What Is a Diverging Bar Chart?

A standard bar chart starts every bar at the same edge, zero at the left for horizontal charts, zero at the bottom for vertical. A diverging bar chart does something different: it anchors all bars at a shared center point and lets them extend in either direction. Values above the midpoint grow to the right (or upward). Values below the midpoint grow to the left (or downward).

That center point is most often zero. For financial data, zero is the literal breakeven line. For performance metrics, zero can mark plan versus actual. For survey data using a Likert scale, the center typically represents a neutral position, with favorable responses on one side and unfavorable responses on the other.

The chart goes by a few names: diverging bar graph, two-way bar chart, positive-negative bar chart, Likert scale chart (when used for survey data), butterfly chart (for mirrored population pyramids), bipolar bar chart, and diverging stacked bar chart when each bar displays multiple sub-segments of a rating scale. The structure is the same across all of them: a symmetric axis, a meaningful midpoint, and bars that show direction through length and position. The full bar graph guide covers the broader family of chart types and when each applies.

Figure A · Diverging Bar Chart Anatomy

Anatomy of a diverging bar chart: five labeled components including the zero axis, negative bars, positive bars, symmetric scale, and directional labels A horizontal bar chart with three rows. Numbered callouts mark: 1, the center zero axis; 2, a coral negative bar extending left; 3, a teal positive bar extending right; 4, the symmetric scale labels at bottom; 5, directional text labels reading Unfavorable left and Favorable right. Survey Q1 Survey Q2 Survey Q3 -50 -25 0 +25 +50 ← UNFAVORABLE FAVORABLE → 1 2 3 4 5
Five structural components every diverging bar chart must get right. The numbered key below identifies each one.
  1. Zero axis: The center line where all bars originate. Mark it more prominently than gridlines.
  2. Negative bars: Extend left in a warm color (coral, red, orange) by convention.
  3. Positive bars: Extend right in a cool color (blue, teal) by convention.
  4. Symmetric scale: Both sides of the axis must extend equally from zero.
  5. Directional labels: Always state which side is favorable. Never leave it to inference.
Quick Reference

A diverging bar chart, also called a diverging bar graph or two-way bar chart, centers all bars on a zero baseline and extends them in opposite directions. Positive values grow right; negative values grow left. Most commonly used for Likert scale surveys, budget variance analysis, and profit-and-loss comparisons. The format was formally documented in the Journal of Statistical Software by Naomi B. Robbins and Richard M. Heiberger (2014), building on their earlier 2011 recommendation.

When Does Two-Way Data Need a Two-Way Chart?

Not all data with negative values belongs in a diverging chart. The chart earns its place when direction carries meaning, not just when a number happens to be negative. When the difference between left and right tells a story about contrast, competition, or opposing positions, the diverging format is usually the clearest option.

When the data does not have a meaningful midpoint, a diverging chart is the wrong choice. A sales revenue comparison across five regions, for example, has no natural zero in the middle. All values are positive and share a common baseline at the left edge. Forcing that into a diverging layout would require an arbitrary center point that distorts rather than clarifies. Use a standard horizontal bar chart for that scenario. The bar chart type comparison guide covers more cases like this.

Use Case Data Type Example
Likert scale surveys Rating responses from "Strongly Disagree" to "Strongly Agree" Post-training course evaluation across five questions
Budget variance Actual spend versus planned spend, by department Q3 departmental budget report showing over- and under-budget lines
Profit and loss by unit Financial performance by product line or region Regional profitability summary across six territories
Before/after comparison Change from a baseline; can be positive or negative Page speed improvement (or regression) across site sections after a redesign
Population demographics Two mirrored groups sharing a common category axis Age pyramid showing male and female population by age bracket
Sentiment analysis Net positive vs. net negative sentiment by topic Customer review sentiment broken down by product feature

For survey-specific use cases, the guide to bar graphs for survey results goes deeper on how to match chart format to survey structure.

The Centered-Zero Baseline

The center axis is the one element that makes or breaks a diverging chart. Without it being clear and prominent, readers spend time decoding the structure instead of reading the data.

Three things to get right.

First, keep the scale symmetric. If the most negative bar reaches -50 on the left, the right side needs to run to at least +50. A lopsided axis compresses one side and inflates the other, turning a comparison into a distortion.

Second, give the center line more visual weight than the gridlines. A slightly heavier rule, or a distinct vertical stroke, tells readers where to anchor their comparisons. Gridlines alone aren't enough; readers need one unambiguous reference point to orient from.

Third, label the directions. "Unfavorable" on the left, "Favorable" on the right. Readers shouldn't have to guess, and a simple annotation or legend at each axis end removes that uncertainty entirely.

Diverging bar chart example: customer satisfaction net scores A horizontal diverging bar chart showing five customer satisfaction dimensions. Warm coral bars extend left for unfavorable scores; teal bars extend right for favorable scores. Product quality has the strongest favorable score. Shipping speed has the strongest unfavorable score. CUSTOMER SATISFACTION NET SCORES (ILLUSTRATIVE) -50 -25 0 +25 +50 UNFAVORABLE FAVORABLE Product quality Shipping speed Customer support Ease of returns Value for price Unfavorable Favorable
Figure 1. Illustrative diverging bar chart: customer satisfaction net scores across five dimensions. Hover (or tap on mobile) any part of the chart for a design tip about that element.

What Color Conventions Work Best for Diverging Bar Charts?

Color in a diverging chart does two things: it signals direction and sets reader expectations before anyone looks at a number. The approach breaks cleanly by data type.

For survey data and Likert scales, a diverging palette works best. One hue covers the negative (unfavorable or disagree) side; a contrasting hue covers the positive (favorable or agree) side. Neutral responses typically get a gray or muted tone. Red-to-blue and orange-to-teal are the two pairings that appear most in published survey research because they don't carry the same warm/cool cultural weight in both Western and East Asian contexts, which matters for research meant to be read across countries. ColorBrewer (colorbrewer2.org) offers a set of tested diverging palettes designed for both screen and print use.

For financial or operational data, a simpler two-color approach works: red for negative (loss, over-budget, decline) and blue or green for positive (profit, under-budget, growth). This color pairing is standard in financial reporting, so most audiences read it without needing a legend explanation.

Design Note

Avoid using sequential palettes (light to dark of a single hue) for diverging charts. A single hue cannot convey the conceptual opposition between the two sides. The two halves of a diverging chart represent fundamentally different categories, and the color encoding should reflect that.

Red-green pairings are the most common default in financial charts and one of the more avoidable mistakes in chart design. Red-green color blindness affects around 8% of males of Northern European descent, and between 4 and 6% of males in other populations, making it the most prevalent form of color vision deficiency worldwide. An orange-to-blue or orange-to-teal pairing sidesteps the problem entirely. The Okabe-Ito palette, developed by Masataka Okabe and Kei Ito and published in the CUDO universal design guide (2008), was built specifically for color vision deficiency and is one of the most-cited colorblind-safe palettes in scientific publishing. For a two-sided diverging bar graph, the practical approach is to pick two contrasting colors from the palette: the blue (#0072B2) for the favorable side and the orange-red (#D55E00) for the unfavorable side. Both remain distinguishable across all common forms of color vision deficiency. The colorblind-safe palette guide covers tested options across common bar chart types.

The Diverging vs. 100% Stacked Debate for Likert Scales

This is where the data visualization community has had a genuine and useful disagreement, and understanding both sides helps in choosing the right format.

Naomi B. Robbins and Richard M. Heiberger comprehensively documented the diverging stacked bar chart approach for Likert data in their 2014 Journal of Statistical Software paper, an expansion of their earlier 2011 JSM Proceedings recommendation, drawing on perceptual research into how readers decode bar length and relative position. That peer-reviewed work gave the format its academic grounding. Later practitioners tested the approach in real-world contexts and began raising questions about specific tradeoffs.

In his 2016 Perceptual Edge article, data visualization practitioner Stephen Few examined variants of the stacked bar chart for Likert scale survey data. Few looked at a design where the neutral category is treated as a separate bar rather than split at the centerline, which preserves a common baseline for most comparisons and shows the competition between the agree and disagree sides clearly. He also addressed the version where neutral responses are divided and placed on each side of the center, noting the tradeoffs around baseline alignment when that approach is used.

Datawrapper responded with a detailed analysis in "The Case Against Diverging Stacked Bars," arguing that for percentage-based Likert data (where all responses sum to 100%), the 100% stacked bar chart lets readers compare the outermost totals more reliably because every bar shares a common left edge. Their core position: the diverging format shows the competition between two camps clearly, but it makes it harder to read the exact share that lands in each outer category.

Both approaches can work. The right choice comes down to one question: what comparison matters most to the audience?

If readers need to track individual response bands across multiple questions, the diverging stacked format shows that more clearly. If the main message is the overall share that agrees versus disagrees on each item, a 100% stacked bar chart is easier to read because every bar shares a common left edge. Most Likert chart problems get resolved by answering that question before picking a format.

How to Make a Diverging Bar Chart in BarGraphCreator

The free BarGraphCreator tool builds diverging bar charts and diverging bar graphs through its horizontal chart mode combined with signed data values. No account or sign-up is needed. Here are the steps:

  1. Open the tool. Go to the BarGraphCreator homepage. The chart builder loads directly in the browser with no installation required.
  2. Enter category labels. Type your row labels into the data panel. For a Likert scale survey, each label is one survey question. For a budget variance chart, each label is a department or cost center.
  3. Enter signed values. Use negative numbers for the unfavorable (or left) side and positive numbers for the favorable (or right) side. For a survey question where 35% of responses are unfavorable and 55% are favorable, enter -35 and 55 as separate data points or use a net score of +20, depending on how the data should be represented.
  4. Select horizontal orientation. Once category labels run past four or five words, vertical diverging charts become hard to read quickly. Horizontal solves that immediately. BarGraphCreator defaults to vertical, so this is an easy step to skip on a first build. See the horizontal bar chart maker guide for more on when horizontal outperforms vertical.
  5. Apply a diverging color palette. Choose a warm tone (orange, coral, red) for negative values and a cool tone (blue, teal) for positive values. The color encoding should make the directional split immediately visible before a reader looks at the axis labels.
  6. Set a symmetric axis range. Adjust the axis so the minimum and maximum extend equally from zero. If the largest negative value is -40, the axis maximum should be at least +40. An asymmetric axis will visually exaggerate one side of the chart.
  7. Export the chart. PNG works for most slide decks and documents. If the chart will appear in print or a high-resolution PDF, export at 2x resolution to avoid blurring at bar edges. SVG export, where available, stays sharp at any size.

For a general walkthrough of the chart creation process from start to finish, the how to make a bar graph guide covers inputs, axis settings, color choices, and export formats across all chart types available in the tool.

Diverging Bar Chart: Common Questions

What is the difference between a diverging bar chart and a grouped bar chart?

A grouped bar chart places two or more bars side by side for each category, with all bars starting from a shared zero baseline at one edge. A diverging bar chart places a single bar per category at a center baseline and lets it extend in either direction. Use grouped bars when comparing multiple datasets across the same categories at the same scale. Use a diverging chart when the direction of each value (positive or negative relative to a midpoint) is the primary story rather than the raw comparison between groups.

When should I use diverging bars instead of a 100% stacked bar chart for Likert data?

The fastest way to decide: sketch both formats with the same dataset and show them to one non-specialist colleague. If they immediately ask what the individual colored bands mean in the stacked version, the stacked chart is adding decoding work the audience doesn't need. If they can read the agree/disagree split directly from the stacked version and the diverging chart just looks busier, go stacked. The debate section above covers the theoretical case; this test covers the practical one.

How should neutral responses be handled in a Likert diverging chart?

Two approaches are common. The first treats neutral as a standalone bar between the two halves, which keeps most bars anchored to the same edge and makes the split easier to read at a glance. The second splits neutral responses down the middle, assigning half to each side. The split method is controversial because it assumes neutral responses are evenly distributed between the two camps, which is rarely verifiable. For most purposes, treating neutral as a standalone element or excluding it entirely (if the goal is only to compare favorable versus unfavorable) produces a cleaner and more defensible chart.

Can a diverging bar chart be displayed vertically?

Yes, and vertical diverging charts (sometimes called center-baseline column charts) do appear in practice, particularly for time series data showing values above and below a target or baseline. However, horizontal layouts are much more common for diverging charts used with categorical data. Horizontal bars accommodate longer category labels without rotation or truncation and scan more naturally when readers are comparing opposing halves across many rows. For most survey data, budget variance, and sentiment applications, horizontal is the right default choice.

What is the best color palette for a diverging bar chart?

ColorBrewer (colorbrewer2.org) is the most reliable starting point and it's free. The RdBu (red to blue), RdYlBu (red-yellow-blue), and PuOr (purple to orange) palettes all work well for screen and print. Skip red-green entirely for colorblind-safe designs. Orange-to-teal and purple-to-orange test reliably across most forms of color vision deficiency. A practical field test: can a reader tell which side is which from six feet away without reading the legend? If not, the contrast isn't strong enough.

Sources & References

  1. Few, Stephen. "When Are 100% Stacked Bar Graphs Useful?" Perceptual Edge, 2016. https://www.perceptualedge.com/blog/?p=2239
  2. Datawrapper Blog. "The Case Against Diverging Stacked Bars." https://www.datawrapper.de/blog/divergingbars
  3. Heiberger, Richard M. and Robbins, Naomi B. "Design of Diverging Stacked Bar Charts for Likert Scales and Other Applications." Journal of Statistical Software, Vol. 57, Issue 5, March 2014. https://www.researchgate.net/publication/289590282
  4. Wikipedia. "Likert scale." https://en.wikipedia.org/wiki/Likert_scale
  5. Domo. "Divergent Bar Charts: Examples, Best Practices, and How to Build." https://www.domo.com/learn/charts/divergent-bar-charts