a tracker. If wave 1 used checkboxes and wave 2 used a yes/no matrix grid, the jump you see is the format, not the market.
Other things that will bite you
Changing the option list mid-tracker. Adding one option changes the competitive set and depresses every other option's share. If you must add one, note the break in the series.
Ignoring how many options each person picked. Average selections per respondent is a useful diagnostic. If it drops sharply late in the questionnaire, you are watching fatigue, not preference.
Charting in list order. Sort by frequency instead. List order is an artifact of how you wrote the question, and if you randomized it, it means nothing at all.
Treating a blank as "none". A skipped multi-select is missing data. Only an explicit "None of the above" is a "no to everything".
Example questions
Four questions that follow the rules above.
Single-select, mutually exclusive bands:
Which of these best describes your household's main internet connection at home? Select one.
Fiber · Cable · DSL or phone line · Mobile or 4G/5G router · Satellite · I don't know
Multi-select with anchored specials:
Which of these have you used in the past 30 days? Select all that apply.
[randomized: mobile app · website · phone support · live chat · email support · in-store] then anchored: Other (please specify) · None of these
Capped multi-select, to force prioritization:
Which of these mattered most in your decision to sign up? Select up to three.
[randomized substantive options] then anchored: Other (please specify) · None of these
Two-step funnel, replacing a long list:
Step 1: Which category did you shop in most recently? Select one. Step 2: Which of these brands in that category have you bought in the past year? Select all that apply.
The last mile: ask why
A multiple choice question tells you what people picked. It never tells you why, because the reasons live outside the option list you wrote. The standard workaround is a follow-up open text box, which most respondents answer in four words or skip entirely.
This is where conversational formats earn their place. A voice survey can ask the closed question for clean, countable data and then probe the answer with a follow-up, in the respondent's own language, without a human moderator on the call. The point for your option list is simpler: write it well enough that the closed data stands on its own, then use open follow-ups for what the list could never have anticipated.
Frequently Asked Questions
What is the difference between single-select and multi-select multiple choice?
Single-select allows exactly one answer and uses radio buttons. Its results sum to 100%. Multi-select allows any number of answers and uses checkboxes. Each option becomes its own yes/no variable, so results sum to more than 100%. They measure different things: single-select measures which option wins, multi-select measures which options apply.
How many answer options should a multiple choice question have?
As a working guide, keep a visible flat list to roughly 7 to 10 options so respondents can scan it without scrolling, especially on mobile. Longer lists invite satisficing, where people pick the first acceptable answer and stop reading. If you need more options, group them under headers, use a search field for well-known closed lists like countries, or split the question into a two-step funnel.
Should you randomize the order of answer options?
Yes for unordered lists such as brands, features, or reasons, because early options attract more selections in web surveys and later options attract more in spoken ones. Randomizing per respondent turns that bias into noise. Do not randomize ordered scales, time periods, or numeric bands, and always anchor "Other", "None of the above", "Don't know" and "Prefer not to say" at the bottom.
What does "mutually exclusive and exhaustive" mean?
Exhaustive means every respondent can find an answer that is true for them, so nobody is forced to pick something false. Mutually exclusive means no respondent legitimately fits two options at once. The classic violation is overlapping age bands like 18-25 and 25-35, where a 25 year old belongs in both and the split between them becomes arbitrary.
When should you include an "Other (please specify)" option?
Include it whenever you cannot be certain your list covers every real answer, which is most of the time for brands, reasons, and behaviors. Skip it when the list is genuinely closed, such as a choice between two products you both control. Expect to code the free text before you can report it, and treat a large "Other" bucket as a sign the list needs more named options.
Why do multi-select percentages add up to more than 100%?
Because each respondent can select several options, so the selections outnumber the people. The percentages are normally shown as percent of respondents, meaning "45% of people picked this", and each option is calculated independently. The alternative base is percent of responses, which divides by total selections and does sum to 100%. Label which one you are using on every chart.
How many answer options should a survey multiple choice answer set have?
Four to seven is the comfortable working range for attitude, driver and frequency questions, because the whole list stays visible on a phone. Eight to twelve is workable if you randomize the order and keep labels short. Past thirteen, expect satisficing and a long tail of options under 2%, and restructure with grouped headers, a search field or a two-step funnel. Scale questions are a separate case: five or seven points are the usual defaults, and the count must never change between waves.
What is a balanced answer set, and why does an unbalanced one bias results?
A balanced set has the same number of positive and negative options, with labels of matching intensity on each side of the midpoint. An unbalanced set gives one direction more room, so respondents have more ways to land there and they read the list as a signal of the expected answer. "Extremely satisfied, very satisfied, satisfied, neutral, dissatisfied" gives happy respondents three boxes and unhappy ones a single box. Top-box scores from a set like that cannot be compared with a balanced one, and analysis cannot recover distinctions that were never offered.
Is "Don't know" an answer option or missing data?
Missing data. "Don't know" and "Prefer not to say" mean the respondent could not or would not answer, so they normally come out of the base before percentages are calculated and get reported as a separate rate. "Other (please specify)" and "None of the above" are substantive answers and stay in the base. Never merge "Don't know" into a neutral midpoint, which turns a knowledge gap into a moderate opinion, and never merge it with "Prefer not to say", which is a refusal.
