A Likert scale is a numbered rating scale — typically 1-5 or 1-7 — that measures degrees of agreement, satisfaction, or confidence. In UX research, it shows up in post-task questionnaires, satisfaction surveys, and as a complement to behavioral methods like card sorting. It's named after psychologist Rensis Likert (pronounced LICK-urt, not LIE-kurt), who introduced it in 1932.
A Likert item presents a statement and asks the respondent to rate their agreement on a symmetric scale. The classic format:
1 = Strongly Disagree, 2 = Disagree, 3 = Neutral, 4 = Agree, 5 = Strongly Agree
You can also anchor the scale to other dimensions: confidence (1 = Not at all confident, 5 = Very confident), satisfaction (1 = Very dissatisfied, 5 = Very satisfied), or frequency (1 = Never, 5 = Always). The scale type should match what you're actually measuring.
In card sorting, Likert scales typically appear in the post-sort questionnaire. After a closed card sort, you might ask: "How confident are you that you placed each card in the right category?" on a 1-5 scale. Participants who rate their confidence at 2 or 3 are flagging the cards they found ambiguous — and those cards often show low agreement rates in the sort data itself. That convergence between self-reported confidence and behavioral data strengthens your analysis.
The 5-vs-7 debate has generated more academic papers than it deserves. Here's what actually matters:
5-point scales are faster to complete, easier for participants to understand, and sufficient for most one-off studies. If you're adding a few Likert questions to a post-sort survey, use 5 points.
7-point scales offer finer granularity and slightly more statistical sensitivity. They're worth the tradeoff when you're tracking the same metric over time (like quarterly satisfaction scores) and need to detect small shifts. The System Usability Scale uses 5 points. The Net Promoter Score uses 11 (0-10). Both work fine for their purposes.
Even-numbered scales (4-point, 6-point) remove the neutral midpoint and force participants to lean positive or negative. This is useful when you suspect respondents are parking on "neutral" to avoid thinking, but it can frustrate participants who genuinely feel neutral. Use even scales sparingly and intentionally.
The biggest threat to Likert data isn't scale length — it's acquiescence bias. People tend to agree with statements. If every item in your survey is worded positively ("The categories made sense," "I found sorting easy," "The labels were clear"), your results will skew toward agreement regardless of actual experience.
Counteract this by mixing in reverse-coded items: "I struggled to decide where some cards belonged" or "Some category labels were confusing." When a participant rates both "The categories made sense" and "Some category labels were confusing" as 4 out of 5, you've caught an acquiescence pattern and can weight that response accordingly.
A practical example: after a closed card sort of 40 items for a SaaS knowledge base, you include four Likert questions in the exit survey. Two are positively framed, two negatively framed. One participant rates confidence at 4/5 but also rates "I was unsure where several cards belonged" at 4/5. That inconsistency tells you their high confidence rating is unreliable — check their actual sort data before trusting the self-report.
Report medians and distributions, not just means. A mean of 3.5 on a 5-point scale could mean everyone rated 3 or 4 (mild consensus) or half rated 1 and half rated 5 (strong disagreement). The distribution tells a completely different story.
For small samples (under 30 responses), avoid over-interpreting decimal differences. A mean of 3.8 vs. 4.1 across two rounds of testing is not a meaningful difference with 15 participants. Look at whether the distribution shape changed instead.
When combining Likert data with card sorting results, the most useful analysis is correlating per-card confidence ratings with per-card agreement rates. Cards where participants report low confidence and show low agreement are your highest-priority items for relabeling or restructuring.
Treating Likert data as ratio data. The distance between "Disagree" and "Neutral" isn't necessarily the same as between "Neutral" and "Agree." Use non-parametric tests (Mann-Whitney, Kruskal-Wallis) rather than t-tests when you need statistical comparisons.
Too many Likert items. Survey fatigue degrades response quality. For a post-sort questionnaire, 3-5 Likert items is plenty. Beyond that, participants start straight-lining (selecting the same value for every question).
Unlabeled midpoints. If your 5-point scale only labels the endpoints (1 = Bad, 5 = Good), participants interpret the middle points differently. Label every point for consistent data.
Should I use a 5-point or 7-point Likert scale? A 5-point scale works for most UX research contexts because it's simple for participants to understand and produces enough variation for analysis. A 7-point scale offers finer granularity and slightly higher sensitivity to differences, which can be useful for tracking changes over time. The tradeoff is that 7-point scales take marginally longer to complete and some participants struggle to distinguish between adjacent points like 5 and 6 on a 7-point scale.
What is acquiescence bias in Likert scale surveys? Acquiescence bias is the tendency for survey respondents to agree with statements regardless of content. On a Likert scale, this skews results toward the "agree" end. You can counteract it by including reverse-coded items where agreement indicates the opposite sentiment, mixing positively and negatively worded statements, and using behavioral anchors instead of simple agree/disagree labels.
How do you use Likert scales in card sorting research? Likert scales appear most often in post-sort questionnaires. After completing a card sort, you might ask participants to rate how confident they were in their placements on a 1-5 scale, or how easy they found the sorting task overall. These confidence ratings help you interpret card sort results — low confidence on specific cards often correlates with low agreement rates in the sort itself.
Explore related concepts, comparisons, and guides