Likert scale questions: good examples and five ways they go wrong

Likert scale questions with examples for customer surveys, the five design mistakes that skew answers, and why the average can hide a split market.

Instant Expert EditorialPublished Updated 5 min read

Likert scale questions ask people how much they agree or disagree with a statement, usually on five labeled points from "Strongly disagree" to "Strongly agree." They are quick to answer and easy to count, which is why nearly every customer survey has some. Their known weakness is that people tend to agree with statements. Use them for attitudes you can phrase neutrally, keep the standard labels, and report the full spread of answers instead of only the average.

What counts as a Likert scale

The Nielsen Norman Group's guide to rating scales explains that the method is named after the psychologist Rensis Likert, who created it in the 1930s (the paper was published in 1932). Strictly, a Likert scale is a set of related statements combined into one score, and a single statement is a Likert item. Everyone calls both a Likert question anyway, and nobody is harmed.

NN/g notes that Likert items typically use five points, sometimes seven or nine, each labeled in words. NN/g recommends sticking to the standard labels instead of inventing new ones:

Strongly disagree / Disagree / Neither agree nor disagree / Agree / Strongly agree

Good statements are about one plain thing

These are written for a business audience; swap in your own nouns.

  • "I can find the information I need in our current reporting tool without help."
  • "The price we pay for our current software is reasonable for what we get."
  • "Our current supplier responds quickly when an order goes wrong."
  • "I would need approval from someone else to change this tool."

Each covers one idea in concrete nouns, with no praise for your product. Compare "Our innovative platform makes reporting easy and fast." It asks about two things (easy, fast) and asks people to agree with your marketing copy.

1. Some people agree with whatever you show them

Pew Research Center's guide to writing survey questions calls this acquiescence bias: some respondents tend to agree with any statement, more so when an interviewer is present. NN/g lists it, with social desirability bias, as one of the format's two main weaknesses.

Ask about the thing directly where you can. Pew recommends offering a choice between alternatives. Instead of "The reporting tool is easy to use (agree/disagree)," ask "How easy or difficult is the reporting tool to use?" with Very easy / Somewhat easy / Somewhat difficult / Very difficult. Pew's 2019 study of question formats also notes that construct-specific labels ("Yes, have done this" rather than "Yes") are a proven way to deal with acquiescence.

Be careful with reversed statements, which some questionnaires mix in to catch people who agree with everything. NN/g cites Sauro and Lewis (2011), who found they can confuse respondents, who miss the switch, and researchers, who forget to reverse-code the answers. If you use them, check the coding twice.

2. People answer as if someone is watching

People want their answers to look good, and NN/g notes that asking for names or other identifying details makes this worse. "Our team follows the documented process" will draw more agreement than reality supports. Skip identity questions you don't need, and ask about recent behavior: "In the past two weeks, how many times did someone on your team skip the documented process?"

3. Lopsided labels push answers up

NN/g's survey best practices warns that more positive than negative options pushes answers upward. "Strongly agree / Agree / Somewhat agree / Disagree" offers three ways to be happy and one way to complain. Keep the same number of options on each side of the midpoint.

4. "Not applicable" is a different answer from "neutral"

NN/g recommends a separate "Not applicable" option, so you can tell people with no opinion from people the statement doesn't fit. Someone who has never called your supplier's support line should not have to pick "Neither agree nor disagree" about how fast it responds.

5. A rating without a reason is half an answer

NN/g suggests an optional follow-up box such as "Why did you choose this rating?" Put it on the one or two statements that matter most. Open-ended survey questions explains what too many text boxes cost.

The average can hide a split market

Here is a hypothetical example. 200 veterinary clinic managers rate "The price we pay for our practice software is reasonable." In version A, all 200 pick the midpoint (3), so the mean is 3.0. In version B, 100 pick Strongly disagree (1) and 100 pick Strongly agree (5). The mean is (100 × 1 + 100 × 5) ÷ 200 = 600 ÷ 200 = 3.0. Same average, opposite markets: one shrugging, one split between clinics that feel overcharged and clinics that are content.

Report the share in each category, plus "top two" (Agree plus Strongly agree) next to "bottom two." If a group matters to your decision, such as clinics with more than five vets, show its distribution separately.

Sometimes Likert is the wrong tool

NN/g compares Likert items with semantic differential scales, where people place themselves between two opposite adjectives such as "easy" and "difficult." Those take more thought (the middle points are usually unlabeled) but avoid asking people to agree with a statement. Likert items are more flexible: "I would need approval from someone else to change this tool" has no natural pair of opposites.

To make people choose between priorities, use a ranking or forced-choice question, because on a rating scale everyone can rate everything "Agree." Types of survey questions covers when to use each.

Your next step

Run each draft statement through three checks: one idea only, no praise words, and a separate "Not applicable" where some people won't have the experience. Then rewrite your two most important statements as item-specific questions and compare.

When a result surprises you, such as a split on price, talk to a few people on each side before acting. If the survey was anonymous, describe the clinics instead, such as practice managers at independent veterinary clinics, and open each call by asking how they would rate the price statement and why. The directory page for operations professionals in veterinary services is one place to start. On Instant Expert, four 15-minute calls at an $80 offer each cost you $100 apiece with our fee on top (20% of what you pay), $400 at most, and you are charged only for the people who book (how it works).