Matrix question

"Rate from 1 to 5: service quality, delivery speed, product range, price" - four items, one scale. Instead of four separate questions - a single table: rows are the items, columns are the gradations. This is a matrix question.

In SurveyNinja it is built with the "Matrix" or "Scale" element with several rows. It saves space and time, but carries risks: fatigue, central tendency, uniform row-by-row answers. When a matrix is justified and when it is better to split it into separate blocks - below.

A matrix is handy when you need to rate 5-10 aspects on the same scale. Separate questions would take up a whole screen; a matrix keeps it compact. But a table that is too long produces "striped" answers - the respondent gives the same value across every row.

Rows are the items, columns are the scale

Matrix question - a question structure in which several items (rows) are rated on one shared scale (columns). The respondent picks a value for each row. Example: rows - "quality", "price", "packaging"; columns - "strongly agree" to "strongly disagree" or 1-5. It is convenient for rating a set of attributes or satisfaction across several parameters. It saves space, but with a long matrix the risk of fatigue and patterned answers grows.

In essence: "rate each item on this scale" presented as a table.

When a matrix fits: 3-8 items, one scale

When there are several items (3-8), the scale is shared, and the logic is clear. Rating quality by aspect (speed, courtesy, cleanliness). Satisfaction with service elements. Agreement with a set of statements on a Likert scale. For quantitative surveys a matrix gives a compact block - you can compute the average for each item and compare across rows.

When there are more than 8-10 items - it is better to split into two matrices or move part of it into a separate block. A long table strains attention.

Risks: fatigue, striped answers, anchoring effect

The respondent sees a repeating pattern - it is easy to "fill in" one column across all rows. Especially if the items are uniform ("Agree/disagree" with five similar statements). The data turns out flat - little difference between items. Central tendency - pulls toward the middle of the scale. The primacy effect - the first rows are read more carefully.

To reduce the risk: fewer rows (5-7), a different order of items (randomization), alternate with other question types, do not place matrices back to back.

If in the report all the row averages are almost identical - it is worth being cautious. Either respondents genuinely do not distinguish between the aspects, or they are "striping". Look at the distribution: how many "all fives" or "all threes" are there - such patterns reveal fatigue.

Matrix vs separate scales

An alternative is a separate scale question for each item. More space, but higher engagement: the respondent does not "scan" the table but rates one at a time. For 3-4 key aspects it is sometimes better to break them out. For 6-8 - a matrix is more sensible, otherwise the survey bloats.

Another format is semantic differential scales: two characteristics at the ends, the respondent picks a point between them. That is already a different construction.

In SurveyNinja: the "Matrix" and "Scale" element

A matrix question in the SurveyNinja builder is the "Matrix" element or an extended "Scale" with several rows. You add rows (items to rate) and set shared columns (1-5, "agree - disagree", and so on). You can enable randomization of the row order. More details - in the element settings.

Answers in the reports - a summary for each item (average, distribution). Export to Excel gives a separate column for each item - handy for cross-tabulations and segmentation. Row averages can be compared - which aspect is weaker, which is stronger.

Wording the rows: unipolar and clear

Rows should be unipolar: "strongly agree" is good, "strongly disagree" is bad (or vice versa). Do not mix in one matrix statements where agreement is a "plus" with those where agreement is a "minus" - otherwise when computing the average you will have to recode. Better to use two blocks or reword.

Items should be short and clear. "Service quality" is better than "The level of service quality of the staff at the moment of contact". Long rows slow things down.

Too many rows, an unreadable table on mobile

A matrix that is too long - respondents get tired, answers become patterned. 5-7 rows is a reasonable limit for one table. More than that - split it.

On mobile a wide table reads poorly - the columns get squeezed. In SurveyNinja you can check this in the preview with the mobile version. If the matrix "breaks" - replace it with separate scales or shorten it.

Where it is used: satisfaction, employees, products

Matrices are common in satisfaction surveys - rating aspects of service, delivery, support. In HR - engagement surveys, rating a manager by parameters. In product research - the importance of attributes, concept evaluation. All of these are typical scenarios: several items, one scale, a compact block.

Case: rating 5 aspects of delivery

A survey after delivery: "Rate from 1 to 5" - speed, packaging quality, courier courtesy, order accuracy, payment convenience. A 5×5 matrix - one screen. 400 responses: row averages 4.2, 4.5, 4.8, 4.1, 4.6. You can see that "order accuracy" and "speed" are lower - areas for improvement. If you had made 5 separate questions - the survey would have stretched out, and some respondents might not have reached the end.

A cross-tabulation of "aspect × region" will show where the problems are local: for example, speed is low in only one city. A matrix gives not just a summary but also a basis for segment analysis.

One scale for all rows - do not mix types

The columns of a matrix should be the same for all rows. "Agree - disagree" for all statements, or 1-5 for all aspects. You cannot mix in one matrix: the first row "yes/no", the second 1-5, the third "often/rarely". The respondent gets confused, and the data is incomparable. Different scales - different blocks.

The number of gradations - 5 or 7 for Likert scales, 1-10 for NPS-like ones. Too many columns (10-11) overloads. Too few (3) is crude. Use what is standard in your field or in benchmarks as a guide.

A matrix with an "Other option" - rarely needed

In closed-ended questions an "Other" option with a text field is often added. In a matrix this is rarely appropriate: if you need a custom answer for every row - the table will expand. Usually a matrix is a clean structure, with no text fields. The exception is one shared row "What else matters?" with an input field at the end, but that is already a hybrid.

A matrix question - several items (rows) are rated on one scale (columns). It saves space and suits 3-8 aspects. The risk is fatigue and uniform answers. In SurveyNinja - the "Matrix" element or a "Scale" with several rows.