Central tendency bias

A scale question from 1 to 5: "How satisfied are you with the service?" The respondent hesitates between "4" and "5", but picks "3" — they don't want to praise or criticize, so they choose "neutral." This is central tendency bias: respondents avoid the extreme ends of a scale and gravitate toward the middle values — "so-so," "hard to say," the center of the range. The effect distorts the distribution of answers and understates the real spread of opinion. It is a subtype of response bias and a kind of bias.

The effect is especially noticeable on Likert scales and other rating scales with an odd number of points, where there is a clear middle. Results "compress" toward the center: few extreme ratings, many middle ones — and the difference between groups or items is smoothed out.

What central tendency bias is in plain terms

Central tendency bias is the tendency of respondents to avoid extreme scale values (for example, "strongly agree," "strongly disagree") and choose middle options ("neutral," "hard to say," the center of a numeric scale). It arises from a desire not to stand out, not to take responsibility for an extreme rating, or from uncertainty. In surveys the effect understates the spread of answers, "compresses" the distribution toward the center, and can mask real differences between groups or items. It is a subtype of response bias.

Put simply: central tendency bias is when people more often pick "3" out of 5, "neutral," or "hard to say" than "1" or "5." Extreme ratings are understated, the middle is overrepresented — and the picture of opinions looks more "smoothed out" than it really is.

Why central tendency bias occurs

The desire not to stand out. Respondents don't want to seem too critical or too enthusiastic. An extreme rating is perceived as a strong stance; the middle as safe and neutral.

Uncertainty. When a respondent is unsure of the answer, they choose the middle option: "something in between" seems like a reasonable compromise. This is especially noticeable in complex or ambiguous questions.

Social desirability. On sensitive topics, extreme ratings ("not at all satisfied," "strongly agree") can seem socially risky. The middle reduces the risk of judgment. The effect overlaps with social desirability bias.

Cultural factors. In cultures that value restraint and conflict avoidance, middle answers are chosen more often. Extreme ratings may be perceived as aggressive or impolite.

Fatigue and inattention. In long surveys respondents get tired and start "marking the middle" to finish faster. The effect intensifies with long surveys without logic jumps.

When the effect is stronger

An odd number of points. Scales with a clear middle (1–5, "strongly disagree" — "neutral" — "strongly agree") amplify the effect. Even-numbered scales without a neutral option force a choice of one side.

Sensitive topics. Questions about job satisfaction, relationships with management, and service quality are topics where respondents play it safe and more often choose the middle.

Non-anonymous surveys. If a respondent knows the answer can be linked to them, extreme ratings (especially negative ones) are chosen less often. The middle is perceived as safe.

Complex wording. Unclear or ambiguous questions increase the share of "neutral" and "hard to say" — the respondent doesn't want to guess.

Long runs of similar scales. A series of "rate on a scale of 1–5" questions is tiring; respondents start putting middle values by default.

Examples of central tendency bias

NPS and satisfaction scales. The question "Rate the likelihood of recommending from 0 to 10" — many choose 5–7, avoiding 0–2 and 9–10. The true distribution of "promoters" and "detractors" is understated.

Rating product features. "Rate the importance on a scale of 1 to 5" — most put 3–4, rarely 1 or 5. Differences between features are smoothed out, and prioritization becomes harder.

Employee surveys. Questions about working conditions, management, and corporate culture often yield "middle" answers: employees don't want to stand out either positively or negatively.

Service quality assessment. "Poor — excellent" scales with an odd number of points are a typical zone for central tendency bias. Middle ratings are inflated by the overrepresentation of "fine."

How to minimize central tendency bias

Even-numbered scales without a neutral option. Use scales with an even number of points (for example, 1–4 or 1–6) so there is no clear "middle." The respondent is forced to lean one way or the other.

Remove or limit "hard to say." If a "hard to say" option isn't essential, you can omit it or place it separately. This reduces the "leak" into the neutral zone.

Clear and neutral wording. Clear questions reduce uncertainty and the share of "neutral." Avoid leading questions that may nudge toward the middle ("On the whole, are you rather satisfied?").

A short survey and logic. Fewer similar scales in a row and logic jumps reduce fatigue and the mechanical choice of the middle.

Anonymity. In sensitive surveys, guaranteed anonymity reduces the fear of extreme ratings and increases the share of honest "edges" of the scale.

Pilot. In a pilot survey, look at the distribution of answers across the scales. If almost everything is "in the middle," reconsider the number of points and the wording.

Relationship with other biases

Central tendency bias often combines with acquiescence bias (the tendency to agree with statements): both "compress" answers — one toward "agree," the other toward "the middle." In surveys with Likert scales, both effects can operate at once. The difference: acquiescence bias shifts toward "yes/agree," while central tendency bias shifts toward the middle points. For more on types of scales and how to choose them, see the glossary entry "Rating scale" and the article "Quantitative research".

Common mistakes

Interpreting the middle as "the norm." A high share of "3" out of 5 doesn't necessarily mean "people are on average neutral" — it may be an artifact of central tendency bias. You need to look at the distribution and the share of extreme ratings.

Always using odd-numbered scales. 1–5 scales with "neutral" are convenient but amplify the effect. To reduce it, consider even-numbered scales or explicitly account for the bias in interpretation.

Long blocks of similar scales. A dozen "rate on 1–5" questions in a row increase fatigue and the share of middle answers. Dilute them with other question types or shorten the block.

Ignoring it during analysis. Failing to check the distribution across scales and not mentioning possible central tendency bias in the report's limitations.

How this looks in SurveyNinja

In SurveyNinja you can configure scales with different numbers of points (including even ones), use logic jumps to shorten the survey and show only relevant questions, and enable anonymous collection. Ready-made NPS templates and satisfaction surveys can be adapted to even-numbered scales or to accounting for central tendency bias during interpretation. In reports you can build distributions across scales and comment on a possible shift toward the center.

Practical recommendations

Consider even-numbered scales. To reduce central tendency bias, use scales without a clear middle (1–4, 1–6) where it makes sense.

Shorten scale blocks. Don't put many "rate on 1–5" questions in a row — dilute them with other question types or shorten the list.

Guarantee anonymity in sensitive surveys. This increases willingness to give extreme ratings and reduces the "retreat to the middle."

Look at the distribution, not just the average. During analysis, check the share of extreme values; if almost everything is in the center, account for possible central tendency bias in your conclusions.

What to write in the report. "When interpreting scale questions, a possible tendency of respondents toward middle values (central tendency bias) was taken into account; the distribution of answers is provided in the appendix."

Central tendency bias is the tendency to choose middle scale values and avoid extreme ones. It is minimized with even-numbered scales without a neutral option, clear wording, a short survey, anonymity, and accounting for the distribution (not just the average) during interpretation.