Matrix questions in surveys: examples, when a grid helps, and what to use instead

Open almost any employee survey or course evaluation and you will find one: a column of statements down the left, a row of labels across the top, and a radio button in every cell. That is a matrix question, also called a grid question or a matrix table question. It lets a respondent rate six things in the space one question would take, and it lets the survey author ask six questions while it looks like one.

Grids are not wrong. They are a trade. You save screen space and the clicks between pages, and in return you get more careless answering, a layout that fights small screens, and a block that is tiring to read. Whether the trade is worth it depends on who is answering and on what device. This guide covers how to decide, how to write a grid that holds up if you keep it, and how to get the same data without one.

What a matrix question is

A matrix question has three parts.

• Rows are the items being judged: statements, features, parts of an experience. Each row is really a question of its own.
• Columns are the answer options, shared by every row. Usually a rating or agreement scale, such as very dissatisfied to very satisfied.
• Cells are where the answers go. In the common single-select matrix, each row takes exactly one answer.

Because every row shares the same columns, every row is analyzed the same way, and comparing rows is easy: the venue scored higher than the catering, workload lower than pay. That comparability is the whole appeal of the format.

Common types of matrix question

• Likert matrix. Agreement statements on a strongly disagree to strongly agree scale. The most common grid of all, and the one the Likert scale survey guide covers in depth.
• Rating matrix. Items scored on a numbered scale, say 1 to 5, with the two ends explained. The rating scale questions guide covers how to pick the range.
• Frequency or importance matrix. The same layout with never to always, or not at all important to extremely important, across the top.
• Multi-select matrix. Checkboxes instead of radio buttons, so a row can take several answers: "Which devices do you use for each of these tasks?"
• Dropdown or text matrix. A dropdown or a small text box in every cell. Common in research data collection, rare in customer surveys, and the hardest of all to fill in on a phone.

Most of this guide is about the first three, which account for most of the grids in real surveys. For where the matrix sits among choice, open, ranking and scale questions, see the survey question types guide.

When a matrix question saves time

A grid works when four things are true at once.

1. Every row really shares the scale. If one row needs "how often" and the next needs "how satisfied", they are two questions, not two rows of one.

2. The rows belong together. Parts of one event, sides of one job, features of one product. People answer faster and more consistently when they stay in one frame of mind.

3. Most people answer on a large screen. An internal survey sent to staff at their desks, or a research panel that takes surveys on laptops.

4. The grid is short. Five to seven rows is a sensible ceiling. Past that, the quality of each answer falls even when the grid still fits on the screen.

When all four hold, a matrix is genuinely efficient. Respondents compare the items against each other as they go, which can make relative judgments more consistent, and there is less clicking from page to page. The survey also looks shorter at first glance, which helps people start it.

That last point cuts both ways. A survey with ten grids of eight rows looks like ten questions and is really eighty. Respondents notice by the third grid, and it shows in the answers.

Where grids hurt your data

None of these problems produces an error. They produce complete-looking data that is less accurate than it seems, which is why grids stay popular with the people who write surveys and unpopular with the people who answer them.

Straightlining

Faced with a block of identical-looking rows, some respondents pick the same column all the way down: agree, agree, agree, agree. It is the fastest route through, and in a grid it takes a few seconds. Survey researchers call it straightlining, or nondifferentiation, and studies comparing the two layouts have often found more of it in grids than when the same items are asked one at a time. The result reads as steady mild approval. Often it is the absence of an answer.

The last row in that grid is a common check. A statement worded in the opposite direction ("I am often unsure what is expected of me") catches the straightliner who agreed with everything, because nobody can honestly agree with it and with the first row as well. Use such checks sparingly; too many reversed statements confuse careful respondents too.

Phones

A row label plus five worded columns does not fit across a phone held upright. Survey tools cope in one of three ways. They shrink the labels until they are hard to read, they let the grid scroll sideways so the row labels disappear, or they quietly turn the grid into a stack of separate questions on small screens. The first two make mistakes more likely. The third is an admission that one row per screen is the version that works on a phone.

Even when a grid fits, the tap targets are small and packed tightly together. A thumb aimed at agree lands on strongly agree, and nobody goes back to check.

Fatigue and drop-off

Grids are dense. The respondent reads a row, looks up to the column labels, finds the right cell, and repeats. Long or repeated grids are a familiar point for people to abandon a survey, and the ones who stay tend to speed up and vary their answers less toward the bottom of each grid. The form abandonment guide covers the wider causes of people leaving before the end.

Labels out of sight

By the seventh or eighth row, the column headings may have scrolled off the top of the screen. From then on people answer by position (the fourth circle along) rather than by the word. If the scale runs in an unusual direction, or a reversed statement slips in, answering by position gets it wrong.

Screen readers

For someone using a screen reader, a grid is announced cell by cell, with the row and column headers repeated or, in a badly built grid, missing entirely. What a sighted respondent takes in at a glance becomes a long sequence of announcements. If accessibility matters for your audience, and it usually does, separate questions are the safer layout. The form accessibility guide goes further.

Matrix question examples

Six grids that turn up in real surveys, each with its rows and its scale. Every one of them can be asked as a grid or as a short run of single questions, and the note under each says which usually works better.

1. Event satisfaction

Question: How satisfied were you with each part of the event? Scale: very dissatisfied to very satisfied.

• Registration and check-in
• The venue
• The speakers
• Food and drink
• The schedule and timing

Five rows on one scale, answered soon after the event and mostly on phones. A strong candidate for one row per screen.

2. Employee engagement

Question: How much do you agree with each statement? Scale: strongly disagree to strongly agree.

• I know what is expected of me at work.
• I have the tools I need to do my job well.
• My manager gives me useful feedback.
• My workload is manageable.
• I see a future for myself here.

Staff often answer at a desk, so a grid is defensible. Keep one idea per statement and check the results for straightlining.

3. Course evaluation

Question: Rate each part of the course. Scale: 1 (poor) to 5 (excellent).

• Clarity of the lectures
• Usefulness of the readings
• Fairness of the assessment
• Pace of the course
• Support from the instructor

Students answer on phones between classes. One rating per screen is the safer build.

4. Feature importance

Question: How important is each of these to you? Scale: not at all important to extremely important.

• Offline access
• Integrations with other tools
• Team permissions
• Price
• A mobile app

The classic importance-grid failure: everything comes back important. If you need priorities rather than ratings, a ranking question or a "pick your top two" question tells you more.

5. Frequency of use

Question: In the last month, how often did you use each of these? Scale: never, rarely, sometimes, often, always.

• Reports
• Shared dashboards
• Email alerts
• The mobile app

Name the period in the question so that "often" means the same thing to everyone. Four short rows is a reasonable grid on a laptop.

6. Multi-select: devices by task

Question: Which devices do you use for each task? Choose all that apply. Columns: phone, tablet, laptop, desktop.

• Reading email
• Writing documents
• Joining video calls
• Shopping online

Checkbox grids are the hardest to answer on a phone. As separate questions, each task becomes a multiple choice question with the four devices as options, which is easier to read and just as easy to count.

For the questions that sit around a grid (screeners, open follow-ups, demographics), the survey questions guide has 120 examples written for copying.

How to write a matrix question that holds up

If a grid is the right call for your audience, these rules keep the damage small.

1. Five to seven rows at most. Split a longer list by topic into smaller grids, each with its own short heading.

2. One scale, with a word on every column. Words rather than bare numbers, and balanced: as many negative options as positive ones, mirrored around a true middle.

3. One idea per row. "Fast and friendly support" is two rows. Someone who got a friendly, slow reply has no honest answer to it.

4. Short row labels. A phrase, not a paragraph. Long labels push the scale off the screen.

5. A separate "Not applicable" column at the far end when some rows will not apply to everyone, left out of every percentage and average.

6. Don't require every row by default. Forcing an answer in every cell pushes people with no view into guessing. Require the grid only when you truly need a complete set.

7. Test it on a phone before you send it. If it scrolls sideways or the labels shrink to a squint, rebuild it as single questions.

The survey design guide applies the same discipline to the questionnaire as a whole: order, length and flow.

Alternatives to a matrix question

Every grid can be rebuilt from simpler question types. The data comes out in the same shape. The experience of answering does not.

The first two cover most grids. A labeled choice keeps a word on every point, which matters for anything you will report or track. A numbered rating is quicker to build and to answer, and it is enough when you only need a score with defined ends.

The last three change the question rather than the layout, and they are often the better fix. An importance grid where every row comes back at four or five is really asking for priorities. A checkbox grid can usually be cut down to the one or two rows you will act on, each asked as a plain multiple choice question. One caveat if you build in Upform: it has no ranking question type, so the practical route to priorities there is a Multiple Choice with a maximum of two or three selections.

How to analyze matrix question data

Whether the answers came from a grid or from one question per row, the analysis is the same, and it goes row by row.

1. Report each row's distribution. The share choosing each label, with the number of respondents behind it. A stacked bar per row, sorted from best to worst, is the clearest chart for grid data.

2. Add top-2-box and bottom-2-box. The share choosing the two most positive options, and the two most negative. With both in one table, rows can be compared at a glance.

3. Compare rows, then groups. Which part of the experience scored lowest overall? Then, does it score lowest for first-time customers too, or only for returning ones?

4. Check for straightlining. Count the respondents who gave the identical answer to every row. A few is normal, and on a short grid some of it is genuine. A large share, especially on a long grid, says the format is the problem. In a spreadsheet it is one formula per response: flag it when every answer in the block is the same.

5. Be careful with a single overall score. Averaging every row into one number makes sense only when the rows measure one idea, as in a multi-item Likert scale. Satisfaction with the venue and with the catering are different things, so report them separately.

A worked example with illustrative numbers: 180 people answered the event satisfaction rows on a five-point scale.

Three rows are healthy. Food and drink has a third of attendees unhappy and a median in the middle, so that is the row to act on, and a follow-up question asked only of the unhappy third tells you why. In this example 22 of the 180 people (12%) gave the same answer to all four rows. On four rows that is plausible. On twelve rows it would be a warning.

A grid saves the author a few screens. The respondent pays for them in attention.

For distributions, medians and why a bare average misleads on scale data, the Likert guide linked above has a longer worked example.

Matrix questions in Upform: one row per screen

Upform is a conversational form builder: one question per screen, with full-width answer options and, on phones, a sticky back and next bar at the bottom. There is no grid layout to put a matrix in. The replacement takes a few more minutes to build and changes very little about the data. It works on the free plan, with all 21 question types, conditional logic and unlimited responses.

Rebuild the grid, one row at a time

1. Introduce the block with a Statement. A Statement screen has no input. Use it for the shared instruction: "The next four questions are about the event. Tell us how satisfied you were with each part."

2. Add one question per row. A Single Choice with the scale labels as options, in the same order on every screen, when every point needs a word. Or a Rating set to 5, with "1 is very dissatisfied, 5 is very satisfied" in the description, when a number with defined ends is enough. Rating runs from 1 up to a maximum you set between 3 and 10; for a 0 to 10 scale, use Slider Scale.

3. Use the row label as the question. Put the row ("The venue") in the question and the shared instruction in the description. The question title becomes the column heading in Responses and in the export, so short titles give you a tidy table.

4. Show the progress bar. One row per screen means more screens. A bar, dots or a percentage shows people that the end is close.

5. Decide what is required. Leave a row optional, or add "Not applicable" as a last option, when some people will not have an answer.

Why it often gives cleaner answers on a phone

Asked one per screen, each row is read on its own, with every label in view and a full-width target for each answer. There is no sideways scrolling and no column heading to lose track of. Straightlining does not vanish, but it stops being one thumb movement down a column: each identical answer is a separate, deliberate choice on a fresh screen.

The cost is length. A five-row grid is one screen; here it is five, plus the introduction. Keep each block to seven rows or fewer, and cut rows you will not act on rather than squeezing more of them in. The conversational forms page covers the one-question-per-screen flow in more detail.

The export has the same shape as a grid's

Most survey tools export a matrix question as one column per row. Upform gets there by construction, because every question is its own column. The Responses table shows each row of your former grid as a column, headed by the question title and its type icon. Export CSV writes one line per response, with the form, the submission time and the location first, then one column per question. Distributions, top-2-box and row-by-row comparison work exactly as they would on grid data.

If you would rather have the answers somewhere live, the Google Sheets integration appends each response to a sheet as it arrives.

Ask why, only when a row scores low

A score tells you where the problem is. The reason behind it is often the most useful answer in the survey, and it only needs asking of the people who gave a low score. With conditional logic, a Show rule on the Long Text question ("What would have made it better?") holds one condition for each low answer, joined by OR: "The venue" is exactly Dissatisfied, "The venue" is exactly Very dissatisfied, and the same two for "Food and drink". If you built the rows as Rating questions, one condition per row is enough, because Rating offers equals, greater than and less than: "less than 3". Everyone who was happy moves straight on.

People who were happy never see the follow-up, so it adds no length for them. More on rules, operators and the jump and end form actions on the conditional logic page.