You asked 400 people to rate a statement from strongly disagree to strongly agree, and now you have a column of numbers from 1 to 5. Turning that column into something a stakeholder can read in ten seconds takes a few deliberate choices about summary statistic and chart. This article walks through them, and it is part of our guide to charting survey data, which covers every common chart type.
Key takeaways
- To visualize Likert scale data, show the full distribution of responses at each point, then add top-2-box or the mean, never the mean alone.
- A Likert item is one statement with ordered response options, while a Likert scale combines related items into a single score measuring one attitude.
- Top-2-box is the share choosing the two most positive points, such as 240 of 400 respondents, or 60%, though it counts agree and strongly agree the same.
- Different distributions can share the same mean, so two statements with a 3.60 mean can differ sharply, with one polarized at both ends.
- Use a diverging bar for agree versus disagree across items, a 100% stacked bar for groups or waves, and a top-2-box bar for ranking.
What is a Likert scale, and what is a Likert item?
Rensis Likert introduced the approach in 1932. Strictly speaking, a Likert item is a single statement with ordered response options, such as “The checkout process is easy to use.” A Likert scale is a set of related items that together measure one attitude, with the responses combined into a single score.
In everyday survey work, almost any ordered agreement question gets called a Likert scale, and for charting that is fine. For analysis it matters: you report a single item as a distribution, and you can average a combined scale.
The same ordered structure works with other labels, usually called Likert-type responses:
- Frequency: very frequently, frequently, occasionally, rarely, never.
- Importance: very important, important, somewhat important, not very important, unimportant.
- Likelihood: highly likely, likely, somewhat likely, not very likely, not likely at all.
Whatever the wording, keep the options balanced (as many positive as negative points), label every point, and keep the order fixed.
How do you summarize Likert scale data?
Start with the worked example below. The numbers are illustrative: 400 respondents rate the statement “The checkout process is easy to use” on a five-point agreement scale.
| Response | Code | Count | Percent of 400 |
|---|---|---|---|
| Strongly disagree | 1 | 20 | 5.0% |
| Disagree | 2 | 60 | 15.0% |
| Neither agree nor disagree | 3 | 80 | 20.0% |
| Agree | 4 | 140 | 35.0% |
| Strongly agree | 5 | 100 | 25.0% |
| Total | 400 | 100.0% |
Why start with the distribution?
The distribution is the table above: the share of respondents at each point. It is the most honest summary because nothing is collapsed. Every other number in this article is derived from it, so check it first for a shape that a single number would hide, such as two peaks at the extremes.
What is top-2-box, and how do you calculate it?
Top-2-box is the percentage choosing the two most positive points. Here, 140 + 100 = 240 respondents, and 240 / 400 = 60%. Bottom-2-box is 20 + 60 = 80, or 20%. The gap between them is a net score: 60% minus 20% is +40 points.
Top-2-box is easy to say aloud and easy to compare across items, which is why trackers rely on it. Its cost is that “agree” and “strongly agree” count the same.
When is the mean useful, and what are its limits?
Multiply each code by its count, add, and divide by the base: (1 × 20) + (2 × 60) + (3 × 80) + (4 × 140) + (5 × 100) = 1,440, and 1,440 / 400 = 3.60. A mean gives one number to rank items or follow over time.
The caveat is that a single Likert item is ordinal, so the distance from “agree” to “strongly agree” is not guaranteed to match the distance from “disagree” to “neither.” And different shapes can share a mean. Compare statement B, also illustrative, with 400 respondents:
| Statement | Strongly disagree | Disagree | Neither | Agree | Strongly agree | Mean | Top-2-box |
|---|---|---|---|---|---|---|---|
| A | 5% | 15% | 20% | 35% | 25% | 3.60 | 60% |
| B | 20% | 5% | 15% | 15% | 45% | 3.60 | 60% |
Both have a mean of 3.60 and a top-2-box of 60%, yet B is polarized: 20% strongly disagree and 45% strongly agree. Report the mean beside the distribution, never instead of it.
Which chart works best for Likert scale data?
The right chart keeps the response order visible and lets the reader compare items or groups. Three do this well.
| Chart | Best for | Watch out for |
|---|---|---|
| Diverging stacked bar | Seeing agree versus disagree across many items | Needs a clear rule for the neutral segment |
| 100% stacked bar | Comparing full distributions across groups or waves | Inner segments are hard to compare |
| Top-2-box bar | One headline number per item, ranked | Hides the neutral and negative detail |
How does a diverging stacked bar work?
Negative responses extend left of a center line and positive ones extend right, with the neutral segment straddling the line. For statement A, 10% of the neutral share sits on each side, so the bar runs 30% to the left (20% plus 10%) and 70% to the right (60% plus 10%). Sort items by net agreement, and the ranking is visible at a glance. Use two hues, one for each side, and a light gray for neutral.
When should you use a 100% stacked bar?
Use it when you compare groups, such as age bands or survey waves, because every bar has the same total length and the segments are directly comparable as shares. Keep the segment order identical in every bar and label only the segments large enough to hold the text. Our guide to stacked bar charts for survey data covers the construction in more detail.
What does a top-2-box bar show?
It is a plain bar chart with one bar per item, sorted high to low, showing the top-2-box percentage. It is the clearest option for an executive summary of a battery of 15 statements. Pair it with the full distribution in an appendix or detail view. When you have many items and many groups, a table shaded by value can also work, as long as the shading is explained.
Should you include a neutral midpoint?
An odd-numbered scale has a neutral midpoint, and an even-numbered scale forces respondents to lean one way. There is no universal answer, but there are trade-offs.
A midpoint gives genuinely undecided people an honest option. The risk is that “neither” gets used for several different reasons: real neutrality, no opinion, or not understanding the question. A forced choice removes that ambiguity but pushes truly neutral respondents toward a side, which can inflate both ends.
Whichever you choose, add a separate “don’t know” or “not applicable” option outside the scale. Do not merge it with neutral, and say in your notes whether it is excluded from the base. Then keep the design the same across waves, because changing the scale breaks comparison with earlier results.
How do you do this in practice?
In a tabulation tool, set up each Likert item with its top-2-box and bottom-2-box as banner-ready rows, then cross it by the groups you care about. Halo Reports produces crosstabs and banner tables with weighting and significance testing, plus charts and interactive report pages, so you can see whether a gap in top-2-box between two segments is more than noise. If you track the scale each wave, mTab Pulse monitors trackers and studies and surfaces meaningful movement.
What are the common mistakes with Likert data?
- Reporting only the mean. As statements A and B show, identical means can hide very different opinions.
- Reordering the categories. Sorting response options by size or alphabetically destroys the scale. Keep the order from negative to positive.
- Averaging reverse-worded items without recoding. If one item is negatively worded, flip its codes before combining it with the others.
- Mixing “don’t know” into the middle. It is not a neutral opinion, and it distorts both the base and the neutral share.
- Using a rainbow of colors. Use a diverging palette with a clear midpoint so direction is obvious without reading the legend.
What should you do next?
Take your next Likert question and build the table first: counts, percentages, top-2-box and bottom-2-box. Then pick one chart from the three above and check that a colleague can read the answer without explanation. If the question is part of a tracker, write down your scale length, midpoint choice and “don’t know” treatment so they stay the same next wave. When the result goes to leadership, an executive brief can carry it as a one-page update of what changed, why it matters and what to do.