"Very satisfied / Somewhat satisfied / Neither / Somewhat dissatisfied / Very dissatisfied" — measuring agreement or satisfaction on a graded scale is called a Likert scale. Psychologist Rensis Likert introduced the format in 1932, and it remains the single most common question format in surveys today.
A common failure pattern: you default to 5 points, collect 100 responses, and find that 60% of them landed on 3. You can't make a decision from that data. This article focuses specifically on how to choose the number of points — including a decision flowchart, concrete before/after examples, and the implementation pitfalls that bite teams in practice. For a broader comparison of all question types, see the question types guide.
Why Likert scales dominate survey design
- Low respondent burden: Under 5 seconds per question
- Easy to compare quantitatively: Mean, median, and distribution work well — useful for month-over-month trends and pre/post comparisons
- Captures degree, not just direction: Unlike yes/no, it records how much
The catch: wrong point count = unusable data.
Point count options: characteristics and when to use each
5-point scale — all-purpose, best for trend tracking
The most widely used format. It feels like a school grade scale, so respondents understand it immediately.
Example question (5-point)
Rate the ease of use of this service. 1 = Very difficult 3 = Neither 5 = Very easy
- Pros: Universally understood, fast to answer, easy to trend across periods
- Cons: Strong central bias — responses cluster on point 3
Best for: Ongoing satisfaction surveys (monthly/quarterly), internal surveys, pulse surveys
What central bias looks like in practice: Out of 100 responses, 60 land on 3, 30 on 4, 10 on 2. The mean shifts between 3.1 and 3.3 quarter to quarter — not enough signal to make a call on whether something improved.
7-point scale — precise effect measurement and A/B testing
The default in academic research. Finer resolution makes it easier to detect mean differences.
Example question (7-point)
Rate the ease of navigation compared to the previous version. 1 = Extremely difficult 4 = Neither 7 = Extremely easy
- Pros: Better at detecting statistical differences; good for measuring impact of specific changes
- Cons: More options slow responses down by roughly 2–3 seconds per question; respondents often can't articulate the difference between points 5 and 6
Best for: UI improvement pre/post comparisons, competitive product evaluation, research designs requiring statistical sensitivity
Tip: Labeling all 7 points makes options too long. Use anchor labels at the ends and middle only.
4-point or 6-point (even-point) scale — eliminating central bias
Removes the middle option. Without "Neither," respondents must lean positive or negative.
Example question (4-point)
Do you plan to continue using this service? 1 = Definitely not 2 = Probably not 3 = Probably yes 4 = Definitely yes
Before (5-point — problematic):
"Likelihood to continue: 1 = Not at all likely, 5 = Very likely" → Responses cluster at 3; you can't read the strength of intent
After (4-point — improved):
"Likelihood to continue: 1 = Not at all likely / 2 = Not very likely / 3 = Somewhat likely / 4 = Very likely" → Clear distribution: 73% positive / 27% negative
- Pros: Eliminates central bias; cleaner output for prioritization decisions
- Cons: Forces a directional answer from genuinely neutral respondents; some respondents find it constraining
Best for: Feature prioritization, decision-driving research, intent measurement
10-point scale — NPS-style and scoring intuition
NPS (Net Promoter Score) typically uses 0–10, an 11-point scale.
- Pros: Captures fine-grained differences; aligns with the intuitive "score out of 10" mental model
- Cons: Strong cultural bias in how people use the top end — Japanese respondents tend to cluster around 7, US respondents around 9, on the same scale. Cross-cultural comparison can mislead.
- Outside of NPS, this scale is used rarely.
Decision flowchart for picking point count
Is the goal long-term trend tracking?
→ Yes: 5-point (consistency with past data is critical)
Is the goal pre/post comparison or A/B testing?
→ Yes: 7-point (easier to detect mean differences)
Is "neither" in the data a problem for your decisions?
→ Yes: 4-point or 6-point (even scale)
Is it NPS or a scoring-style question?
→ Yes: 10-point (0–10)
Still unsure?
→ 5-point (most versatile, most benchmark data available)
Label design — the often-overlooked factor
How you label the scale affects response distribution.
Anchor label method (recommended)
Label only the ends and the middle.
| Value | Label |
|---|---|
| 1 | Very dissatisfied |
| 2 | (no label) |
| 3 | Neither |
| 4 | (no label) |
| 5 | Very satisfied |
Mobile consideration: Long option text forces vertical stacking, increasing scroll. Anchor labels allow horizontal radio buttons on mobile without overflow.
Full-label method (for academic or high-precision surveys)
Label every point ("Somewhat dissatisfied," "Somewhat satisfied," etc.). Reduces interpretation variance, but makes options long and hard to read on mobile.
Recommendation: Anchor labels (3 points) are sufficient for most marketing research. Full labels are worth the trade-off only when you need academic-level precision.
Three techniques to reduce central bias
If you want to keep a 5-point scale but limit the clustering at center:
1. Switch to an even-point scale
4 or 6 points removes the neutral option entirely, forcing responses to disperse.
2. Move "don't know" outside the scale
1 = Very dissatisfied / 2 = Dissatisfied / 3 = Satisfied / 4 = Very satisfied
/ [I haven't used this, so I can't say]
This separates genuine neutrality from lack of experience, and keeps the scale itself from getting polluted by non-opinions.
3. Strengthen the question stem
Weak stem: "Are you satisfied with this service?" → Easy to slide to center
Strong stem: "If you had to choose — would you continue using this service or cancel it?" → The forced-choice framing naturally disperses responses
Real-world failure cases
Failure 1: The "everyone gave it a 3" problem
Situation: A team ran a 5-point post-release survey on a new feature. Mean responses converged to 3.1–3.3. No actionable signal.
Cause: Central bias, plus a vague question stem ("What do you think of the new feature?")
Fix: Switched to 7 points and changed the framing to a comparison: "Compared to the previous version, how has the ease of operation changed? (1 = Much worse → 7 = Much better)" → Mean came in at 4.8, clearly above center. Improvement confirmed.
Failure 2: Changing scales mid-study and losing the time series
Situation: Quarterly satisfaction tracking was running on a 5-point scale for two rounds. The team noticed central bias and switched to 4-point on round three. The before/after comparison became impossible.
Fix: Lock the scale before the study starts. If you need to change it, add the new scale as a new question (keep the old question running in parallel). Run one or two overlap rounds to bridge the gap.
FAQ
Q. What's the difference between a Likert scale and NPS?
NPS (Net Promoter Score) is a specific question format — "How likely are you to recommend us to a friend? (0–10)" — where 9–10 are "Promoters," 0–6 are "Detractors," and the score is Promoters% minus Detractors%. A Likert scale is the broader category of graded response formats, of which NPS is one specific instance. In practice: use NPS for tracking customer loyalty over time; use Likert (5–7 points) for detailed product or service evaluation. (More detail: NPS complete guide)
Q. Is 5-point or 7-point statistically more accurate?
Both are adequate for sample sizes above 100. The 7-point scale has better sensitivity — it's easier to detect a shift in mean (e.g., pre-release 3.8 vs. post-release 5.1 reads as a clearer improvement than on a 5-point scale). That said, more points also mean slightly lower completion rates. For consumer-facing short surveys, 5 points often wins on completion; for internal or research-grade surveys, 7 points is worth the extra ask.
Q. Doesn't a long scale break on mobile?
Yes — 7+ points in a horizontal radio button layout becomes tiny and hard to tap on a 375px screen. Solutions: (1) use a slider format, or (2) stack the options vertically with labels. If horizontal layout is a requirement, 5 points is the practical maximum.
Q. Is a "neutral" option always worth including?
It depends on the question. Behavioral intent questions ("Would you continue using this?", "Would you recommend this?") benefit from an even scale — you want a directional answer. Experience/perception questions ("How did you find the onboarding process?") should have a neutral option — forcing a direction feels unfair if the respondent genuinely had no strong reaction. Avoid making your entire survey even-scaled; some respondents will find it stressful.
Q. Can I mix point counts within a single survey?
Yes — mixing is fine. Just keep the same point count within a thematic block. For example: all six product evaluation questions on 7-point, all two intent questions on 4-point. Consistent scales within a block make responding easier and aggregation simpler.
Summary: quick reference
| Goal | Recommended scale | Reason |
|---|---|---|
| Long-term trend tracking | 5-point | Best balance of continuity and low respondent burden |
| Pre/post and A/B testing | 7-point | Higher sensitivity to mean differences |
| Avoiding central bias / decision research | 4-point or 6-point | Forced dispersion makes direction clear |
| Loyalty measurement | NPS (0–10) | Industry standard with benchmark data available |
When you're unsure which scale to use, describe your goal to Repoan's AI chat ("I want to track monthly satisfaction" or "I need to measure the impact of a specific change precisely") and it will propose the right scale and question wording automatically. 20+ question types are supported (details), and changing the point count after setup does not break compatibility with already-collected data.
Related articles
- Handling central bias and acquiescence bias: Survey error and bias types
- Reading statistically meaningful differences: Statistical significance basics for surveys
- All question types compared: Question types guide