← Blog
May 15, 2026 · 4 min

The matrix question is a cognitive trap

Open any enterprise survey tool and you'll find the matrix question within three clicks. A grid of rows and columns: statements on the left, a scale across the top, radio buttons at every intersection. It looks organized. It looks efficient. It looks like the kind of thing a thoughtful person would build.

It is the most psychometrically damaging question type in common use. The research on this is not ambiguous.

The straight-lining problem

Couper, Traugott, and Lamias published a study in 2001, expanded in 2013, comparing matrix grids to the same questions presented one at a time. The finding: grids increased straight-lining — selecting the same response for every row — by a factor of three. Three times as many people clicked “Agree” down the entire column when the questions were in a grid versus when they appeared individually.

Matrix grid · straight-lining pattern

 Strongly disagreeDisagreeNeutralAgreeStrongly agree
Onboarding was clear
Support was responsive
Docs were helpful
Pricing was fair
I'd recommend it
All “Agree” — Couper et al. (2013): grids increase straight-lining 3x.

This isn't because the grid attracts lazy respondents. It's because the grid creates the behavior. Tourangeau, Couper, and Conrad (2004) documented the mechanism: visual grouping triggers cognitive grouping. When questions appear in a grid, respondents treat them as a single task rather than separate questions. The implicit message is “these are all the same kind of thing, so they probably deserve the same kind of answer.”

The grid also introduces spacing and position effects. Items in the first and last rows get more attention. Middle rows get skimmed. The physical layout of the grid changes which questions get thoughtful answers and which get clicked through. The data you collect is shaped by where you put the row, not by what the row asks.

Why creators love it anyway

The matrix grid optimizes for the wrong person. It's compact for the form creator. Ten questions in the space of two. The analytics view is tidy: a heat map with color-coded cells. It looks like data. It patterns like data. It exports to a spreadsheet beautifully.

Krosnick (1991) called this the gap between the survey designer's goals and the respondent's experience. The designer wants coverage — ten dimensions measured in one block. The respondent wants to be done. The grid gives the designer the illusion of coverage while giving the respondent every incentive to satisfice.

A 2020 study in Survey Research Methods measured what straight-lined data does to analysis. Removing straight-lined responses shifted regression coefficients by 13-39%. The data wasn't just noisy. It was directionally misleading — conclusions drawn from the full dataset pointed in a different direction than conclusions drawn from the dataset with straight-liners removed.

The alternative that produces signal

Pluck doesn't have matrix grids. This is a design decision, not a missing feature.

Instead: individual fields with follow-up probes. Three open-ended questions that ask the respondent to name specifics, with probes that nudge shallow answers deeper. “What slowed your team down?” followed by “Which dependency had the biggest impact?” when the first answer is vague.

3 fields with probes · richer than 10 rows of agree/disagree

What slowed your team down this quarter?
Cross-team dependencies on the payments API.
Probe: Which dependency had the biggest impact? Was it a technical blocker or a scheduling one?
Technical. Their API spec changed twice after we started building. We rewrote the integration layer both times.
What would you change about onboarding?
New hires take too long to ship their first PR.
Probe: How long is "too long"? Is the bottleneck setup, code review, or context?
Context. The codebase has no architecture doc. People spend their first week reading code instead of writing it.
How would you rate vendor support?
Mediocre. Tickets sit for days.
Probe: Can you give a recent example? How long was the wait, and what was the impact?
Filed a P1 on March 4th about broken SSO. Response came March 7th. 40 people locked out for 3 days.
3 questions, 6 data points. A 10-row matrix gives you 10 identical radio buttons.

Three fields with probes produce more usable data than ten rows of agree/disagree. The first answer gives you the topic. The probe gives you the detail. You get quotes you can act on — “Their API spec changed twice after we started building” — instead of a column of filled circles that all say “Agree.”

The tradeoff is real. You measure fewer dimensions. You can't produce a heat map. The analytics are text, not numbers, and text is harder to aggregate. But text that says something specific is worth more than numbers that say nothing — and a 5-by-10 grid where 30% of responses are straight-lined is numbers that say nothing.

Who the form is for

The matrix grid persists because form tools are designed around the creator's experience. The creator sees a clean grid and thinks “efficient.” The respondent sees a wall of radio buttons and thinks “I'll just click Agree.”

Every question type is a bet on who the form is for. The matrix bets on the creator. Individual fields with probes bet on the respondent. Pluck bets on the respondent, because the quality of the response is the only thing that matters once the form is sent.

Make your first form →

get started

Ready to get answers worth reading?

Your first form takes 60 seconds. Respondents can fill it in a browser or hand it to their AI.

Make your first form →